MIME-Version: 1.0 Content-Type: multipart/related; boundary="----=_NextPart_01C48638.50519170" This document is a Web archive file. If you are seeing this message, this means your browser or editor doesn't support Web archive files. For more information on the Web archive format, go to http://officeupdate.microsoft.com/office/webarchive.htm ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT) ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_document.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT) <body> <p>This page uses frames, but your browser doesn't support them.</p> </body> ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/preview.wmf Content-Transfer-Encoding: base64 Content-Type: image/x-wmf AQAJAAADDwYAAAYAPQAAAAAABAAAAAMBCAAFAAAACwIAAAAABQAAAAwC7wLpAwMAAAAeAAcAAAD8 AgAA////AAAABAAAAC0BAAAIAAAA+gIFAAAAAAD///8ABAAAAC0BAQAMAAAAQAkhAPAAAAAAAAAA 7wLpA/////8IAAAA+gIAAAAAAAAAAAAABAAAAC0BAgAHAAAA/AIAAP///wAAAAQAAAAtAQMABAAA ACcB//8cAAAA+wLU/wAAAAAAALwCAAAAAABAAABBcmlhbAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAQAAAAtAQQABAAAAC4BGAAEAAAAAgEBAAUAAAAJAgAAAAI9AAAAMgqvAFsAJAAAAER5bmFt aWMgQWJzdHJhY3Rpb24gaW4gUmVpbmZvcmNlbWVudCAAGAAcABgAKAAMABgADQAgABsAGAAPABIA GQAYAA8ADAAbABsADQAMABsADAAhABgADAAbAA8AGwASABkAGAAoABgAGwAPAAQAAAAuAQAAHAAA APsCEAAHAAAAAAC8AgAAAAABAgIiU3lzdGVtAAAAAAAAGAAAAGxpEwACAAAA5AQAAAAAAAAEAAAA LQEFAAQAAADwAQQAHAAAAPsC1P8AAAAAAAC8AgAAAAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAEAAAALQEEAAQAAAAuARgABAAAAAIBAQAFAAAACQIAAAACKgAAADIK/wD+ABcA AABMZWFybmluZyB2aWEgQ2x1c3RlcmluZwAbABgAGQARABsADAAbABwADAAZAAwAGQAMACAADAAc ABgADwAZABEADAAbABsABAAAAC4BAAAEAAAALQEFAAQAAADwAQQAHAAAAPsC1P8AAAAAAACQAQAA AAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAEAAAALQEEAAQAAAAuARgABAAA AAIBAQAFAAAACQIAAAACNAAAADIKqQHSAB4AAABTaGllIE1hbm5vciAoTUlUICAgICAgIE1jR2ls bCkdABgACwAYAAwAJgAZABgAGQAYAA8ADAAPACUADQAaAA0ADAAMAA0ADAAMAAwAJgAWACIACgAK AAoADwAEAAAALgEAAAQAAAAtAQUABAAAAPABBAAcAAAA+wLk/wAAAAAAAJABAAAAAABAAABBcmlh bAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAAAAtAQQABAAAAC4BGAAEAAAAAgEBAAUAAAAJ AgAAAAIiAAAAMgr5Af0AEgAAAEpvaW50IHdvcmsgd2l0aDogIA4ADwAGABAACAAIABMADwAJAA4A CQATAAYACAAQAAgACAAIAAQAAAAuAQAABAAAAC0BBQAEAAAA8AEEABwAAAD7At//AAAAAAAAkAEA AAAAAEAAAEFyaWFsAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAC0BBAAEAAAALgEYAAQA AAACAQEABQAAAAkCAAAAAg8AAAAyCvkBywEFAAAASXNoYWkACQARABIAEgAHAAQAAAAuAQAABAAA AC0BBQAEAAAA8AEEABwAAAD7At//AAAAAAAAkAEAAAAAAEAAAEFyaWFsAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAABAAAAC0BBAAEAAAALgEYAAQAAAACAQEABQAAAAkCAAAAAhIAAAAyCvkBGgIH AAAATWVuYWNoZW4cABIAEwARABEAEwASAAQAAAAuAQAABAAAAC0BBQAEAAAA8AEEABwAAAD7At// AAAAAAAAkAEAAAAAAEAAAEFyaWFsAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAC0BBAAE AAAALgEYAAQAAAACAQEABQAAAAkCAAAAAgkAAAAyCvkBrgIBAAAAKIYLAAQAAAAuAQAABAAAAC0B BQAEAAAA8AEEABwAAAD7At//AAAAAAAAkAEAAAAAAEAAAEFyaWFsAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAABAAAAC0BBAAEAAAALgEYAAQAAAACAQEABQAAAAkCAAAAAhMAAAAyCvkBuQIIAAAA VGVjaG5pb24WABEAEgASABMACAASABIABAAAAC4BAAAEAAAALQEFAAQAAADwAQQAHAAAAPsC3/8A AAAAAACQAQAAAAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAEAAAALQEEAAQA AAAuARgABAAAAAIBAQAFAAAACQIAAAACCQAAADIK+QFDAwEAAAAphgsABAAAAC4BAAAEAAAALQEF AAQAAADwAQQAHAAAAPsC3/8AAAAAAACQAQAAAAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAEAAAALQEEAAQAAAAuARgABAAAAAIBAQAFAAAACQIAAAACDQAAADIKOQLJAQQAAABB bWl0FwAbAAcACgAEAAAALgEAAAQAAAAtAQUABAAAAPABBAAcAAAA+wLf/wAAAAAAAJABAAAAAABA AABBcmlhbAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAAAAtAQQABAAAAC4BGAAEAAAAAgEB AAUAAAAJAgAAAAINAAAAMgo5AhUCBAAAAEhvemUZABIAEQASAAQAAAAuAQAABAAAAC0BBQAEAAAA 8AEEABwAAAD7At//AAAAAAAAkAEAAAAAAEAAAEFyaWFsAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAABAAAAC0BBAAEAAAALgEYAAQAAAACAQEABQAAAAkCAAAAAgkAAAAyCjkCbAIBAAAAKIYLAAQA AAAuAQAABAAAAC0BBQAEAAAA8AEEABwAAAD7At//AAAAAAAAkAEAAAAAAEAAAEFyaWFsAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAAC0BBAAEAAAALgEYAAQAAAACAQEABQAAAAkCAAAAAhMA AAAyCjkCdwIIAAAAVGVjaG5pb24WABEAEQATABIACAASABIABAAAAC4BAAAEAAAALQEFAAQAAADw AQQAHAAAAPsC3/8AAAAAAACQAQAAAAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAEAAAALQEEAAQAAAAuARgABAAAAAIBAQAFAAAACQIAAAACCQAAADIKOQIAAwEAAAAphgsABAAA AC4BAAAEAAAALQEFAAQAAADwAQQAHAAAAPsC3/8AAAAAAACQAQAAAAAAQAAAQXJpYWwAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAEAAAALQEEAAQAAAAuARgABAAAAAIBAQAFAAAACQIAAAACGAAA ADIKdQLJAQsAAABVcmkgS2xlaW4gKAAYAAsABwAJABcACAASAAcAEwAJAAsABAAAAC4BAAAEAAAA LQEFAAQAAADwAQQAHAAAAPsC3/8AAAAAAACQAQAAAAAAQAAAQXJpYWwAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAEAAAALQEEAAQAAAAuARgABAAAAAIBAQAFAAAACQIAAAACEwAAADIKdQJbAggA AABUZWNobmlvbhYAEQARABIAEwAIABIAEgAEAAAALgEAAAQAAAAtAQUABAAAAPABBAAcAAAA+wLf /wAAAAAAAJABAAAAAABAAABBcmlhbAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAAAAtAQQA BAAAAC4BGAAEAAAAAgEBAAUAAAAJAgAAAAIJAAAAMgp1AuQCAQAAACmGCwAEAAAALgEAAAQAAAAt AQUABAAAAPABBAAEAAAA8AEAAAMAAAAeAAgAAAD6AgAAAQAAAAAAAAAEAAAALQEAAAcAAAD8AgAA AAAAAAAABAAAAC0BBAAEAAAABgECACMAAAA4BQMABQAFAAQAPwKeAXcCngF3AqIBPwKiAT8CngE/ AqYBOAKgAT8CmgFFAqABPwKmAXUCmgGBAqABdQKmAXUCmgEEAAAABgEBAAQAAAAtAQMABAAAAC0B AgAEAAAA8AEAAAQAAAAnAf//AwAAAAAA ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_next_active.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhHgASAKIAAP///8DAwJT//4SEhCnO/ylj/wAAAAAAACH5BAEAAAEALAAAAAAeABIAQAMy GLrc/hAaQYuJOGtYut+ZIY5gaYZASlynUxCw1c50bW8Ttd5LLuy3V4xF83SIvKQymQAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_next_disabled.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhHgASAJEAAP///8DAwISEhAAAACH5BAEAAAEALAAAAAAeABIAQAIpjI+pyywPm5yUjYvr hBHoDzbcEy6jEHjlyraudL5H/MZqabvAzsu+XAAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_prev_active.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhHgASAKIAAP///8DAwJT//4SEhCnO/ylj/wAAAAAAACH5BAEAAAEALAAAAAAeABIAQAM0 GLrc/pAVIqqJOOu9iv8cZoxkaJ4SAawX2hgTQbRubd84NFUC7e6WHEw2yylgH59xyXQlAAA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_prev_disabled.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhHgASAJEAAP///8DAwISEhAAAACH5BAEAAAEALAAAAAAeABIAQAItjI+pyzgCYQCt2ovD 2DzXGHni6IDCRCLmmbbuCzcPiKZzSLlrbOw8AAzyhqQCADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_outline_collapse.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhFwASAKIAAP///8DAwISEhDFj/wAAnAAAAAAAAAAAACH5BAEAAAEALAAAAAAXABIAQANG GLrczgTISeWIkrzNVywFAFak5pHkNZkK4b5wDENiOKIrjQND30esAGZX6hgdQyIlmFRmdM7ciejD MJ2q55Q6ZMq+saM4AQA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_outline_expand.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhFwASAKIAAP///8DAwISEhDFj/wAAnAAAAAAAAAAAACH5BAEAAAEALAAAAAAXABIAQANK GLrczgSUIim4d8RLnv9LNFVYWXahqQ7shipELM/0DGHUNKrAG2ySyg6gwfiAPJ4PxIQln8dnMipV UaVFzu3J6kauXOCxRqY1zwkAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_space.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhAQABAPcAAAAAAIAAAACAAICAAAAAgIAAgACAgICAgAQEBAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAMDAwP8AAAD/AP//AAAA//8A/wD//////yH5BAEAAPcALAAAAAABAAEA AAgEAO8FBAA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_endshow.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252"


End of slide show, click to exit.

------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_navigation_bar.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252"
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_outline_navigation_bar.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252"
3D"Collapse3D"Expand
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_outline_collapsed.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252"

Dynamic Abstraction= in Reinforcement
Learning via Clustering

Shie Mannor (MIT       McGill)
Joint work with:  Ishai Menache (T= echnion)
    Amit Hoze (Technion)
Uri Klein (Technion)

Hierarchical RL=

RL with Options=

Good Options Lead T= o:

Getting to different regions of the space

Method

When enough is enou= gh?

How to Cluster<= /b>

Slide 9

Experiments=

Performance=

Performance=

Car-Hill Problem

Car-Hill (Performan= ce)

Multi-Valued Maze

Performance (multi-= valued maze)

Future Research=

------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_outline_expanded.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252"

Dynamic Abstraction= in Reinforcement
Learning via Clustering

Shie Mannor (MIT       McGill)
Joint work with:  Ishai Menache (T= echnion)
    Amit Hoze (Technion)
Uri Klein (Technion)

Hierarchical RL=
Divide a problem to multiple = subtasks
Learn each sub-task separatel= y
Learn when to start a sub-tas= k
How can we learn sub-tasks automatically?

RL with Options=
An option is:
1. Starting condition=
2. Sub policy
3. Termination condition
Q-learning with options

Good Options Lead T= o:
states with unusual reward
states that were frequently o= bserved near high reinforcement
bottlenecks in the state spac= e
(very) different regions of t= he state space

Getting to different regions of the space
Balanced exploration
Meaningful options
Options will go a long way
But: don’t know how the state= space looks

Method
Perform learning as usual, re= cord transitions.
After enough time passed – co= nstruct transition graph
Perform state space clusterin= g
Options learn to move between= clusters

When enough is enou= gh?
Tradeoff:
Learn too little and cluster= s will not be meaningful
Wait too long and options wi= ll not be useful
Bootstrapping
Heuristic rule – activate clu= stering when only a few recent changes

How to Cluster<= /b>
Solve the optimization proble= m:
(E is the set of links between clusters)
g measures the quality of the= two clusters
Clustering problem is hard fo= r almost any
g – bottom-up heuristic

Slide 9

Experiments=

Performance=

Performance=

Car-Hill Problem

Car-Hill (Performan= ce)

Multi-Valued Maze
Multiple starting points
Multiple rewards (negative/po= sitive)
50 x 50 states

Performance (multi-= valued maze)

Future Research=
Is it a hack?
Learning what works in a prob= lem
Forgetting options
Large state space
– Approximate clustering
– Local clustering
– POMDPs

------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_blank_notes.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/master03.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii"
Click to edit Master title style
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹date/time›
‹footer›
‹#›
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/master03.xml Content-Transfer-Encoding: quoted-printable Content-Type: text/xml; charset="utf-8" ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/pres.xml Content-Transfer-Encoding: quoted-printable Content-Type: text/xml; charset="utf-8" ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0002.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
<= /td> <= /td> <= /td> = <= /td> <= /td>
Dynamic Abstraction in Reinforcement
= Learning via Clustering
Shie Mannor (MIT       McGill)
   Joint work with:  = Ishai Menache (Technion)
    <= /span>       Amit Hoze (Technion)
      Uri Klein (Technion)
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0002_background.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh3AFlAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAADb AWQBgQAAAAAAAP///wECAwL/lI+py+0Po5y02ouz3rz7D4biSJbmiabqyrbuC8fyTNf2jef6zvf+ DwwKh8Si8YhMKpfMpvMJjUqn1Kr1is1qt9yu9wsOi8fksvmMTqvX7Lb7DY/L5/S6/Y7P6/f8vv8P GCg4SFhoeIiYqLjI2Oj4CBkpOUlZaXmJmam5ydnp+QkaKjpKWmp6ipqqusra6voKGys7S1tre4ub q7vL2+v7CxwsPExcbHyMnKy8zNzs/AwdLT1NXW19jZ2tvc3d7f0NHi4+Tl5ufo6err7O3u7+Dh8v P09fb3+Pn6+/z9/v/w8woMCBBAsaPIgwocKFDBs6fAgxosSJFCtavIgxo8aN/xw7evwIMqTIkSRL mjyJMqXKlSxbunwJM6bMmTRr2ryJM6fOnTx7+vwJNKjQoUSLGj2KNKnSpUybOn0KNarUqVSrWr2K NavWrVy7ev0KNqzYsWTLmj2LNq3atWzbun0LN67cuXTr2r2LN6/evXz7+v0LOLDgwYQLGz6MOLHi xYwbO34MObLkyZQrW76MObPmzZw7e/4MOrTo0aRLmz6NOrXq1axbu34NO7ZsdgECdLA9W0Lt2hx2 536wO7jw4cSL4/69wLjy5byRN2i+Abrz6dSrW7+OPbv27dy7e/8OPrz48eTLmz+PPr369ezbu38P P778+fTr27+PP7/+/fz7+4H/D2CAAg5IYIEGHohgggouyGCDDj4IYYQSTkhhhRZeiGGGGm7IYYce fghiiCKOSGKJJp6IYooqrshiiy6+CGOMMs5IY4023ohjjjruyGOPPv4IZJBCDklkkUYeiWSSSi7J ZJNOPglllFJOSWWVVl6JZZZabslll15+CWaYYo5JSwEAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0051.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Hierarchical RL
= <= /td> <= /td>
Div= ide a problem to multiple subtasks
Lea= rn each sub-task separately
Learn when to s= tart a sub-task
How can we learn sub-tasks automatically?
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0052.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= RL with Options
= <= /td> <= /td> <= /td>
An option is:
   1. Starting condition
   2. Sub policy
   3. Termination condition
Q-l= earning with options
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0052_image001.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhrAFCAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQABACj AToAgAAAAAAAAAL/hI9poe3fGJw01gsl3rz7D25aSIZjiSant6ZK675xNtf23cH4re8i2asFXcNY 0YdM5pS7I9PhtDyjIGrJ+swqA1xA98voerFIstZs/qQx66U2104WNRKw945/P+LyCfePwPdD9Ed3 BlcoJ4hzZIhXNyZ2lniw6NNz4pjFl7nwhiWpd8nh5Ah5Z0mqUinKEtFyunkFmzfFktoqw1YaVtg7 FrlJW5tbMayLS9EGE/YptZdcDMwmTQTlWW2sK7VCKRJ6vWCl0/wC+D10DIUZ6m0HRlfXO3IOPU3p Wwkubs6aAi+O3rlE9QLtwRao4JRyBleF64TwgqZw93wpjLiqmz+J/xspNrT3UB4/gvjmRRJZERag XxUb7gPHkNNAMTRR0jz40CGxLfx0YvwIcSeNjs9c5dRZ6qNHov1elezZLh5UqS3XJZyqz5u+jD9N UK0q6SLTjUGj4RQadJtGVUqLuuHmVpnPZxbF7uvpFO80sFRrglThV29Trka+wox6Vu02WWPTKlWn re3irBYnk13qZ65loVczqqyKN3BYUxkysRxd+pUQw1IRY44ImaMJycQ+d5wod6zmQYCJ4h7adXeQ l8Ee2RS4FfTou8mx7h04uHPh5qINMbcNW3ctQb9/Ot5+WTrCm0WZA8dY9nbZ7sz8rJ/pPPTx5I2Q t5R38SUu1M7J7/+du9Z4vtFmDERf7HZSewHFI5BpDJYHilbqLZcVYA1uxlmC+VS4YF0LalgZSRJS Rhl8HG6WiojzrBiiWAlS5E47Fb5z2oNXERcXho3lpiNvr2k2R47ZkLLhFt8NmVmSOHloyi/l0OOj blCeF5xwKCiY4YCREYikbNrxwGWXGE6ZGHjYnLLilyhqFaBlQQqJjoxhHpmjWYxV2YSaYnYoGTmL kfZIbWyVdp1VcPSohl1/fYPgngWaVwaejppTqGqPseikSW9Nymmnnn4Kaqiijkpqqaaeimqqqq56 CaSsvgprrLKqIemstt6KK6zd5cprr76KGsuvwg5LbJdFFotsssr/Lstss84+C2200k5LbbXWXott ttpuy2233n4Lbrjijktuueaei2666rJqZ5brDkrku6q26668VNJqr6n01povol72C6wtAD8KDI2C 6cnEvtAeW68RA3sp0oH4ETSkws2mJ/DDSqLpyYgNZ2zvXcHypDFwHKNyby6ulqugxV257DLCm8bb YsH6xazku37u5DGhPIa5y39whmZzFTYMt+t0K/OA8687YwxxzjIrw9CPHftzIL4zYJIHvVlPIu/O gvKbXcpxLJ3lkwfPLJ5rPapdRtPNcVsXTGe1uaWkDFam9ToeWpA01X/P7TaIhe4r99zpYolWlMMN Tc2/hpIt9T9UuydcDNrfYmfmx52XSXnKVhNeoOhAXE5ylEFHp+53ZDKz3s9Tg266Xlx/uZ97mCd2 tqVhe9wgmcWd+PLokt9+nvCf+2tU7Vu3pXzfs4Objs9ADz99FZrrWFD0mevR/YyZggZk4r0y/vj1 xafOtkNMIpn4xCth7aRx+eDdOpuNlp136FcYTSXvtSJ+PgEUJIC3v3Uh0Hi0U1gakHcb+ImiTQa8 Gu9K1j/FSYROTOsgULZ3K7ttpWYAsZADCgAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0053.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Good Options Lead To:
<= /td> = <= /td> <= /td> <= /td>
sta= tes with unusual reward
sta= tes that were frequently observed near high
reinforcement
bot= tlenecks in the state space
(very) different regions of the state space
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0025.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Getting to different regions of the space
= <= /td> <= /td>
Bal= anced exploration
Meaningful opti= ons
Opt= ions will go a long way
But: don’t know how the state space looks
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0015.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
<= b>Method
<= /td> = = <= /td> <= /td>
Per= form learning as usual, record transitions.
Aft= er enough time passed – construct
transition grap= h
Per= form state space clustering
Opt= ions learn to move between clusters
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0018.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= When enough is enough?
<= /td> <= /td> <= /td> <= /td> = <= /td>
Trad= eoff:
   Learn too little and clusters will = not be meaningful
   Wait too long and options will not = be useful
Boo= tstrapping
Heu= ristic rule – activate clustering when only
a f= ew recent changes
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0021.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= How to Cluster
= <= /td> <= /td> <= /td>
Sol= ve the optimization problem:
(E = is the set of links between clusters)
= g measures the quality of the two clusters
Clu= stering problem is hard for almost any
g – bott= om-up heuristic
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0021_background.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODdh2wFkAXcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAA2wFkAYAAAAD/ //8C/4yPqcvtD6OctNqLs968+w+G4kiW5omm6sq27gvH8kzX9o3n+s73/g8MCofEovGITCqXzKbz CY1Kp9Sq9YrNarfcrvcLDovH5LL5jE6r1+y2+w2Py+f0uv2Oz+v3/L7/DxgoOEhYaHiImKi4yNjo +AgZKTlJWWl5iZmpucnZ6fkJGio6SlpqeoqaqrrK2ur6ChsrO0tba3uLm6u7y9vr+wscLDxMXGx8 jJysvMzc7PwMHS09TV1tfY2drb3N3e39DR4uPk5ebn6Onq6+zt7u/g4fLz9PX29/j5+vv8/f7/8P MKDAgQQLGjyIMKHChQwbOnwIMaLEiRQrWryIMaPGjf8cO3r8CDKkyJEkS5o8iTKlypUsW7p8CTOm zJk0a9q8iTOnzp08e/r8CTSo0KFEixo9ijSp0qVMmzp9CjWq1KlUq1q9ijWr1q1cu3r92hAAgDxj wSooS9YsArRp1RpgiwcuWLl36Ha1WwfvVr1z+GL1KwewVcFxCE81/AYxPrGMGztmzERxG8l9xFKw nAazA81JKK/xvIczBNFmSJ+NzKDxWw2qA0h+DNkGaHum16I+fSD2Bba6LfDGMZt28B9yi2cwjgHu 8A/L6dV2onwBZbqeyzYHcd2K5di9W6uGjHl7eM3jHed+fgW9kMe5E1zPbn1Hdirex7+1jn+1a/Hl 7+//j++ffupVMaAPv7WHoAfztZbDfPShRV5++53Hm4QTBihahAx6UaB87tmWYAfw6aeDg1NEeJ57 nK1oIYoXBgjjGCaicGCI74UAYImBQchjiiyu1WKPLw7ZIRYz0tgbia5hhyNxO6aIYYsv2gcjlUP+ p2QXR9LYQI0gXrmZdMd9CGUMWz7h4oVUCmmli1Yu+V+RUZxpgl7RfZllmSHCqSKQuO3JZ53U7cjf WN/BViiiccaJKINJbkGnoLUpt2GgKhZX6WoVbmoeniUYKqZDkZLwI5kR8IXYdn9KcKaqqy406gjI eRpml75lqZhgeKGap6WFcXcZsJWNZmCfphK7aK8P/7gK6Kn3mcZraqC+WhiQhG067LKx7pZps6EO lyNryjLaqF3hejsZtbYe29ZmcvYA2LljgjvutWKaS8egzqqbV7md3qZtc+92+VyqZ+Gbr7iD+Psv wLYKzJyvtK7Lbru7tTPtBo/OGzDDB29LoMK+VehnMhBKPMG0s52867L3jhuYyBWUGqMxQnIw8J/q tewuH6CB7FaxuG3cK35ABw3DnSfHqDSCOSNJ6tEHo9Rwgt11a7SHIhBdc9RSH3QnqxPPMGPBKG/9 2kdh7wuzmZ+am/EJOX8Nz9rItu121KnhrWCudLsj775/u5ystT0mKZ50cVN4OI9wN571s4Y6imJ5 g/9XU/Wp9SkuQ+BqTt4m6N9OPeF3AiK3JptOH1jokkuLBK3Unn9eZZAvk/4meyWnDmWGjnNHMux7 18C1tbUfTzrBFoKpZ5S965amnt1qhDANi08dOvIV+5n7rphqv3nXvj8NK3WXX7+77VNS6r3zFOIa nvtXhg5g7PxMz5/hPrKXKJE/Fj+z9phPfpXrDpy+RzvDYQp4UWJWfoyWO3TJ402SwxLtTNefzy3N cl1LTvMYB5vCJY5zH1OU8RQHwWdpMHgqFCEA40FB5PGuTNlbn/aQBhz71YyBNFSfm26GQ+JNjoQ2 7N7zjkjAywUxhjZMIM1q+MMPBjFpy0OiE0k2xNL/NfFm5JsidnSnQAWCsXQMpNzvCudFeCkxjUTo IhvB4MY3aumFcqyjHe+IxzzqcY987KMf/wjIQApykIQspCEPichEKnKRjGykIx8JyUhKcpKUrKQl L4nJTGpyk5zspCc/CcpQinKUpCylKU+JylSqcpWsbKUrXwnLWMpylrSspS1victc6nKXvOylL38J zGAKc5jELKYxj4nMZCpzmcxspjOfCc1oSnOa1KymNa+JzWxqc5vc7KY3vwnOcIpznOQspznPic50 qnOd7GynO98Jz3jKc570rKc974nPfOpzn/zspz//CdCACnSgBC2oQQ+K0IQqdKEMbahDHwrRiEp0 IKIUrahFL4rRjGp0oxztqEc/CtKQinSkJC2pSU9aggIAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0027.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= How to Cluster (II)
=
By topology:
Use= value information as well
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0027_background.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODdh2wFkAXcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAA2wFkAYAAAAD/ //8C/4yPqcvtD6OctNqLs968+w+G4kiW5omm6sq27gvH8kzX9o3n+s73/g8MCofEovGITCqXzKbz CY1Kp9Sq9YrNarfcrvcLDovH5LL5jE6r1+y2+w2Py+f0uv2Oz+v3/L7/DxgoOEhYaHiImKi4yNjo +AgZKTlJWWl5iZmpucnZ6fkJGio6SlpqeoqaqrrK2ur6ChsrO0tba3uLm6u7y9vr+wscLDxMXGx8 jJysvMzc7PwMHS09TV1tfY2drb3N3e39DR4uPk5ebn6Onq6+zt7u/g4fLz9PX29/j5+vv8/f7/8P MKDAgQQLGjyIMKHChQwbOnwIMaLEiRQrWryIMaPGjf8cO3r8CDKkyJEkS5o8iTKlypUsW7p8CTOm zJk0a9q8iTOnzp08e/r8CTSo0KFEixo9irQmgCVLkzR1avCpEqlEqCKx+g+rEa1AuFYl6LVIWAxj NZQVcrZe2q4i1l5w6wMAXHhW5yrQavcA17wM+Eqoy2HvBsF+1QEO3Hdwg8IIGEM4bHax4sQBHJuD HHmBX8EeLD/ATFZyZs16s97tgHcy6c4qQL8VnYFzZdMJGKcefRr1CtcWZIembMCzON69gf9ePVi4 A+IUfL82Lrcf8+bGnyM3qxx2adW5Y2uPzm/6hNveqx9PIT6C8+LVszsD/7g77sbcawfP69584fUV nOf/ZwbfZ/LdB95Y5DUmV1P8BVdZgP2x4FqCCgqoX4IMamcffftgRZVUFlI44Gx6OahhYiSOt9t1 IhII4oAenrgddOFdl96KGcaIY4j0waheivKl55t4Xnk4Y3cc9kijijeSxmN8rSF3ZIs3Ronhkv/1 Et2HHzZ4X4MdQlnej+ytFuCWLC6JoJdnEqighVqCaZ2RcaJ55S4SnpllngV+mSFf5M3FGXxtjrgm h1ru2SWXivKpIX6i2eXfN4I+haiill7Y6HwxAvrooIlWimltlI766agRhvogZZBiWKedpJaa6KU2 XugncJyq6umiuc5KaKwOmkniYY6qeh6dkr6qq6+M/25XK5ypkpmroKjiheyklwab27BkFlsiqk80 2QZ+1Xpqbbe0QrkgpUqi2Wi0pi6bprKwMvhrtuiax+y67ObY6gv9TgHpuPPKChm8hC0L77Wd9qou vpaWWymyvDbML65T7lsmqwDrAaibHksI8lJ35mgjyPuWBqyoZtJr7o5bAmsyyV6K/DHNIetJMY4x y0xrxiOuzDLPs/4LIcc21MhriUT+JXLLQmu2M9FIB+l0X00LnXATRG9Mw9TLTYtiyVVq7K0JXtt6 8rZYpz3VHlvr9xebO+oI9dBjy/h0W3Rn+7K3Z2WZdtZM8fG2kn+XfeDXdjtsdb0+Vm1v4HEvrmLh J/9YDgXmLZc1ob6HE2fZvzV2LnmPoLMNFeE1CK7y3X9TiTp1j5d9F4wL1u557Edo/u3qul+dO5J0 h1405HPrm3TwLnLt9tHGo+z6Y7Yjz63Zzwe9t4HTL8+8yqS/9WZ/N/OeNwwGQn/y+d8nHwLm52NP O/vos379VTSCq/iaz8rfAvnUY3qixC0HaYjxl5Nkxb60YEtKmQuRtsrXnhv4b2+yo6DwUFC4/dwt bBvUGsaqtyr+GVAHtmHc5J7UvwL+74ArlODOpOSZGMYvBhN0Qw1/cMMWKbB+HOwM0LrGphf+hH5W AoEMRTjCpOisgkjsYQFzOMPajW+KVKyiFa+IxSz/anGLXOyiF78IxjBScQcj6+ERWdODNt2MKCFE XK8uCMHKvdF8SoTcFCtmLmpxT1RzBBsPAwM7OkBRd5Ao46aWiKCV7SVYhkSfmn72wn4BL45rGOQf GVGmnK1vh/ljGuXgSALAEbKSONwEb25HQbh8zy2c/N/6mkjKuJiSj9kLWxlVOUlVys2Vi5lkFN9g SUo2Qoh2tKVuYEmsXl4xmZdUQzB/6YkhIckxvmSPNE94vFHGMo2pCAvn8Dee/TSJlbizIByeicxM vO9rxEOMN50IzTugE5021I1twBkf7S2zdYN40RppiIoSwq84r+RgNZ93pIIarYgyoKczGzkmfEpv /zT4QyUCVZfHGTi0jmjMl+84mgU+TXCjIK0PROlY0issrYYkTel5FJpEl3YPii2V6f4YJVEj2pQK OftkCXy20yioMZLoEVtQhYrNUObyqEgFJePeaVSmOuFwM+Ol1Xwq1bZNU5lWZGY8s8oWJ1nUadWs KVhN+LMOGi6QZ92d6TrJwOSZta2/418WpYhWupYSRAeNYkL9qFe38hWqriRgYHMwz8MqdrGMbaxj HwvZyEp2spStrGUvi9nManaznO2sZz8L2tCKdrSkLa1pT4va1Kp2taxtrWtfC9vYyna2tK2tbW+L 29zqdre87a1vfwvc4Ap3uMQtrnGPi9zkKne5zKJtrnOfC93oSne61K2uda+L3exqd7vc7a53vwve 8Ip3vOQtr3nPi970qne97G2ve98L3/jKd770ra9974vf/Op3v/ztr3//C+AAC3jABC6wgQ+M4AQr eMEMbrCDHwzhCEt4whSusIUvjOEMa3jDHO6whz8M4hCLeMQkLrGJT4ziFKt4xSxusYtfDOMYy3jG NK6xjW+M4xzreMc87rGPfzyPAgAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0027_image002.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhpgFVAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAMACQCg AUcAgAAAAAAAAAL/hI+py+0Po5y02ouz3rz7DybBGITmiabqyrZaOY2uBQN1I9v3zPf+DwzSdhCi UFHLLXbGo/MJjUofSkkzeqsimNOu9wvmXXFhm2h5LqvX7DaZpuamD2O3/Y73aa1xERO2lyc4SHhS x3D4RJSlk1j4CBnp4DgXtrglmam5OUmJuXZpxjlKmun52Zd2WsraKrVqAPsTKutqeysWEZqKiuv7 i6WLhERCojgMnKwMdFgiV1kLsrtMXZ3SLEqHrO00bf0NvlGX1BvLfVwZrr7ON3mefR4d4q0SuC6f Ya9OOf6+WBwsXT12iGYRRCPsTbw5+KRtW9HwVkQKE105GvMs46uH/wMPcmxRsdVFhZ+emUNX7ppH hDNCikzIsmS5LC6JfYTXcWVKiDoFxtxSLCigWEJNothjpObNlUoL9ty5FGYRnvRwHn0K9SrWk1Ty tZv6E03Vplk9kvWpExuGkQDDlnW29CzXrXLfYVV7YaRVOkWFkqm610TdTYMDMwW7FuwVvTbjgtxq NyddxHmL6PP0z+1clX5y1GoLqyhKFv885zNmuFPQ0U69usP4tWxryfBQw+Eri1G3ltD0dcWdOLLB 3yrH9V3dmOLjyLmjEt/MjDfz4LJJppbujrOZxdkb86PY9mvSys67F6YcIzzi8bfRrg9mkmySfpr5 KtdF7r5v8A6hB/85pQRmRHmwXwX5oXOgf6Uwdh0yE4WEHIKiPSRggzHMo94R5ABGCmOrDEVgfxZi J56BCpIX4onD+XceFPwUiNCDIrb4nBUwqvJBRQnuZs5Yx/0IZJBCDulXedW9wZ6JRs42ImlEFblk jPV1d+RPNDq145UaxtaeUTcaZZ91NJHGoXX34fQHl034Fp9cSG2mG5cbyYmIbU8CYmc2EUKlxZ5+ hHlnhm6WWSdSfaHip4J9ZggUoI34OeiSFXrxHRXzxTkdnSoWhKlUhliF2aWI+nMhk8J0ip5goHpa 45ysftRpkq1WiQmqVN4q3BmECnRglppuKhatI6451a5NHttDpZ3/NErqlLXa+WFtJjKqImzGMjtg VGxBC4etr+WZa7WrvmpmQOR29uaJejUEl2uLCflXlECBG+66HbQb3LtBxutev/LqUapr16kV0Y1Y lqtZIErB9lvB0YDobK4K/xrxfwHnmyoxDotz2ZmOhivOrJ0RKA+9wKY0McX/JnsxeYn0UzAH+Fa8 V4t4cZriC6kxCLLK/gqhbGIxs4vPzEvAm5wLRSNbp8wDI/1nyBdPGJ3PluZstIEQK7kdwopurZ3A WfPHNFdji3v0yabe6q3SLbfndLYci01feQa/YLJ4RNMtsnE6T/0zbbhCRjguk+rJ47mFL96hxzuu mDHjknNyeKIAt1s9eeaDTPo45CJrDjohnPfsNuahnw7KmW1C1PbKqL/eRYVgnlVF0LDf3oaAhwbu Veu84w48a58HTmxXvgsbfPKeD/9symtrA6bXyk8PtMfZrQn1MNE/T333yw/uIPdopWuZ9+YnHvnI NEPnPPPnvy+49NDLL1zx6cOP/1E844xD9h+TroP8CbB09zNbYBjmHeINcIFOKqCeEOgv8tGPgRQ8 jQMttqwKahBD7sNg0zYIQhFFYhUFAAA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0022.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Experiments
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0022_image003.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhWAERAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAABY ARABgQAAAAAAAP///wECAwL/lI+py+0Po5y02ouz3rz7D4biSJbmiabqyrbuC8fBTNf2jef6zvf+ DwwKh8Si8YhMKn+IAMsZi0Kjhin1ilVYrShuNuWFhb/k17Y5K43Lo3XLzY6bztXaAnfAV5voW95f Z/MHmDNoF3goULiXx4dIYwj5KKkIyDiZVmk5paeZuOhJ2TlKKogZKQnaqegY+kEnFwLL2qrV4DYG J8G11uvgywAcfEs8XGzb+PhaG+sxq6vWAd0sB5uWKc1MvfEsk70NnmCdyIE9Hf7QLfaNHv6Mzd4e P/sWL99Mf++eda7vn3zpH7h+cwQahJDvT7BcDG81vNProRaJE48hu2PxokZx/xTFZdQGMqBHYyQ3 hqRVcmRAVZtamgolyuUnPSxflrJZi5PMmDvhbaE5CpVOmzd5ziTac+hRUzXJJTyIRZ0Ze1Cjnqzq DSBKFwRTYqXydNnXCVK5Uh27TqsIeGh/5czKre2VsM6uyhUJ9+4/uvP0urXqd+9Jtgohvl2I2HDi iooZNzapdvBfjIs5foyMOfNWlZA1C+uMV+vPpEJF90G61OihpqrNBTV9KjZM1y+3LlJVWjZr2ql5 91baOtPusl0Cww6toitl40/symJu23m0ctCba657fSxx63GrJ5eewSdz8XXyYiDsnQRf6tm/bl+h HHR6EOu7Iw8cf+38fZ47Fv9e6VBlaAT42IAFGrgcaJ8pSKBjDHolGYT3bRZhgg+uVJRv5uSUIXAa CtdhbrNxcthsIg7HISGvmahbiC2i9tuJQJF2HBjQvfcde/tNN6F+45Volo47tgEedsbhaKOQQz7X Hn3jUbRElDogOKKUVl5pZZFnyYVkcUouKVaPTKaX35hgshEWepUIOJh/5bH5poNUWiYhhZ5NpqCb ax543YIVWvgnZxfa5iKL0QnlYaKIYqLoizgdVyiKos0I44cASWpoh41mSmmMNRLZZFVdnlCmfWcm WCpCWho0akFpnYoqqST++Cl3XKWqV33h6XoQeXtO9SqsnOGKJ7EDAVlPsML/1uoqmfwsS0aaeqpp Z5x0yjnnf9duy22g3n5rbbbhaitouXfWWW26fv45WqWbvkMjpyrGq2mN7ya1KaHzunvioo4G56+8 v0GB6aE8potWqwcnCy2zToYKlcLqmQctr+fNeiSytr5hbFvtRkllx/iMQzF8DWtspsj4oJwjsCcb nKSzgL0sRZtw/prnzdSqybOeOJt737rn8vkzuUb3eVnQeAKt7tJAYrqivvz+m0qk+Noro7v5flw1 vZsEbLXWWQ/MqIgwgyomVhKjHeTLFl/wtkBr++jyyXFXsCGtACr7hMrurYr33fr4Sq2XdTcsOFmA C+ZwzG3bvXh4+/kNN81g/9lM9M4Ndjvu0UV3/jnoTdeJS9Kh93wzwkKPTrrppZfEtaW6pTi1wKuF Tfbel+DuKYb79r4i1F4Pn3vBie8S+eAsO84w5BCfl/w9c5vZPOLRR5B3xo0bfivNxyP0PTqEU459 ycKGn8717Uz/8OEVq08B+dTIH7/lNfen8+qoE336tBG5DsAAZs5/pmPdoA4ILqVNiniioB3wrjai sQUHYwbjHcAWyBQLYkx4teNg8Qp1NrohTDvI0p/T0jaxq5gQXSsEVEjQV6wbLY971Xvf8+BGQb+w L0zdcxv8Nke/L4zPfMyzIQold0O5FZBURDwVDJFnv2ZFsXlB/OEUXXhFJv9i8IGP8l0HIejBCSYj jB/a2mlux8DskfFeaUwUwUD4RNjJ0HMGZFq1qpiS12GxjkxrIRbjGCu9RauJZwKkRvA4s0HybVmG vBYi58IRLElyCY6cpCUv2cgZkjCJCbNcJtGVqyxOzorlm88jRSlEB14wU6qUne269kWx6e6VbpRg GctGtVrm0pauZGMG4xXClM3xiKiknh8788lZ6pCUxVRcsOiXzDJE84Q+fKasOOmPaaYOcafcXl8E KQNtKtGaJmsmEolpTuhhU53ptIA4iwmvWH6ilb4UYwR3iUta8vJd9MynBuvZy0jt05+WCGb7RtjO UlZzaPLJBTMPmVBTWY//nAfrJiHLiU6Z1Q2aObxL4eDzzmxa1KA8HGkPM+qMkCqPokX06EOdidKI KhShB13mOoVkUvvFk4vz3OIq63VPffqzlQNVoBfRKE97CpWW/MzlG4EZTZWmUorUhBUvBPjHl2rS pVRdIphMGlWtxiGsM53oRq8ZU+mRVVXeG+laq+OrFLJ1oY/LaShvWlOZGomm8BQr3vQqwv8NMHX7 w5bnCss5az0FsYnl42EJmL/ISnaPqzNg7ACKWcz6a41gI6pTL1XQyxI0qT9tY1OZOlqn+NWqWpyr 884KWFB+SbZLAiv87NrVyuV2j/XjklvPGISQTQmTShAucX0wrCsZ97hY/0ouc9E3DZVBc0vZbC1c bwvFvFKPnRGz7jClBjyzjQOqwAURvYp61IC2cbPBM61nUQLUpd4Gji91qFo0x1u29fGulPWqSCp7 TBXKEaN8Td82cLu+7HYXr39lcIHvN8Xo/s3BEqYjTOWqAQTzUL+bY5X6PupNBa8FVxqWxnTBN+G0 gge63uVur1pMW/Fht6zU3es5F7wwQQrWQWmqqv4gq0wBD3ayjUVsj1lI2AIGGGkDTm0DZVNB+rL3 vJ/1aWZ5OeUu/vO9V8bnPY232mHVeKoRFrGHHXxhEh/rihUW1YxdC2LXcli3bt6tYavL4DiTtLfq UbOoTpyOEn9Tohbxc/+O6YxjDFd1pSoOlXRh7M4UK1rHHdbWkVtHOiCTVI+J3ZmmO40kACcZyZjm 7WdEOzsoj7d24IVleKuc3i7Lt1NPnnVoywtaWdr6l6wOKYsPnMU2J/qRv55fsM08zkaj8Cez1W6k 1QbpOzP6wdvcKo3pY+heAfoXgsZpOZST7RGPeX3bXrT03vy4OT+7zofGD+Z4nEDHBk3Tl/6jzxhL RyMnkNPiMmp/S/3vWJ92y6ou71M7iN5Wy5rguPEyKBSe69ztGqm92/Oajc1mZDNO2ZcDdsavvXFq x1jdG3bxmdst7XPn+RzNVUO45Vbuasuj2OkW4Y1fbGc7qpzjts35hXH/jnL+qgtK2NtxkSsNuriK eshMLzLRGWrHeysZq1EvtMH7tRBdRw3VUEt47NDL8PaSVulfg3XYyYtmW3V75IXUOJ45TsWLR1HY QBf5XDq6XUTXneSN3bndpZBMPRs4YjHn37TXLuc5vHwvhc/vzMN88gxLmu9cnTeR+90/wtI73p/u N74fu7ylS1v0CGRyIJ2sXomfHeGwhrjZ3RtUzebT9fCF4MRdvUqLG7H00Aa4zu87dd8LGaKxdfz5 3H54v9PQ5MmetMxlnPbkIFLw5iZ3tDGffOtHP+U/h/n1IU9gxA/+793Xe+QpzyXwH7TxDWa3891t ZdmjNtUCR/0GbU/7//iuvqe3J7uWbb9/uSdVeWdnlbV3+yZipLdfpEZ8IRcL4lcNyOd3EGh1PieB j2eBojOBMxd4iydS36eBG6h9cPd8icd4IEiCiUSBTfZ+7Zdo6Ocx6ldSKGiCDmhzlAY262V/9kJr veR1uPZ6Y3dw/EdwsRZxShWArqR7GEdgC5h+wudoThhwvIccA+hs0nRsIPd25JcX7MdnL3iDbOdx KQgGHThu4uOF2IeBZNgFZthsg5OG29dxbMg9caiF0waDnSSHdUiD43eABGhT9edwAFh2gth1rfd1 QchTFFdrRZh/YAd7YLaHG7N8ULdJU/h7/xV8mDh8YtZ8D5iFNbiFK/94ei34hX94hX23hlxoBm5I aCeYgdQ3hqx4UrTIbWfoDnZoi2S2i2rXh9WXYBmoh3RoXbroh+cXhvAniP2Xg12kcEP4Uz+IZTtY ZQ3XfzulRoQohNQYZKvYUv42jFRIdwqIgEljhSU3VqF4jM1Hig1oiuVHV8AIfcTocn4li1n3XPnI XJWkj1kyifX4j6Hjgjp1gcFIjxX1i/7FWhrVi+GXkGJYW6bUT4OodYX4jNwoa1G2jRSperBnhIpY WjoIM+0oj+job1cFhZaokispb4pkKudoYxEYkGD4hN2YSCUJYQ0Zg9ojcP3IbIjik0Ygg3joURSE eHrWjjApd0KnlAr/mZTq+F3iR3cyWWYMaXSed2+htoni6GNUJ4VfCZamB2pIl3n883mzxFlmJ16t oH8eqZGLGJJwmXpKlX/aKJfyt0MuqZMz6D04eJCOI5VDqXyVt4TuAztHKZjeWJQh1orm9ywziYrh 2JQQuTIf5x2TaXzzOHcSGX/TOH+1N3bNeJclUoQaJJpxqVqemZaxR2U2iSZemW+wqUxLxpLkWJuz eGOYGYh7eYm7qZtM6Ul+yZu4mImwFY+L6ZqA55gzc4+XeSR5WYmlqJckeVHu95srlph3uHtRSZZn 6WmhN0O0qTqyOZ4pWZ6c+EJZeXn5JnWdyZFIuJZHeEvvSQlviZr3/7lwbpmIn1marVmYoDiL1LlI O5mcS0mVw9l7yvifNcSC0xmcz8mYtRhD6RiCXwWhEVpzDcqLAmqYBLqglKihj1mVzok/TYd5SleB EyKe/FaO5mmb4Ag0Z1mWd/adZjmRzDhUBSefunSNpOmRpomR+LefbbmR8mVtlcmECOp+wAmZD2mc r+Wb2SmKnviaTfqJyPmh0qdggVmhEYkf0EmD10mZ18WTWdoy48ehUMqd65l0RldvdLGiXWmeLfmi LXlH6mmi7DlqywiksRefu2OX+AmNczmfOMqaRVqXFVmkAoWhB1p8AyqZUhqTj8qgHiqmyyap8LiQ X9qoILWlB5amlf+qHWAqjJdqpmSqoKYqbFz6l5HJpHnapkDELp0YlioKngzIleTJorHKpjNKo6d2 dfTJo2a0o6dJqCgqrHgZqMeKjfm5qP5pqpqqlwRZptFakJBkmVFqpeuGpgcaqiDqMUY5VqtKlc2J qrlCqn32qRQqkE4knKzareyqpDSZqt3ZeZY2q2JZq+e5aVu5r7sKo+mpeb1ao6N3o31qjauwakio n2z5oyK5mg83pMvqrK9WoLxIqyeqq3JqahvrsZmZk7mZqW84rQ9ar9s6kCmKhShLlL7VqVo6U6xq rt+FpaeKVnBWrt/ahEVJMCMbY6rqs0iKrpAAtDHrrSaLrnvQrMpNqnV/CpodiagLy6yqSbHSmKxU u17XcA1BW4MAK5Bx+rHoGbC3OabgWj9Ey7XMV6UjOrRby7LJuLJsO7Q9GZR1a7d3i7d5G5SUyrfp UQAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0055.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Performance
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0055_image004.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhVQHxAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAQBV Ae8AgQAAAAAAAP8AAP///wL/nI+py+0Po5y02ouz3rz7D4biSJbmiabqyrbuC8fyTNf2jef6zvf+ DwwKh8Si8YhMKpfMpvMJjUqn1Kr1is1qt9yu9wsOi8fksvmMTqvX7FPg3Y7LL/C5/f541/UBvB/f NxD4R9imdzBYqBiXuOiY1viowUdZaXmJmam5ydnpSRUpiRFqCCraQcqWurR6KtEKUYcAO1sRKDs7 uEpbO8Xr2vC7cJvQd1gpCGesjGucbMDXe0v8nLy3PGwKnKHLWQz9fWhNrQztjMhsDr6ODi4bLVit IExEv907Ku+OqB7/PO3PHLZqutbt4aev4Dxt9yzYK4Yr3jF4AKdFdAYwoLxl/wPHuSP1MEjIeyNL iMtR0kdKVyu5tNzxUlLMLDNx1Fx000rOGn0E+PwJ1GfDhZNC0dsZAukMWAKGfuPGDuFEZsesWbVp SlNTpwjzaTR4DaO6iigZNtjKNWDQtT/DRaLmT+y/sWVhZEIVC63TlCfRRfsb9xy8G0pl0NI7tDC3 iDQU242AmAHbtYz8OI5FCS+EyZQfREZz2UnoYPs0c2BaGVBjg6MOeuD02Yyn2bRr276NaXXcV7nw oQgQO+3efphzuU4BnHNb4cyDMbao+LByoc0Tf1SIvEPw6mcym/6tnXvqrpNYrNwufkyjlqPPp1f1 dEP78O/VeD9tnn79tO1x8/9Bv18XuZVXD4AButRVSaOhMt1yB16xXlQOFcFLgw8yEeFXtlCo34VJ eMcXhxwY6KGAIm5AYolbLIhfhyqqdyKKL8oWowYpzqiTES3diKNZQuzYYz2Dyaeji0H2kCGRNWLA 45GP+QbVkkw6+UOSBNbjQZNUtnCfklgauSWERWpnYZgvWBmlkJuAaaYJA9JB3hNAtsnla90wpsSc dOp4HBQvabmnbho28WegefqSZZmGwuQjCYAu+hSecCKKwqOQsjbohJSeYOml5HznZwqdQvppi1Ks yealE/R5JYKpqvqkiVnCStgXMY26aJetrvgBroGi6ZUXhdK6IZTByjoisZr/ZrossjIqu+qbkwqb qKLQjsCiDaH5OmNVoPKqArcqluqls5xeuxuzFGTL0wrilkjurq5Wiu44L7Hbrqj1YhvGZe/uqy6E qCZ7LbDT2jerstIyMKQ4+AIxLLQjdWNsdwkXTFxxAkGH8Ku5YhrBWxKCdrHCIDtXzkkPV1kysfGm 2bHHhrIKMyS9AgzCyirdjPNrpbTcc80kAx30wTYTPbObP8tsJlw5L00wrQsfezTTdOakM5I8Fy2v xUh7OmTXXkftcnxi02h1mwY3G/OzJlfMdtUoTjdz2EIPfS7XeYyXt94M823Cv0lDHbjfpBFeuNr8 Il5Cg4I7gjXgMjyOEzKm/66RtQGUB0iR5DFs/qDTbc8AeiHetrYx5jiUbtnJmmbOA+wDsK5aurY4 zMjA+sLrOtV/jEY7IHb73rq71g6uiOzBD4fT6uLpenftNixPxtrFJu98ddavC3n22scZ/R3KM3fX 5YSM72n3N1BfPfRG/+49+YtjHz9/89M//fciyH4E+s1ZXi7T1Y8rkVNf/v53PwGujzs34d+YDkg+ AJ5NfAO82iOUd7xuXZAI7KuCJb4FvyF0kCY+M2AQRkjCPPQFdyYEAgqx8AuKge933oCgTNz3N79w zDo2fA9IRkaSCv6vKioTjv9uOLy4BXGBoiggAYV4vgQ25IgtLOETmYjEJP9ej3nTyyADjchBzslP CC+s3hhPKEYwinA/DtwUGk2nxb2p0QE13F0IZxiLMx5uj/SKItzkaD865jFcycMhIK/IxxzaMT1t PJUgB0lIH0awfIfso2UkuKowAc+LZgxgkKjIEk1CcYNbAqWwLuHJT45SC9hRUNNWCcNIrYuFpYSl mKQxqUaqbn2c9FPnlqXL0f2gjENIRBw9FEwFEFMktKRSMhOwTDccM5OiDKMHzRY+DVrTR+xpmu5q EE0SbI+asOKf41gxTd6UjYPnBJfU/APPeGail/0z3AWn9j578k6fOHomP231T20GdJ8DRWZBCXrQ A/kzobFkqEId+lCIslH/ohOlqCQtykiMThEiossmOVOpxHyGdIsT/GhJbWJMY1rxpOoE6Ui5Zz6P ttSljNIhEEVKUpmGbKU6hSRNTdpTbWXDXvIsqlGPitSkKnWpTP0SRzQK1ahKdarpnMdUFNSXaHHE HqicUFd589WdGucoG8vq3sKqQkqOyZUdBetNH9m7QdKsOBmJYUrfalV+tNU5AdnrMATjV74S5EN4 lWNg+XrYv5YjYAwTDJzqClOJFDZSjuUeiGC6WLJCFgmJVWRnrTpXw9qOmqE9az+EUSrUpo6x4Tit ZV1LWtj2z6whu2o+qspRj9wOk2D9YCbRCsitFsu3veUtyvA51eQqd7nMhW2uc5+LM+L+kbWRhe6h iMdS67IMu0HVbuwWco6rSGggCiFvQZrRDOp6dwVvIRd2XFvezbqmlQtdrshG9t6+6ne/toNLfteb L3zcKb/nlUqC2AGd/wJ4KUTpb5wKDLLyGrjBCw7wgQPzYL1OeL8SRnCFaxUfihFYKrSML3pV+uEU q7g+BQAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0056.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Performance
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0056_image005.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhZQH7AHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAABl AfoAggAAAAAAAAAA//8AAP///wECAwECAwECAwP/SLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bz+i0es1uu9/wuDwZ qAcW9jtBr+DP/4ATenyDe35+gYmKe314jYx9iIuTcoWPloYMdpScbJiWhHWOnaRpeYynkI+rpa2T kq6xc7CytW60trmmurxtuL3AYL/BxFvDxchWx8nMUcvN0EzP0dRH09XYQtdJAt3Zvdvf4jrhTALj suXm5wrs6K9X7hLe74rqQvL1tfdB+Rbd/vSt4ccgoMCDGwi2W2AQoUNBIdw1fEjRgcIGE1tk/Ffx /8xFHhs5dmySypAoVZtYQQkpcuSST41OwvTA0mWymZhWEfRX02YunKwOndzA0yc1oCgbfCxIo2dL o0RO3Uk1KKWqlSecQn05RWsFr1uR8AMbFt1YCGTLMjv7IG0/hvQIAIQrMe7cdnbz1t2rFsNSpjHc Tohb5K5cvXQT4+W7WPFhxo8dG55sl8vfGoL7hrncVLM9mp59sm2bObSt0WgZmv7GOUJpB6+JZMwH mXJtxI1zR9ZtWzLuLq1dZ/36+7AegHaQ11EegLnzbsmhL5fenDr16gKi57mep7v37+DDix9Pvnx3 y01/P88+Pbt769zFu5vu+7b9+vh5q99/X3//3f8A9sYbehxoFd56x0HnX34BVubEABBGKOGEFFZo 4YUWLoDhhhxmaMMmwd0T0nm6EdHhiSim6OEOEX4BygkiChIcDBCutsKLJoToQmwODGAjCziWoCNs JvCooY8/AlmSkDC2VSQKSCbpTAcNwWLkCTVKKUWMEPBxZQpRanmjUk0W2KVqRmQppgrhlUmClyqU puaaQGpQ0p2a5LgjBxRaIWCD//0pKH8MUkaHTpccoqdGGsxJ5wzeUYDUVeoYNCMIYWZA6IKcAsog azFVMKlQ4Vi6UA+ZYvDlj0FGMGqeRJ0plw+pPhpDq12mRCKlb6KZglO12nprJCj8dSlGGQQrLBX/ xr4ALJXLFlHpA8d6oOxg0X6ASGvTknkqD9fOky0HbS6qgT9wgjsupC0sle6vya7LbiZuglDttxeE K++iOY3w0b34VqCvcPvaeUm9F6AbxMCpFZwBriKUo7CvODDcsMN+9etvrN4GbIPFGBvxL2Ao1ARy yNKGALDHEZyM8hAXrWwyaC/zIHHH6kJb8w43w8ryDC5TvLME2yL81FWrYhD00BeUy6S9JA8nMNNj 1gn1nlNT/YRCK8+aNcdaE02s0RTk07UFS/8c9sFXbawpzl5/rPPasmqs8rlwJ4023WXabVIokeKp LbIl7813jn5DQsjYUyE6ONYTpH341YvTW5XP/x3I/LWmetP5DE6FfDLUV3BXDPbkFlkAeqhJvT2K 0ECj/mak1IL4N1WKuq5S5xRILvu8mRNeeO+q/n4DQV1vBDLvj1a1a68eJE/80cYDnzDmOfhufNel Ys98y9VHr1KxeL+utgvXEha+ReCR7ZfwT07v5PqStvs4o5HTn3LwzsqvfxDbkMfZ/Pe/HwTQezdI XwFT94IDmi9uNFDgAmF1L34gYlUhUdb31uU06JGrfwScoKjsxz/8SUB7IiRf8R6YwBTaaUkefIof NvgACbrwgdzSHdJaSLAbqiSHJYRgCTIoLh/+0H2qg9wJjag6GLrtgyYEHxPJscLxyW2KObiGAP/h J4Nw0TBsd/pOJConQy7GwItYTGLrZNK2NjYtavGTYg/TCJE1hm5099DcEulIuVGwUXTlAJjy+Eg0 UeBidahwXBl1gEL9QWyMjWsfvaw4xxnWQF9f5CDbnrbI860Ak1h85NUSAkcSZASUoXRixIIYp/wR 8kOshNceX0kDLWIvgrTM1fOeSDpWZFIBqCTkkFpiSVxii5ZANBjsRnDKYyITiXVkwUQaWUDameth SuxRLmHZyStu03rO3GHsishE8UATfsXsYji/GcP3LTMis5wjO3npzhVMc57sU+UosSnNeOIzdCrs JSVphE8KDlR8xCxlC4KZRlHer2zmyyRDGzr/tmu+8Z0guCc5C0pPNdpTjhxtICk/ClKOji6gED3o QkOqlJO2k5zpfMFEX+lSSOJBjEG5ni/PKFBh1tSPfpyJSmXlydNpM6V8/Km3KsfGSZqEn7KEADVz icjLsXCErSwpS7HKVFAMZRrsGGADpurCYUjldlbJXS9julKkPjMFYN0pQd1KUzYps6jlk2pPt/pQ jw5Pr3wN3DmplVWt/pOBFu2hWDW0VUk0LrEXY+sn9zpFx7pxnzBVKJgoW06ivrR+GDXqUddZWc92 FLRSM+xhS7fKeqYWsIHd5Wfnt1hg8nVYazXjZul6W4Sm9IIy5WxvR3pRvOpUtbod7l39OkRX/yrX asVtrnOfa9fcGpO61f2tZrFEPexGt4iS3S1vvctc1ArRTCG8GHlda15mbm69nMSqdP0J39kSbrFk ZWcYaZdW64aWBAz7peyqqrirEqltzAswdQn8OqtE07gbRW5yWcpgXn0XntOtLwO7usmh3jKj6dWw TaciRsEdU6z5FfFl77vdEih4CuaJsYxnTOMa27i2ItXuf0VgMQEPthIEki+EIzza8ZKEFDh2wTRa 08wxcM2l2wjHYx37VVcRNwtLjiNsxYA8HKIERI9N5E0XF0nb+cx5N31qKJz64Sos4xwDTLHI7ret MLPuj0Gy3ZTVOuaKRjImhwwywYA72eNOaf/M4+lYnb9qSMZZTieN3vMt+wtohyaZhOCdMI+ruKXH UdmgViXxJvWsOLWiWcx2/jNhjfFdwWi0u0548klF/bdSA7rPjrgjm/tr1Uzg2cqsLi+It+zkqABQ 0A8WQUPkbA0AKjWLyKZtcD2C5GgTSRJfcpmPUQqPYIsL21DKK4ztBWUo7k8Lnzvw3BgQtG3/WGxH DKpM8PjlS68a3cL+gD+YrQQ/nEjRAM8drVExZQNa+97uHetA6Kzoeev61onjJr6xVS1+99vTiGXq rSHuYRnYu2rf1jSf0Otmcs963rUmeK5P3YOPZzc14Oaxozg97nPb7OClczUBtOfud9fAwTf/9zaw pWtxktz46EhPutL1yaz2KnvnKm4XpV1zjWkw7DKcyTpkfXvavkpDkerVrNXzPeSOs/du8UW7fWPJ ldbVziCmKpvFfmGgrXO9tV2/u9pjDeml+/3vgA+84AdP+MLLtu0uj7qSnq14LAPd69pifNMeD/no ST5j/7r85DXfRGWYXVLNWrG58b530ud9pO76/P5Cn/bIz252xuK8kCMW+sSjnfWnHz25fQ7a1EMv 9rbvY+6XK3za+/74w+en1plFebZnrvkJWb6/sI572sO+8djPvva3z/3u63dcOBW9OIIfjZWR/yWy t5H5Y5F+9W/YcqRylcNxrXJP2LRgloX/XKjvLdRQF7vOHaZJS9VgYMdCZCRUm7FqEWcr+XeAkEQL U0ciZJSA7yd+zVOBa3RRE1iALqKAqpckDUiAbndmiuR/XNZGznN+0BCCJahUaQU4K+d9nieDV6CC Y5AAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0057.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Car-Hill Problem
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0057_image006.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhPQHgAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAQA9 Ad4AggAAAAAAAAAA//8AAP///wECAwECAwECAwP/SLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bIYF04Mxuk9buuDyT ntvvErh6z+/7/4CBgoOEhYaHiImKi4yKSHsKcB6SHJSVH5YbmRqbGJ2emCCfRaMWpRSnqKGrk6wd qXmiTrB4ta1NtLa6dGe5u2y+VsElAgJRwyXIb70myr9DzjjRzdPPQNU12MsOxhHd1kLaM+Iim98P 5+A/5ErkdRjpE+zq8swj7wTn8ZzzBNPR/6TZu7dAX4h+9CIgbLHwkgKDshK+aLiCIqeCDPY9xBhL /6ILiylAcoCIjqNHgWYsatzI8uQNkdfQgPhWrMLKihFf5cw2sIEaCitJSrjp8uOskP6AestYlBSe NWtqoqNJtWDVhzStNo0Bs1lIfEtN5ks6VmxWpj63oujqI1VQtJE6kUx37t3PHWzX9Qyb72rLfHCk cusm+C9ZqDzypn25FwJdsQ0ESCJq2IGewztpKGYQEJeLx3/rvqP84O5hsGpdEYFElgXosn9/GjuL 1bDpBbdT30py+YbQybD1EV5subXu3UdUwm35G3VowqIz5WZM3cZmh5z14DPNHXVUf2m6qREfGZ/k 793LY5T0h7NPsNvvwo/Pvg5U+sR/3q+f3ZJ2/P9x1UcJgOD1x5oRB+Yk1Gs3QafQcqZkNg5KyLVD gmBCBbecVMWQFtlxOHl2oVkfhgYhbX0ZZgxC113RomozLFgibCnSWCA8prwojBlzUQWdg7UFSSOG NXW4mF+FeWichDK8qKN1js1oI2QoDkUchFjmhxuToMDY5Bkk3ZVklsw9CJk57qFh14BbVkfAABN6 mdxMcIkpZY9aZTAXmeCxZwSccPIGZp6tEfkjVkCOdWhfs/mYJ5GELvkkBQNUaqGIHmg0HTcZVYnk cJ2aJCNu2/1pqaA8RrkhiTZdVdiUsFZGqjOBlgZBrRgAyoubYwBHJZmgvcoXSxnyiVhxFeiKrAP/ ynakQLMR8toFnkKOCRlawrYp65BpqrIJLJXiSumpzvIkJ4IbjMrtqqL6dcGe14oln3MThJsruTiI m8xRGnzKKrbwKkoohzNmGyuN8i1J6Zu5MqwDtNswMQyS6YZapr82coiicPFGonC9DlsAsQ32njCp Cp+89W+xjrn6Gp8YKkQvdubWi2/EWmyaMrucwuXhyx3XeHDC50Zb4a0h78vFsRgLnWSVAJu57W9c wgDSyB/vKnGEgQ1M8LUbP2qwldVa3LOJx518UQV2eo121KIW7Oqd+xisLttF6PslpjjyaZPZDM4N OG2Qxrbmt2v+mfTeWmjMFKiMIiq50xRbdaiR/ylCjrmhQiZ1WVTRaVsuDSVrNq3UZ6+buoZ/Z2k3 isfG3q3HqhytQelxbsGgxYWXrerQfsub438BFk97iB5gbbROfJPte5DFhl2tkmZBLvTZ8/amdgbK G4XqUC5rfLm/jS4as/lFOio555QnWlhvHm0vj5+/q76tBRyr6HbGlVM+8Izwk4kt5FcPpQQtVupi 2VQeZT9gQY1t9PqEk0igt9o9wj3EW1F4jAeYrpFFg2MZj8cmYx5/SCU+PqnLbPpUE/hgaDSIEc/h PEegAu1nVrPTT59I1Z9uHag7A/LTGgAVRANtChr9KhH7zvc/VY3NEnVbDiyotqVvIcVNuMNb8/94 5jTeYUl6XaTekKw3tqYJhmjH40IWH4ArAi5PSpHBWPQ6BzX3WW+MC6wftXZoHP3MLArdW4Cy3KjF kqysLGJ8TtwQKTiMlK9djqSjz74hOxwe6wqBfNapCJkpYCESjoTz5OA28higRfJgsQigEEVnOhmU jJN9U6St6se61X0Sf3C83uPehrAZ9tE+pbKd1mAwyCmojCNLzIq73vLEXNpvilLETCm4UzTvcc9h sHSe/fK3xC/6LJJF0pPlTKSRpgVAMiMcZhmyOSwF+g963hQYMnu2QoG9LFv1TF2YfJlGKwLDCQsK 2P1aR8rKDOegGWNk3BB6vX0Sj486M1AY2An/FPUZaUyac9wj39lFrTQqch+VZ+T8F1KOvk+a6VyW 57JmMgplAxmUCJwADblIW95Rm1Oy2y15yabPpfRKiIEFm9YGpRwcUWYepGmZGMjQhoZlJVA0YBq1 WZceQjSCGOyTdHTISmsyDy9/TEtSt2lHMuoTW4rKJy4nZ5Cbuu+TMtTeJS6ZTvhszajBrMdYg4I+ sypxXWdh6AoDF0W3/kqelYTpSrNq1yU4CaajMVstBypVecZDsCItqFMtC1d21ZNo/kylpGg4uwvy ChvuPGYt3fkhhKoVgU9d3lE5OFsWTPClfoil6vDpK56Vk1NjW51aX4vWfxWKazikmXJzV7OK/3Xs su3TbD6tVZJmfpOna43NAbM6VaI2d7lW02ptR8nLenIznmgzLyqVOthd5lS6nV0vbcP61dUAlIut la5f0QoR6HL2tW0FrkE9u1M+ttIpTQjodeMrzlMGFr7ETSvr1NtUwK7vp8xFIoW0wbIKvw2MoOws gN/rRKOBqpKj7aowVXrg7+YigZqi6nPpGbw5quxuaEKjf0pbSBajrsVAfrEoi1tQCp/yYPtV2E21 S1CEIfV4owjGbVdstQRNYHOOg144IYW5d140c2DessvGWZsui9mjYR7soq5Hv5lO9BFt42V+/9vA IhOZVe1dFXQN+ysP+3B0OGuLafcq2QJTVv/GfCZWZnf5YFDlWX/DyR4HfxzoHuiIDx1xSy79HCzj us7O/LOley242MSqlEWUZpyLzcQedHawIBs8Dwm7FlPzbHCxr46EZE54zri8h5IpEuF5sjobGCYl rsXbz398quxJBzGoKEwulIEo7Rvh+pdGpC9DrvQesDyQxAFWL4NvHM2/KhWawksuaI8o5TdW2rtB fpdCyyvhMtd5sq718++2zN/q3nK6khbrHC5t5Svjl9Fqzmxh4xtKRY/Y3n+NIoncRVcMt4Gi7B21 oeXbzo1PeNHgnCedtxmc0HXbzWC4DiBGkFpaai54AhZ3vBrEcZaOd6hccSkTRkVcV5NY5H//o67L A7wUHD/oFDi/QGeKet+ZA7cqeXb0ZAdDRSJX3eDptpUv2i0teGvGPjjCLJ65NdyMK3rklX00g91r 5M6hsYBiuG0AsS7nfrv203eHVbgTeuR5SpyckXr7yVOCkrl7g3yZM7NmTSrHmLE1fZ96ZEix3NfJ 6ZLH1e5V4bvbgZYrdaes9bfMZUVzD02T86gPb9ebVPC099vHJzq0xokCtCUv/plWFbxVNVG1iUTh gQsPOWcVXvRy39n4dHdyFS3e0tV36eudv5yXxefR3lF/pJS3fJZHmvjwiW3NJ/XpLx47sYP7XcR4 h3DG1U52Gv891OrWNsrmpAP5c7HsP0/i//nPyn7hqx+eVGJy3QYTFKU29tdk/nV5wwdingQkdPF+ g2Fi8yMdV0RqQGYdyCA9PddbbCd7ZEZZD3h8ccRFHIgsFLhiQqVz43CANAWBtxd1/3ZvL8hjtldV CIh5/cRu9cVtFbiDXOEMecdz7UcZMoJZPwM3zFFOTXNVSMeCLtJ0dYdM1Bdhc9Z2+DZvkMR3CkhW 3GV4Akd4CUZgR+Yu8rZ/V5goEaghZrQhVSVXnFdxz8cvWeB5ArZ4oVeHjud0hydbdjFCEeSEOBiH F7gERagiJWiHNWd58nVPAyV0Njhpywd7gsh7zocEc0R1aNZUasdXVzeGyCdjKqZ1zcd0Pv9oif+X XkPoeg62dqBIXa8TKaYHiD22RfZFbyV2d0TIcHrHdoyohTp1WAGCcfXHG4HReOC3P4z3fe2CPrDo gNa3hOkwXnbATn1YhvfzW8rkaUrxVp+2hz3mhR8zZd+FYL6mQZS0aw9RQ9uBjibUa9rSHSsEFTL0 HmVRBxrkjj63WMUAQ7R2Iys3X35YRDyUbAb2SwL5Q+rIH5FobeIli11XRy0DN9MFesWXdRr3iR0n OhLEUlTmVeBlBIVjZuA3ZtOjjNh3fTXCPilpRiwZiHAYB8J4JzDDVJ3jjVZINzXZX3KjRx8CjgU5 cFBoYX23XRlJhdzEVxeDjPoQcCcRkw3/pj81iYgMGDTbZ0qo1Ezi95GqR4pQSHR0cYgiRYUtc4e9 2CBUk5WUWIrqlGFhOGB5RCyaSJFR0om3h5H1w5RsqWpaWQRzA2NI6DxeWXLZhYWCeZe+1Hq64JTG dUzXF1w0GZZsmJMMNEkZ+VPSCJRtiYqeGJU3CJnVw5kAWJKL2YWpVwuKGUuqNWoTSZRpmIdTg1NP plWTqJd5uXOfOSWlMJFiJHQut2Bmp13OABC955GW2JINRZLRlYyiuZIfaJybxWYc+QynuVt+KYWd uZyOiWd3I5OZN4pYMJ2QQIchSRrPqEdoKJTayJrI850ANR7GuDLOaX3k1TvMiZ1kWJtT/wCepUmH I4iI6sk7EClq6rl0gPZuOYcLuZE/NnIKTAQPZHk2sqYkBUegqVaa7kabv+dM09eMUShJKraa/aR0 ziGOg7iWpviW1Ck216mcUnVG5kc706lhtklLlKllrYKMOKlLcyEbxnKZcCcH08mYhfafvvVbKsoX VFFbiGmi/6Q79xeRTfSaUOqIdUML08F1obiV45ihUQoYz6Wg+HOHeMQcKagwWGqhFpiWWwoFj4Em 1klHSdafTJaGqfNDpXGlXldN26anQhCN4WFyPEeekbmcyQdiPqobGMcaj+hkSTWeN4qdg+qLhhqd IIKmeqF8ZHpsU3mdrqmgSJmE3FWpBf+6Dj6ZoDxKo3Canlc2lbfhp1s3nHvKlbXYolXCoGD6qLIX jTXVkbOYp0yqpatBbSwEQmi4j1jxp+BxQitZjPYYoeYoHhrTrMM2rG+YHdLWWEClkNU6kAaGrdya HiomkN2adLPikMPIk6O5hdxpn4PJQlgSo7RIf5/nQKApp/QZe1AZjFJyqKmhn+i6hYmUonL2bZnX hqL6o3M4ZHLaUVK6bxAydwY7m7/qe5VoTHa5sNmJsU0mm1LkhBTqqxILrFgAVQv2NB36nFmTsR+L oQfKq+EwmEKqioR6hJV1sL1qWk9pr9l1r76loT47qi7hrzKrVQG7R2flgT9boUF7Omf/cmt0yrCw haQwh7Q5BKu2pYJXQEXu+LQZK1mCurGyZbVAq3RYm3KnyoUdKnRyQRonWC4kip9kG6/JVx5OW6iT dLZza5jweld8q1vEQbPYdUhTG7E2O0t9a42SCIpzyi3foX/MV7iW+gWyFmIn+5OWG7D8qhZ767JW abcLObVSsLebW1/FqkTQGHhL0mWQexByKAJQ46a2aDiwu7piGxMjIru/iUp+NLu067JBQA4xNrvu 1FO924NsELy7anqhCrrFO7ZdkAqmkUiuarmJCAWi2xOhsztUqbRsW7vGC7cXKq8h4KIrKqYC2rxx iyC6lynAVr50gr5qqWELoaRINbrO/xura9oW2oEXeWCuAQu/kWtpBLEWurabXQvAiUuqh9MIDNzA DvzAEBzBEjzBgYDAFnwHUpY9QiWsEOSQrLGk9UsHQLTB5oqQJLxsexmsa+OGteOTrBaIJsiDLbwr bVtA5GrDAQyjOswP7dBmqIBi8rduQVyNDBmbN6xFr5piSMytS8yxM+xrKTyrccvCOIwZ0SIgV5yl VTzFMBzDkJhpO/zELny/MiqiwHSZkHBDw4NiP3zGgAjCdwp2cFyuVrzGNHTC4hXFF7zHfNzHfvzH gBzIgjzIhFzIhjx+c7wYPnzIadPFKMfIlZp0t5EwQoxB9gvJFAtlbbJKrXFJCnnJmEBsWwlCPzvW yaasyaFsDZ3QZqC1kKycyqo8UxQIP578jrAsnbKsybR8yrx8ywMUm8azv6eBc6Dsy8Z8zMj8AQkA ADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0058.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Car-Hill (Performance)
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0058_background.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODdh2wFkAXcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAA2wFkAYEAAAAA AP//AAD///8C/5yPqcvtD6OctNqLs968+w+G4kiW5omm6sq27gvH8kzX9o3n+s73/g8MCofEovGI TCqXzKbzCY1Kp9Sq9YrNarfcrvcLDovH5LL5jE6r1+y2+w2Py+f0uv2Oz+v3/L7/DxgoOEhYaHiI mKi4yNjo+AgZKTlJWWl5iZmpucnZ6fkJGio6SlpqeoqaqrrK2ur6ChsrO0tba3uLm6u7y9vr+wsc LDxMXGx8jJysvMzc7PwMHS09TV1tfY2drb3N3e39DR4uPk5ebn6Onq6+zt7u/g4fLz9PX29/j5+v v8/f7/8PMKDAgQQLGjyIMCETAAAQMFQY7eGAhhIZSlwSACIgi/8HKkLJqBFRw48OQIbUM5KHyQsB Vq48iSeljpcVXtKEaUdmjpseeOJ0o3NLy59zgoIZyiajT6QGmDbwWcIpMKNeoC5QesCqVQYuTbb0 yvRr1rBkwbpEIBZtArNj0ZYdy7bp17gDbL6VSzdt3bt76S6iWmNrVL511/rlq9epYKIlANNYzPiQ YxmQsem0iDmz5s2cO2/+O8+z6NGkS5s+3Rk0roujI6B+DRv2RMy8JkeibSC27t2o39mGUVkH7+Gf Z9smDfH3i+AtiLfueFFTWMN5pUpWOYM3qOrcsyqwjkl5ldiNEJsHGxXCWeZ7xEd5Xqh7UxB65+Nt Czc//r7ezzb/ct+EZmqA9xRN1Z3gl2HqxQKgCuw5UNwUiZ23VoX3qSXXBvLxl+EMAnwIogAHhEji h6rt5EGEQ0yYIH2QEXghh3vxUGKNNt5Y44k4POgQbsB1teEP9RUW4w44HonjAiBa0mAQHLkwJI+P tUgjklYuWUp0E+X2npZtsOhdDleOKSKU+gXikUdPPEkCmEAkViWZN8ZwYCJsbmlck21i4CUHQ/ag mJQkyFniDIoROeMneqZ3QZ8W/BkYonBi2EKZAxCK5Y9u2UTpKIvuSYGPT9QnqAiYWrockBbW8ikN oiYB6Q1WXmoiCoFuuupUSjg6QawhBFoqCklG2ukyrWZ3rAlS/wVbApJTMhtMsvRNwCuxPlzJAlZF diPtB5BV+wC0Z+IwJnDFjtNtTxKAG666PThra53spMsCu091AOMLw67g1T70qmBvBC/mWym8ypIF 0L8pBBymWw6P6+GRKRAckMIaCMYwXlqRu6+ZClmMAcaLijtCx/xSXBDILLmmJ8kfzNmCryepTELA MkdMIpQu/0PzCPairG/O5kYGoREZcxyizjsT1DMFIgsB88RAE61A071CwO7SFeSI4NRUF33DVuB6 HYLQB3/tcxHVku0B1xOjHYLVH4ztQtJdw22C3Cnq/YDdtuJ9At/tPnD0oH6PwDbgEMqkpg1WrS3s 4d72qzgIav+OJHhJhAOWOAWSe1t53qxBV/gLvHYOweehB0jE6SWr7ifqq3OQeQalZ5Dp7FLUfhXW Vuc+ue418P7d5pq3XeuvwttAfK4LnB4s8MvvPoSXQ0WfPL6yTy9C89QaBVKp2XNPhffrqjx+BjeT D4P5iDJwewTps0+9tc+DPH9N9P/gPlR9aouB/PVKa/tLkZPA976toaqA5TvgVaQkQAY6wX2Le536 JOgk+1XtXxEsCQExODcNJiB+DOggCBciwh5xJYALPOF7+Me5C5iweC4UAgXhR5XEzfBcNezBDe8H gh02rIdA8J7/uiVEIhrtMRVEnhIbaKgm0tBzLXwi63zgqOD/CPGDVuRTUH64wQdS8WpdFA7mSAfG 3BwriVwsowUewjgmNoCEBkiiG20oRxxuwI53zGAU54hACeywjX3sQBqhs0JBVrF3hSwiFgMpv0VO sZFYfOQeJUnJJf4RiLjDZCbVlscRtoqPn+SBEfXIyNR50j6l9KEPcxhJCRCylbPBk2tMmUPBDJKW AKMItXB5SV7i4FVgoxMqWSjMG2Tml5Q5pgxXmcxKNpOTClKlwKKJAmLOEZjf2coMZylMODLTjGLs GzTByUtxroubz8Tm8LQJv02KspPuRELzeBI/UtbTVfJUYYWsYkJ0htOWt9yB9RKogA4KNJy+XKdB gwLAEnpy/6EDFc8hS6fPffZSjY0b2jzbqVFuqvOe1BxjSGWwzB5R5Enw/FsYi+eTgJ40B2k8KEJH tEqKppOgWHsoTBvwzZk2pqEFpWmDMipUA1p0mi81aVLxaMySxvKp7XsjUz8aJp4ElapxuxPhyGkh reaUq131ajGVCVFWJkCmZLWcWbdpVAsgta2NsqpH/alAuvIvqlh1ql4DV9e7qjGVN2HrX2n31nii lZFiHdxhaWdX5qVVrQhQ6GP3FqphNmmuly3oUgVbS7lCs7MaOKNDl9NUv5LWZ+6p3U1selNaHW+1 K7ghbAs7UdrWS7O5wu1sdbsw3uYVuN0jak9Rm1pFEre4PP89a7akas3llvWzHuurpJSESZ3W81UV YalwG+bbe0mXTx/FjW11QrnK5na8gSVdnjhqvoxxlr0qzSxyrWtO+u4tsc8DbWiVq19D8rdq9+1r etX72wBXkLrvHGJ4E6lg4zHYVSnxD4ITHOH+MoyC8h1thi1XYEROYL4KbimBq4vX6H4Ysqd9blMt jFPHrvh+G5Zsp2gSQe2GFEDuiw6MZYvhGY+wAj0WD4mF3GJ+4dcBRy7xgIccs+TmF8l2rXGUrXvg Oq6Xyg7pyDhRKpMfAxnCXK7vl2Pg43PluMyGbPCIPcxm8l5ZxACOc2NQ/N+p2rlmeE5zsSy7ZzmD eYhqBXT/oNuL5gpP0tCHJjKy3tzo0mKuoXqjCcOazObuarpn1rk0nCMtypXiqXkXMdCFxQtq+yaa soXecqqT7AKJ3GQljH71cREEyDrbus24zrWed81iZUFoMZh+NamhYpJaAxuuaEa2lmW8bOfG2jEg UXa0+9trZwL109fGtrB9zWRud/vE3+aKbYpt63vGcMyoHre3l5PFZy/I3YpdNSMNS29yg0rb2M23 7bIN3bWK299cKvd3wIXuXfMuAK7bKsG7DBxqs7vdD2+uu6TYO4dX3OJ+YpmKNw7rjns83CB3NLw/ XvLBvpfjtpK4xq/d3VHDm9q7dDdH2GRi/fmuJDUv+b+e/0bylJ9Z5DvfttBVzS+ED3zjgmP4ZHru cyhVa4tHTxfKpu5hHY+XXkCX6NGRrqHzTfnrRSV6SbBO9qGH7JpYX4zWt/5cXpnI7V8XnKPmTkah 8+3uS9f7xY2X0L77HV9nN6v00l72ixU+KMCje91b7vVJBpnpgaNK/hxf9cr3G/GlJTy/Ew7yZK0k MzgWPOdZ7jQ0AqD0k5f84OemaIG33nmvR2yfYEdoijPdtLbXSaFOb7uRnApTwEesloc/q+LHzezM J/Pjd3RNxKuMOVt5e4BpZv1xi173yu9887uftu+vDNpkv+HGgN80zDs/7VbbXvellX3Kex78d56/ 4tcPDf/n6F80/8EzD1mdf75jXFAGRJPGU45xSOMRYrkHgAHIGV62UhGYJyxFGyN1RhF4Gf3nf+gy geYFRx/oSx9YcBZYgSAYHQkIRdqDD/RCR/WmCN0Sf8WAghCnI3+Xd7Q3DTPoZYzwc7L0fzFICg1h I+6FgHmjgdPig7EFhEFIK1wTgvDVgXnGazU4Zxy4OZR2c6ImhcH2gsLWNF+YTcEVhmP4brC3fDq4 JggChmRYf4DFhm24Qfsnh8uEhq0zaR2FOJMRJSbnhkb4hnz2h90jhn5AgTLXh87HNmvoh4cIh40Y fozIB1mYEnNIiZVoiZeIiZmoiZt4ifSHDC3oa6BIYaKhQ4pb6FakKIrMBoip2AasWF4qQzMGuIir GIuA5YpLZoamiCYgA4ILcYs0SIuwKIxPxoVpQ4yeiIwVJyBgl4zBqGnz5IF1+AXSiA4GGIIneIcX qAkiyD3WuCVUoYVcQo0pGI7k440SKGLRSAnYuFPN6I7vCI/xKI/zSI/1aI/3iI/5qI/7yI/96I// CJABKZADSZAFaZAHiZAJqZALyZCdUAAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0059.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Multi-Valued Maze
= <= /td>
Mul= tiple starting points
Mul= tiple rewards (negative/positive)
50 x 50 states<= /font>
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0060.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Performance (multi-valued maze)
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0060_background.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODdh2wFkAXcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAA2wFkAYIAAAAA AP8AgED/AAD///8BAgMBAgMBAgMD/0i63P4wykmrvTjrzbv/YCiOZGmeaKqubOu+cCzPdG3feK7v fO//wKBwSCwaj8ikcslsOp/QqHRKrVqv2Kx2y+16v+CweEwum8/otHrNbrvf8Lh8Tq/b7/i8fs/v +/+AgYKDhIWGh4iJiouMjY6PkJGSk5SVlpeYmZqbnJ2en6ChoqOkpaanqKmqq6ytrq+wsbKztLW2 t7i5uru8vb6/wMHCw8TFxsfIycrLzM3Oz9DR0tPU1dbX2Nna29zd3t/g4eLj5OXm5+jp6uvs7e7v 8PHy8/T19vf4+fr7/P3+/wADChQCYOC5ggbLIUw4biHDcA4ffososRvFitsuYsymcf/jtY4enwEo OJIAyJDOSKoceRLlMYQlW3ILwCAATQg2bxKgqTPnAp8KgAbtafNn0aFEkxpVinSp06ZQD8lMdXSn 0KtVsTK1SrRmA6FWn3IVq5Vs1rNby0ZVOzaqoal92MoFKldD1bZ4w65FSzYe3Cx896bNObjwV7NM 2Qa9oNPj3yV1KVz9mtXllsdEwG7QTNkyGcw3FHO4K+1uYLxzG78dwjlEz0qKT6dGTHux11SgVbQm QVrLbhaiJaiGGLp3id9QTA/XLTaKAAELnkufTr269evVEeX+sNxEdyuqv9sefRbE8wfSFZxXj729 +/fw48sHAH01DOPeAeOv0Lh6Tun//8kn4IAEFngde/UpwNKCAAzA0gAQRijhhBRKuJ0cFwpng3g7 GOghdQEAaNOHJK5nXXT1pRcBgwswuGCFMMYo44w0RigVbz8ZxUWJPBqog4tANljjkEQWaaONoGR4 mG0cioAcBj3CF0WQVFbpopFYTjiMkjkOEeWJwJlg5ZhkjmlkQ02I92EQ4pXpppsbDLkOly00Ztx7 TQAQwJtwsjAjPXSuwBMDeCbBJ5A4EMmGYTe6lh+hhRrBJxByFjGbYLXpNQmXTdqFYHZHlIkEjEIQ 1txmdjKKWnmExOQob2AS0adkO1RYQ2qy5ZrWkp8sRBIPsRJkpQedphAjCmURpqoS/1gVEtOCNoDK w6yvulBUjJeuulVNxRaXaXCNBOrBc92+ICp53+qark623lecquV2wum4AsSLQpV2yaYDhdZ2dZxy p/qipL2fEjxClfDuRymS1vKnrlu8IjOwBtTVAKS/NBiMAYR1moopO+JWMJ3FLGqKg8YWMNyBUsp2 mU/IEowsg00lA4GyBFpGoBXGOz0EMwQquuwdzS6WISG4FtzsFwhNrnebk6QVjYVcHI/wmkQZNp3g 0yuvWNIVTy6gMtMoZT2B07wtV/MRSjswNncu/Qyp1RNACxlOCj/wtmcfyM2eAwTvZzcTbYtdNXd5 Y60C2uNNkLhJLKV5WgZ78w2C2f9Ab924hhREfrcJlSf9uGNwo6e5mF8zi8LhlpeAuQOMS9Zt6plV FjEHoTveet0oxM55BZ5PcXPuOhfOkNyMtx28zaNvrgHru7t+Quw3L79D2KtDHz0Jr8/N9fde+71Z DsRv70H3KPJnAe0nG/+B9uqb77VJxGZO6/riy+7+7RfAH7/8DIDJr0zguw1YDzy4G8D4AAiBZx2Q A8njwAOHFrAUEMx/DLwcAbfGMwlM8FGjUmAGT3Ch5aDtJrNjHwiVgMER9q1+DSjgBVT4Lye08Hcu fEAJYXc6DNAQMM8T4QJzqEMSyBB4P8PezCgnRE8RsYFG1NzofgjEjTXxiQeLIv//Ghiy/ZHvilgM wQ69J7T5Wc2LiQJjGDXYNTL6MH9RuKHO1gjFNqYvA1RkDBqZY8XR0LGIIzgiF0uXhHjJ8QF7DEmg TtikPFaxf2r8YwcWyUEOOXINh5Tk+uz4Nx2ZkZNayCT4NKkgUAqyAZfE4RZE6TxSltKPbuzcdhLZ sJRFUpUudNWKAtnDQZKhXKz8oy7raB4pehCObOqjK+v2QfqFgJGfxAAtgxDMNa7NAbN0I35SCbgx sHKacXPmEAlwSgWBBpw/+KYwlydAcTqxk7I0Q7fUuU53BhByzaRAOSEnT1suE3/2ROUzjenLMczz lv+8Zx65FEEdYgadPqjmOpuZ/5v+EBSbD/UmQreY0BbBEp7MzABEe0DPjgLyneTsJUbLcFDGuHKA uywmQH8JSZOGFCQMrWRBdVeFltrUgwG9p0w9icqpjHRhP33hOKG50vsZbaNlTKpHl6rSACITCpk8 6kCuKVCqxpOl/kyaVE9Kr68aNKxjLeJJcjpKR2rVXRWQ6EthStYOOK03bk2DXF8a1Kmi9K7DyeNb kfq/tAr1nafkphmyalKZnDOWDgVrXKHaSk0Orq4bSOxVf9CpvZKSq4e1q06LihMudJaySb0sNofa M9VQcbARRW1qgefVoTS1m3pFLWwtM0xpQpa0cmBsY4m5EsVaNKQ0pcAhd5sQ1f++UkmCVCwCJ+DZ ZTqWtVG7CHPhijPZStWRGWJqi7SLSe/+FCZARSwHbztKMiz3u68kpm9BGtngUna7zV3rEI8o3elS 976G5V1ZoxmG0wZYjB8ALHDb69/uHpiNoiWqOXGrhupal66rVa+E+clRjYr1uxjuqkjvuFM3CDe1 FN1vVfsrvMkW9rwpjvAxk+vgB9M2phAc7Xg73FPlmtewOB2w89iH3339GMT6daKCrUrhM5zYxpiN H383mwRRFtkgoPWrS+m74AL/98NpdW5oKbZiKl/Px1C+aXpzvGEid6FJFu5obzPMZtvumMdgiHNC VRLZjMLzLiq88g2sHODLttP/zyklcHnBfOCLOJCbOpEhi6MAZ902OsZ1rq+Xv5xmCGe2PoAm7xYq 3WkBr/nTdp5wk917ZCB3TsMmC3RuGV1qEW850Zpuw3sfLGYtg7mAk34CqV881nbiGNW59sKwa33S IMvYtRERNLLQzOzwmRrZ0MZzF/RcakSvJ9sM/sKTqz3m/wHbzEUYN7nje+tzg2HZTi10r9lN5hL7 xsXrBqSz69wYWU8N3/ku6kJl3LM7RzUM3K71YyKdoH5H28mylTbWQlxunuKavWggdJrnbGtz9zLY qqM2seU98Ex3+d4iD7gBn31yidMt5Sof7we38+2WPzXm+qYIxy1ecyaHe5UR/++2mCn+O3dXOOid HnpxZcJweze4xvGG8s59DaWPozsGPsU5k/dd74Ib3OstBrjWPVryroPdzawe+8rne1efH3zbSO/2 jbds9Ixbmtnz7uscL37yHovd4mrX+9ms/kiYB37q9L7Atx3+80LWNPBbP3XVU412MSQ8zJj2OOUf bvm4KzzzTvUdyKmg8XVzPfQ9HH2pHk/rpINe8am/ephYD3k6R1mfoPZ65dOkzNpbNckZqPlN/C3u Vitc8Inn3OK/DnYvlBTv4GV77O1+63yfnqeil70FmThf3yPffhgXdhC9v/bgg1onuw9578kfT9As X9Vvx8LlrU/00uK6KukHev/1If/oltwk+ze3f773F/83feIXJ97lcrmEbH2nfuvHfj6EaoMCf81n KQkEgeU3eRPIYZVlBfOndQT4NxOYf1dQTQq4gHR3Oqq3BBJ1giMUggBIOBc4Yhj4fTyUbGzTAR84 dpjxfhxYgTbjATsIgl2Hfw7hgiMwhET4Ye9HglVggjWYfPYXOytYZT+GhCgoZSqofaDEfTQYhT24 hY6ng8YXhTYYQ+dHgUB4ZmRohudThMPHeSVYhmYIFwXYWj+4hnHUaliYhffDOFWYbnzohlKISPQV iLPXhuMEhmanhsmkiIQ4SfU2KE4YGu9Dh3V4hnNDiXIoBeUTiYOUZV1CLov/UYksuFd9uIB2eIhc GHXj14WMKFZU2Iq45IWwWIOPMYtBKISYCIqFaD8oRHw21Iu+qIkiGBTC2ARKiIFGdYw7kYynSEi+ 2Ix8l4c6MDx0mIpYRI2ASIuGmEDZWIxUN3gNaIngKI5ZVHXBSBHaCAHLmImwV3B7wo7KSIzo+Iuj CB3ruGpYB4m3CIpMx4o8oDFx1o7WpIXlOAMoY2EGeZDxyHxM8In3iEca6IgNKZETGYHxGIf8OATv OI2oV3DQmIivSDYZOY68gjaI+I//RYwNKUw8pzkr+VFWZI8vSUdw0Y0RaY8nuWM5qY/MN1IGg5E9 GYE/qXudyEdBxJNF+VzO/3Vc6JeU0zaDTtKUFSc0KumNDyiNVgkSxwWR1HR5N2lZD+mI6USUVpmB gxeVHflyNdkMhxKXcoko9iEy9WKWiWQvH6kKWvlqdblqpGiNesiVaOWWrVAQBzJhEbGYtveDcNGX VNARquE0Mwk6TDmYp4CYiWlsvxI5JWM3z8JP0feXt0OZ4qNEenOZY+kGLVE0nhmaX8NOopl3xngH RymYtXKZtAAztMMlkDkFORmHw4EyzeOOaCkMv3lsgzAVgTkSw8mStjR/qwkHrYlhjNlxHQF8rbJ3 6hGVzzkDx9l6sLB0wcOZ9POangOa5tR/+OQQyTkl3HlxlUmGuomZrECe6f/JIug5m+uJUexpPe8J BZLpPfOJgPWpC7zpnvL1ex0nCDJhmjiylAeKoJLSKGs5YeGBKky0l7dQneilnNxjoTHjnW1ZAuH5 hR1KAKRSSujlQO3JZ+noLPbXnRhaop2xoRMaf7JQEMeCTyy6EudpbCIQoE+gEQWYIKNXLD0qKLwg NytJpE5gpCmFpI8xFTSyoGKiAvmzpVrapUUApXkSAVmpAQWxHH8yd/fipSnApWuqpXP5pnLZKkS3 HnaTJTUiQYaJkkO6Amyapmrqp5xQEnY6qDmjVCbZmNLzpySkqFnKqJjQf3AaqZI6qZRaqZZ6qZia qZq6qfmUloDipEKqBPOzWaCxCaokFKqJuqgF+gekiqp7Cqik2qa4iWCLOqt9Y6qr6getCqOGYquS 6KedKkE/46KAijq56qnImqzKGkaiKHmZKUmdWZ69KUDEOgrHulXnma0+Cn/QUn+acK1Ylq0D5Gjc WpuM8KHgiq2d6UzSKnNAGqjuCaaKs6z0Wq/2eq/4mq/6uq/82q/++q8AG7ACO7AEW7AGe7AIm7AK u7AM27AO+7AQG7ESO7EUW7EWe7FvkAAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/v3_slide0039.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="windows-1252" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
= Future Research
= <= /td> <= /td>
Is = it a hack?
Lea= rning what works in a problem
For= getting options
Lar= ge state space
   – Approximate clustering
   – Local clustering
   – POMDPs
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0002.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
D= ynamic Abstraction in Reinforcement
Learning via Clustering

Shie Mannor (MIT       McGill)
 Joint work with:  Ishai Menache (Technion)
       Amit Hoze (Technion)
  Uri Klein (Technion)=
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0002_image007.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhLAAMAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAMAAwAm AAYAgAAAAAAAAAIdBBKmy+3XEHq0SiWz3rzn44UiB07VyXxYgraGaRQAOx== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0051.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Hierarchical RL
Divide a problem to multiple subtask= s
–
Learn each sub-task separately <= /div>
Learn when to start a sub-task <= /div>
How can we learn sub-tasks automatically?
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0052.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
RL with Options
An option is:
1. Starting condition
2. Sub policy
3. Termination condition
Q-learning with options
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0052_image008.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC6VWS0wTQRj+ZzoFdltoLS0itLA0BiUKERMvGmIvJB4Ag3BRDmZNGtzEAilNoB6a 3kx8oh40e/TghUQPevGV4MWDxAfxoAdvHgzGGI0HddXWmdkHO82SbGWSzf7f/N98880/M5t9++KZ DryV05kgSCxaKyIWBJqDABiSPEtDkLAdYYSsCOFqtcqjAdRu9YUcXhhnguV0jEa9DRHogCojQ5Ti FRo9oU8mDfCX0DEWJwyjauH0ZHEuC6AgNvdvHKx84B6W2KSYYIhOFHOnZs9AKcMc7JR/hLFcStBH eR0uAewDKgj9lN1E3/vlm2iU+91d9KO3vku+6K2H+PurS/cGYRJ9RZ8+f27m09RFju7zxrr8XvLr Nx6qy+8vv36HknX5vezX73hPXX4Nn37J01Bdfq/49EtCEbffcgU2dIHpTmq57Lwyll1Qjs3m1Bnw 51cn96w6SBZeiDKGQpvNGG5mjL2KbOGVxqDFUOidMvuAyHYfb3Z/BsR+SxOtd5ua1qzoRHuARjl1 0WHgBK+jauM7cYYV1R5RjjGfebWQtRm9MZMRsvDLMMuMZNX8jDYzbbPGJD6zM2owyPBBB69hM2/W Wan+X51r988gfbzOg1nJwj3djFGgaibjS+B4ijM0ycLjnSLDIN8TboaBDnTZDGsMzLXZPR7+vc71 H3/nxCAPFSbRr9n4WhfDezTb/aekmC8n3HkD3U6Z2OP8et2Lq/7qqpOJHe55dPKgTcSJVhEvR0Ss S27fOrofEvLoAj8fQ1ptPdGW7p1BDvH6FZx69SVErMZEfK5FxG/CAkZHU25cxjleF9XJL8ki/32T iFuJiE9iE9fuF9rSPdDJKveZ39ivuIi/bRPxK/71mXXwYb7ueQff4usad/arqLjXraNHSYGPujpF /jtJ0EdHmkT+coPInwqI/FUk8gfA5HvUzev+Vfx+p6c6mO60s+5UXMDobMLEm92v0UlleLGQV8G8 X9f97tfd7UxiRJOtvyiZ/2nxzyGYc0WhkaPH/N8MYZyeKM4XsjmGZOiB8yaZts8fY9BSkUq1awzw 7D/0x/OiMAoAAA== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0052_image009.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh4QFKAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAUAAwDX AUMAgAAAAAAAAAL/hI+Zoe3/GJy0Gmmzw7r7D4biSJYmxJ3qsqZr475TLB90jef6d+967+OdgDUi zvhCBpfMkrI5hJqeEerU6sSSAlqp92sD77jiEHnU3UrT6LKZDTZyGeT6xQZv5r/y+fmyxxOI4PcH YKiHVTiYgehmoYQhUUd3ePd4WLjAmDjD8ee4FIr2KcnZ2DWKmaP1JHk5makKZVh7KnoT+hqnohvG CzK7mqTYsRtreRsc8zksBENDKSaMukGnDBnt6RxEreBN+ItcaTkdPYndzXyeCU2u4Vdhu2zd/m1n PCevjQIUj6cPkCx7AQUaLAhoFiV0ATUB5HePEDpZ+qSpOlMQocN///8k1rO36R2fhZsi1nOEkQoo f6UEVQGJx2A1mR+/bXtJ0BRFig0ZJkwZc2fIayCBeoRJ82HSLRx7yjR6qeRHlBPj+JR6FBpSgvmy 1pS4CGFQrVSjZvOqFcbNtAJXJp33tCXMUYjczpU7Nly5U3AP6jz4VWrZciPRIr1Y1fA+rIFdTnUB jvFWyb9wunMI8C1eu34VT05IM2yKi0uzeKVail3NweYo182V2HOV0SyZ4tSl8vVa0WI/+5Z2ePPf znkVA+/bWFwRvHeDJ4/LtrBh3Wxxu6Jdm9TqdddDrmXRz7Lg2M2hj4Vq8ih6YaSVIbdLPT1j61Yp O6+u+qzs4j/Dyv/PSld3+5VGWHJEpYdeeUUx11F0C4aWnYMtMJhYLZMF6FlkMsSHFm6n8RceiP85 Nl8zBH7lDV3ftWUhfLQBOFxO+p0Xm0b2mbUhheYRxlqHJvqGwhVgSWgLZDTqtVSCEJ0YImjgPVjZ VkpuEMGKKzXlJIecwSVaYClVxVBvn6XzHpgYuZaaWjsWaJMI1HH05FAYJmPjkEPF6eMyDaonFI69 7Zlfm5f5p1RmhAoVT6I9fYlZoYZ+CVagUebI6E+gLeLoQ4jd0aCRk3og1qIvUrniPZ5SIKaGqKLK 0qmerMNNL2rSMmusoC7Wz2iPtoPMmCTiiSSbteI4A6lTVPmpoMn/IqusrW4aS2uwzjbCLLHNFngM UeS4Co81od44qUoSvvHtgLp2O+K0oHIbLZDqmlpXukj6dwwsfVqLa5yYXivtrcM+qykzQQLM77v5 NMoHvgYru2/B1v44DqctRAHtM8DqUa3F/S7Mccde5FJrvXuNrIas4860LMYnz6iwxy6/zEpjLWFq UR7pwIxzzjrvzHPPPv8M9MJiBk100UYfzU2vSC/NdNNOH5Hy01JPTXXV9lqNddZaI33g1l5/DbbL wIVNdtlmn4122mqvzXbbbr8Nd9xyz0133XbfjXfeeu/Nd99+/w144IIPTnjhhh+OeOKK/31zy4s/ TnjjUUMeTMmU/xct+caXo2va5kGr2rnn/h4r+s+ZX1x6rrxuK+Lk06Qu56i0nM53trySXLEztLtd 5se76y1yLE3BOe3vafviesywQ6LcyHK8a7zZc4Ley5JruEFmG1cL33L08Hgf9vSQ4fN9ita7WdHz u6XvxA/5ykiztj1oxD72m296n6qcPccIebnfORB3MU95ufKf8qiXo8vh701daR2BBmEH9SHLJyIx g/uKhTsfRPARCAQebIpDPQ6dDIIaUh/CtAc1eTCJGOCT1914gxys9GgfstsOb7Tzqi6pqWvk0mFm fugknjSsTbTr4BFaqLU5JamE2CmV48pnDBVeo3+qA2IAHcaCIvcacTmUU+KFBOTABw7BCn1YIerO SDpWza5yKHRh4TwEoLxEaC78GyMan8ijc61McxRb1Rr5KKw3iGdxRdrhcMY3Hj6CLHR4NKEeoyTB IgyQCfGSFgQz9rghlmSKpmhipOwErj0UEoMDu8sjI7khFf4xiNiqGe7Y8xzFlbGUd1zWFtlouX/9 SF1IZJMeJ7LBQNpSdIucnMAmWUtJ5nJW9DJYL/eyr+BxSlRnwiLiEIbI/0UilgREHy0hCb1eckua yVAlIK8ZKRX9j3PJTOUy+bVLWz1znM3jnjD9tLwo7lGN7UzgD041tLJVMibxW10V84nLCyI0bQUA ADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0053.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Good Options Lead To:
states with unusual reward
states that were frequently observed near high reinforcement
bottlenecks in the state space
(very) different regions of the = state space
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0025.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Balanced exploration
Meaningful options
Options will go a long way
But: don’t know how the state space looks
Getting to different regions of the space
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0015.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Method
Perform learning as usual, record transitions.
After enough = time passed – construct transition graph
Perform state space clustering = ;
Options learn to move between clus= ters
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0018.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
When enough is enough?
=
Tradeoff:
Learn too little and cluster= s will not be meaningful
Wait too long and options wi= ll not be useful
Bootstrapping
Heuristic rule – acti= vate clustering when only a few recent changes
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0021.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
How to Cluster
Solve the optimization problem:
(E is the set of links between clusters)
g measures the quality of the two clusters
Clustering problem is hard for almost any
g – bottom-up heuristic
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0021_image010.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC51UPWgUQRR+M3Ob5HaX84hJjFHP8QfRcEYJ2ogEfziwSASTS6fiRhc9vL0Leytm DRw2NmJhLRbbqAiCgqCNRarDRrCzsDkVFcTCxkLUrPNmZ5f1TOR0YNj3zXzzvW/ezM6rF61bIFtQ aGuQxehEiWDAnlIAChvlrCZ6lsYRJURFhIZhKKMxMqzGjIRn0rYWFPpFtKNnDYxAiGTIC7wkojui nykAXBNUQ3FMmLK8C2V/3haOCOb+Tr/+eCc93MSkNEMhP+M7c/UqNA+JgfujOjWp3hwUnb80mwB7 ISPGdwt2n/iO67fp9gFcutPvRs/fpt9dWY/I75eU7uYCSuxSuleXoUN3qsxLC55rQeSTreYz0iVK d5Ed3oQSVT/G7XVp/DDzyEjjFnvOEE+u6qNjf/e6298im92AEq+V7tJP+Ps5ZLrb3zf6yUSJ93/4 xVHULVccu8GP25f5dN2xatCt3/0jqHBR1SWgx9YjLiX4zRDio0ldG4O/82sD6fmAHlmL+HxS92Jf mt9ivTTNb7ExiPAK5/AP++qsV0AfDKNEMfF1aijChvL1JIczZ6uXGp7tNmK3czlk8WpW4cBkIrJc O2FMynPgboxPGhGOVxzIyhW1c8mKD7ICvKYrJ9M9+J9zbLYpxya0gwTXz1uuV/Eq9ZraxYT2VlaH 8xjvk9jicfVuyFtcTPBjEmFD5TotlRxroeJUrthRjXn4fzXuvDt7MkxmH7diPEoirKuXSZevl0wD Ue489Er0TL53hNKtM744AAeRDlvgekQW7fPHfsgtZ5ud58vk7C+BLskGhAUAAA== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0021_image011.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhoQFRAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAMACACd AUMAgAAAAAAAAAL/hI+py+0Po5y02ouz3iDwD4ahJ5bmiabqKpLsC1duTNf2jZdzzrN7DwwKhzqi sXVMKpfKH/PpcEKn1CpHaq1is9yut/PlbsPksnFsHqLT7DYssHbn4PK6/UW/n+P6vv/B94cnSFgY EWiIgpjYBef4CAlpl8eIQ1lZ53i4SHWJSeN5Egox+kmhCcg5VWqqwjqy+NpKKguW6SRpe/VooNoR WcsTPNso6/sUuoO6oTzcMHOMF028tFx4OTqdzZwwveJMnWUtiM2gbc5tRUmHap3Ly/t7wJ7n7gI8 D544nhaPsP4s3YJo91aRiNfu4D2F8/IprAexF0CJ+XqFU8CPDLSN/xYVpYBmsCLFkesW6hopD2XJ gQxRXpToTYiybgc/fsyFzKTKH/Yw6kwJNNlCnC9dauxGs6Mom0pzigw6FOivZk9XGsUWk1wbkElP FlnaMOROkVYtUh2L9mpWP2tlZrTV9p9HMf9+RlTpsqxejnULFvXaD1BYjH1loPN5iHDhGqX4QiVJ r25Vu5TlTtV3LRMttQOf8TyL1PHg0ZZjeMA19LQmhPRYt1b9+rJsYFH/AjbjD7FDVlgyEg29GGak w6ZvY7ZdTQ5WloYFX6iJ9Hnz6ImNIw9MSlxnxRPG+IJO+pR44s5xsWxn4d11sNnV6aYuAXY5DOzg p78shTa+KKfJN/+dOh1M6321mTi/2RdFeNys5Zd/CQaUX1cI8rfdgCNsEhcRfIB3IYP/TfhgZxGW BqKDH1qYgX60sRXfcQXCsgUn+Y2o4Ikm2oiiDCrup8cab12hQy2I9OafNxnmuNuOud2BRn8muAhh dVK+Z5+RSF7pw20LFhGHkj862Z5AKYJmFJYvFdQWhyk25GNA7Xn3gTZkSmXmmU+1ENNPYZKYyp5r 1glooHNx96OWDwmKaKKeSdjgcpLpoqeiktZ54GDjVAoplKBousmLjB05qRYlulljEqACmOCpnqka KjJ8xvcqE6oao6UP37UaBlcBsorENzHi6OufuNIlJrGu/FprsM//8TrsV7cmKyt+5610aWTmgEnT Mq05tCpfCUUlX20UfQtts3gWyh+6TZRKbUJk8WaoSfC0hFpaVunk6EOqZWpuGbwx+yaFkL37qIjX ElywJ/M9RpJle10Va7/q9AZwKhvelZKjzNHSUpIKzziZgAkPR6e6Eq9LqobBrNYxxnRGzK3LLxfM KGUPUyncydqlrEa59+IrGs3vyfxKY5UtVpbQ3Oos1sZn+Aw0wmWCEWPUwTXT6NFcleSXwjAzjaHR qCZZ0YoZs5YtRyaPR/WiAGoLdGMnQgSbT77l9rO1cDmJ08elgt1nmWiPNS9UNj9Kd8dx3nkefrEJ WDTF4QoleV/ggdYE90apybU24EPXKzXRjmmcb6Sen37E3Uq7S3PSrrdcMeqyczwz6db6HXLuA089 e++bMp4X7JzpzvDuM/uOvK3AS4Wx6CM/H3rsySNPt1esn123ZMIr3rzp03+vyLZJVTu55jmjinfm 8oHPvjDStw//xPHPf+b79N+feuf4719FAQA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0027.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
How to Cluster (II)
By topology:
Use value information as well
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0027_image012.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC5WVz2sTURDHZ99Lt8021LRNf9rabVpFoZpSPahNQa09iKkHG/XirwhLDSZNmxbs +iNEQRTpQfDmyf/AUhUPFcnJmyIIKtpDoGg9eFCwKiqNO+/HshuqbgObnc935s28md3lvX729A6w n95dUMGP1pdLChr0FbX+oI15K6zLT6RFFEVYCimVSszapjQLrdqOC5CCqnfXWdYmdR20QgmDIWhx wbLmrKvYBZBVrTUiJgDDiamzcXPcABhXsPYvsvx7ke3hFhYlPgLBETN9JpOC3B5L2KtqcwGi5Rqs S38RyAH0gs/St1rRVda9T1ugOwku3Wx6yXejW9u9ej6F3T878m4PYIotJvJPwjvDn6yPMwiRMdhf i1Y96aXcQi1ZLzVuodYfkhq3UIs2Si3ayPd/73/zuL/6/onYvyL230MuBDHFRdNLXmsu/f+eC7Hz PmrEFJdN2cPBdtkDt1Ab75Aat1Dbp0uNW6gNhqU2GPbY/wOv/V9pW1P/Ua/9Pwk7+5+lp8QzzKrc Qi27XmrcCpE8hRqu5embLr6n/AqU7Wk4rg9NT2UTwHt96K3XmDrTiilSJmeNnuh089EGJ8/7jlS7 +WsF57/tq2xWA95mFVPvNmOKd6JOlOzoQv4uuI9y/iF4VuH8zY7f2IK8ZMdz/mjHc/4guEifs292 wO4r6ef1y/tCFfuKJ9PGpH7IOK8fzqQTY+B13vkmzHBO1JmouM2+iSHBEXJSRx6055/a4IyPkNF2 pz9CjoXc8UadO/50rdM/7ztQ5fQX6fFKZ/0iXfa5469Td/xV4vZXsYnEbX8MkEdXex/WMLfy9yFC Rti3o9t9ve1w8wzjHpsX27jfL/h9i3VsQToxbUhlVxMqqcyoIdfE2LPQM5Jf1rv5GuMemxeC3C8z ttawGskxQ06joLIVSckTPs6aONc0dvax4QCfWBAqGT1mp6VCSHjEnJwy0kgadMJNFozOT0t1ULPi z5W/aZSt/wMLSQQ0wgcAAA== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0027_image013.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh4wFbAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAQAAADb AVYAgAAAAAAAAAL/hI+py+0Po5y02othCLn7uH3iaITk+ZnoyrbuCydcTC9zTd/4ru/+DwxWekIW sSg6Ik/KpfMJTTWjGBX1Yr1Ktdyu95D9DqdiGbnsCKPX7Je63XiTznM6HGy/6/cPax6UNtIkt4WT 9dcXmJS2gcj3uOfHpJgS5yhBaNRzyTDFqeDJkQlJytcINhkn2DmK9Zlxirqqusg601qauxYL8ApK 6aHEm+oz7IvgOWujcqzr7MRMbKNsZiQkSf1buyzz7K0XPQfcIdysDYS9Pa1ejfz9zhae3c2Oh4Lb kl5Ju8/dDg+Qi7x67gj2Mmfvmg6EyQweBIUwoERxsuYV7HdOUEQL/5sodrJYMeHEkXWwXHTYC2RK kVWQdFS5EmPGkCRrwnL00kwsOg2RNbqlQdEwk9dmHtwJiJ9PpOOMbrQJNRG9k0eTfqQ6dB0tfFfR aS2h0+pXsEuljnsaNe3Uk2rs9Gwr9iPXsYa6wo279m7Tr2jV+o0ZUhgmSoKDCvXVF2/Mwoa7AqZK d2biv15C7PRzw7KcnIMGb1UKeuqhrEMHiTJh7GeJU6rXku2suCLsvZAnU9bS+vLqW7odA5XZ7djs hwmx5bbFm/fq4plBi0L5G9Ze27evMMuMXeTmgs2SCVd63efy8Y/Fk0edfXlPsN2FBpvuprpHcumJ 2w/PSlZ7Vd/9nf9fqNx5lmhHRGp9LFQeUVchNqBR88n3XiH31Yeff9GRE0h/EJln34Rkfbjhf3ic VlaD7EEHooIhuiYNhBHSR2CM5K14oXT8ATdjhRMeYiJ+Oj5nYkr7LfjiipDB5CJHwVDIJEv/NBeZ Y9owZtY/HhYXmozEVRigfytRWeU5YErJ4ZEoJTnEklpyWQ6UHXaYFSp3jRZkmyyhN2JrCYo4Hnok Orkbd5sMWlgY9c1YpmkFWoPmgxSoBilrkmrW1kBHvdEZaUsRAmSKI+4G4Gg/1ZjnpaZWNWpyBeoz apZf8rhppmZWxeKZjT7ahT6eabWokgjmlyY91DFSa5iBFZvIr8D/zorhrbYqxGyUr0n7GU1lKkYq DzntKqaDUinrpacSOgsjbuDGdiK1RFoL6IGiFQHrBLAdgROIoXhbLrn5UqFrUK2yC/CA21qJyb/i GjKwv3quB0KnByO6Z7P6FhlFv4RRi4jDD89lrLbI8opvxHKFbOgKw1J2sop/ZCtyxuUN92x80QqL cZq9kimyKxOPW/HMJdb8KL2uSuyVz58CLa/QoT3F8c4EQ2EpIwx/HGU5JsN7rtRD5+ytaSFXkbJf Tdfgpr+0PRwuwGiFzZe8EU9NrLRKM8F2WmOTjXZwZ3MtJ86J1f3uYDBT7c6Y4q69KlNLKs5RqnfH 8HhdVmmMtsuP/7ENOM4g36zulIQbPYZrcQZ7tI15u5Q56WZhuveyAbuQ+ueLzd25kae/nq9eYNd+ VmWxu+0o6B3DEDmSuG59vAZ9RX07bQzu8ju3OOIOfDHRM9R69XEvMzZqyA/P98VfFO90DtE/s127 7i4Pu/BB+Ol4+U+cnwvDl1iWD+wG50q//P7n7z71AJBR4UPd/w6IvgD2jYD3UOAP7ue4CEpwghSs oAUviMEManCDHOygBz8IwhBSsFTSYR8D0YCdCCJwhbtIAuXsxRXDtYJztEJY2ljoQJeUjwwS3BCd uraqpwnpWi/TXP7o1T9vJFF7+tJT4F6WwvRBBImjG2LpLrW/HP+ahHY4/JpAdoaZJxoOeZwgVRlV R71peXGFS5TerbhIvZVlr1ppNKLckgLHLtaRX2C0XRrzcBpefAJ/Wgwkpi5IrDzqsXl8BGMWjeaW PQIiImVb48b21MYEwiGTkAAkHhtIycqhcYGyY+MdOPkITyqPaRupYgGPpLtFWrKRi7ScJdZ2j1HY sliodEYvvye/Vr6yYKBEFxBlqaRNIlOYzCxmwyw4MmQmD1SP1GEtG5iuWQzphfg6A8ukWausLeGX 8XDii8gnRAXhQ5F+BCddEjZOd8pzntNkZFHoic98OkWSPNCnP/W5LU6S858ETaVOXAmvgipUmpzJ 1UIfqkdwoXKroBCtaGW02E+LajSYtQmLx/i50ZCqpZKY7Kc4RYpS+aiqmsN8z0lTClOxJZOJzbsQ RWOK0xzMFF08sVdOfypTm5kzTyMUWCmBilRd2BJuRqTcTZMK1enRkXdqK2JUrwqPjHFKRe16Kla/ SlOtgU9zYwSrWU+Jq566EWL2PKtbcZPWgkFzGmV9q10vuituIsubKfLqXe2qyjfBp2Gz/KthE6rM wyr2i4nVVwEAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0027_image014.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC5WVT2gTQRTG387O7jbbUGOaWm1rTdLgP7SUXjwEQagpWFMRGwTRw0YIVUza0gZ0 EWPBmycVPdST3jwX9eTBUwsSQWqroJeCYIUKValVtCTum5ksmZDKZmF232/em28+Zt+y71/PPQR2 LccsA3wY3b2hYKDOq84NulhWc4aPVCKiKCJSSLlcZlGvslPMNbt1fmIZy7GgE+3Vt0EHlLEYAg6/ dKJZZzyKAQQNZ42o8cNwOn8pZU9kAKYV3Psv+bn5iXm4g5sSSiAwYucujmehcMyZmImYC35iFtqc EX7jLwD0AXXmDzvVTc6z37xAhwgu3W970VuKmSfq6yns+a1KN+FHiQO2R59vt/LJdRVX1ww05HfI q1+toyG/i978rpLB3Q35PenN7yq5GeV+kf8Q3jl4VXxgj4VIH93owqiV/NB5hHO+aGWORyECtF/U AV2Mcq/TJRBelfpnsFTfK6k5g7wy24NeJ+1a3S3OIPn/MyBCt6RRpvvL5vycyjyoyryuc/4t+DTh vGFX9LKdyCuunsyDqszrOucvrh7nz4KL+pE9yB8Fv1KfSbxGkl3IxwVbdIL14FPBcS3eXl1v0WX2 TR1189d9PL/VuQ6nwolr+ck08Pf1ztv7Kuo9EZTIur4fSxzX9u2Sea5V5tFmmXWdc52+AvSZupzL TIVPZa6Gz4zn0mPg7f0X9fZuVLgi9vmuHWTfWsL1/UDKr5EnUj6ude6ozlu0t606b9H5IPKAW39u u1yfCsj5Dy3IKTffzXjUzZ9vktePGfJ+hibrzahy/X0i5zcB+aybTwLfr04/NHDOtf1g0XshlDjk 7jMQlHlBl/kW5WyKP5fJ/m7MBnBvATAYvWD/Q4WQ6Ig9lc/kkEyIwG1WjLevK0FoKfkKtT2gsvX/ ALsvbqWkBwAA ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0027_image015.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh2gFgAHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAMACwDT AVEAgAAAAAAAAAL/hI+py+0Po5y02ouz3rz7D4ZfQAbiiabqyrbuq5owTBrljOf6zvd+Jvs9ghRi TYhMKpfMGXHybNoQUcXRVpVqt9zuMgsBfxNiwNVc9qrX7DYnzYAnn+Dz2Y3P6/dWIDuaZcc3SFjY Jbdwx6VYdWNmCBkpyYNoVfnDaIl2Odnp+RnBSeXWSCYKipqaejrVVnqgqCo7S/vo55pIxVrL26sW KwG8FRsk7HuMjAu0u0Oc/Ay9J9fIrOMcjZ19GJxIV/JdHXKtTV6OBOeomwv79Wr+Do+DzuhuO1cf n69/Mg8IyM5k3L6BBDekkXEnU6t263KEK8jiIcSDWExZSqdEIA2I/3HOcexjoYyRkWQABsT3QuIz lRpYHuMkUh0sOiYZgpT3sWEPl75gOvA3s+QilC14IjNaJKfOUD811bS3cE5HnEpvNqsqFEpTizVp ynSh8WlRpD3JhjFby5hVU+C+mWyrsChUsVHBok17t4HagntHtax06R9ToEvlYl3r5HBWpn+1Ji0M kjDiiIoXU608d8gbx27jwG3blGhdw5gzbywNeDO30X7DeDZdx0levNZmz+rLOuRqMan7iM6dAjdf 218rCw+HDs3Wx8VbQ45BXJZwcdFXIXp4cG9vy3Sdj+12BLkjVp+r/zRPbPwb9RjKZ7ye0vTk54E0 0/Cmi5md6TrNn/+vzRUWu+ynn3eYwAeWfBeVl94pxAkyn2Pc6dYdJegZ8dxqE1aA4XsSrlCMggYC ZwJCHIaEERQhZshYhBr6NwR5KQazoov22egacDuZaFlexSA4olPMaahcETVuSGSQEsL43wUkGdkd eVPdIqKFXeEo3ZBVdqMjlSKA08QV7q0FE48gIAXmSX79lpaWZXbZHgpHekhhhXqZOYKcdu4YIJJt fihKePzoCWczgSrZpEtGzXlPnyMyCGmkkk5KKaQsgucgnnkOOtRMcHnGX1Bb1inOIR2ySUtgqoUF JZbLFXoZqXEW6WeMl+a4pzXOCfbhIif2OiutdQlDz4QJ4fNkSqj/6tXBjz3WgV+Q+y0VrV0N8Qpr rlJpGWODP0LoKasCYlqKN+amqW2ey4IayGfPzqjgjOiWNNK5nSFa6rrp7ovJr2c1l5lkLU6ZlL74 IgafwDQCXGO1A9+aC7bA3phuiUI+jOs23BI8mleOvgpxZK4evGFyAfvU45oXUxyyUB5vDHKumgqb McZaqIoxtrHthpGDATspILSRamawgTszyzO4J7782L0LMhhay9lO7QPOocx7JcfLYWdxs3ZOcyds 7WFdK5Bcj9oib0kGVTTa/Fbtr2pdgu22axLNTHKUYlPGMtJ3122f2hN3XavUgO8Ut0EyD6ZSqMwa g/LHVAeLNNGN//8Nuaw0vw04k4nX3BKs2bF0OUUTy3cX3eeRfqZ2G9ened83f+5H44fLXrvNGcIO orql+g661rkdCnOYtI9t+92OP07j0OBRAnyzy3c0jaTPb9rrmHAXj+KZ7FTTNfjf66423r17EKL4 5jNHeMwk2oIc7R32y/3xVE7PpdzHuYq/k07ffzmvVWRt80Cf/PJGmrWVZoHksNpbMmI/BkowGQ5c iOcKh8EJavA2nyMb4uq3wRCqooKM2t7pRIjCVSSuhB88YQpf2AkSzg6EMKwhIWSIwATqzoY8vGHc mNY/Ds1PgT0sojT85a4Mem2IOzSiE/8QQalJLGpvu+ATr+ilJpuODHWMYWLwsAjGYUTRZfDahKWo WMUwqtELFbzR0XCnHKZ9cY10FEIbt/LGL7aPalasox9z6LYS5dFm4nrNHw9pRxqOy32VK5kiEQnJ GDxyjyEb5NT6GMk6bsduV3PeRYTXyEyKMj5EvAkl51O+YblwlKz80irforqKTTGUrawloeYYkFLa cpcGxOUYfMnLYIaOg8IspitHiIECAAA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0022.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Experiments
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0022_image016.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC8y9f2yc13nnS84M9TpslAS72TrpNe6OA7ibbKuQ884PctINiki2kyayLdjZ20p5 A4aWaJmxRKoUZdlxUxS7RE/A7hZbBInhixSwsPtHcFuExYFU3AZB0LWRGxRWkXuRixgUDsogWAS9 RZ1t3Djzi9nc73kOfww573NepdhY3wQ0pXm+nPnMd97v85458/DVq3/9f/3vY/K/b4yfH/+I/Gnw m+NjbxkbK3/pf5TGkrEfJ/62Y/j65SNvHzs19tBjp/7t8QdOPVY98cjJRx69/4EHqx9EbRxf/2bn +/unp6fHbvt/d+GrX9r/e8U/2vjP8xH9E9ouHXy88bEJfC+N+3vzf3r/+Lvx33u8G6Wf/jQo/5Mv jldK42MfWbjw9MLq4tl53PC//crkvW8tTf7uv8BX9f9+6++OjU3v3Gdp5+ffIY84NvY1eczxUuk9 jz17eXXhov/b5Ngvj637+x37Kf53z+f/+djb/sdbfjf8fFnu5x9wP57oraUvTbwz8X/6hdJ3f3xu /Ij488tjJTiw/2fcj/z5vrF/HP/h+H0PPeZ9O7WyMH/x8QsL1U88tvzE6sJT1acb75/+5NHJqUuX 5i7PP71Qlf+cW3iiOnX5yeWrl+bPL1Sf+6y/AZrLcysLjy88tVxttarnFs+uim7vxscXzi8u+XtC Ze7x5WfmnqlVZ9vhvvZue7ZWTdPmzv3tK9Nqs5EeluLG2cYOy9n5Cwtzq08uzPnylZWF6hPzFy4v 7N7PpbmzFxYvHbjtxPHqc/i/1J6tLjxz9smdu8ffn9n7+84PP7l6SHH13P4Nnz06eXl1+dKlhXPV peXV3fv0j7e6ciV49dnqov9PuMPdB/bPA0936I79Lc+M3ILneVizf8tnqyOPPupFLsfPPTXjB1JT +jk+3m4f8Ed8ZfwXdgj+l8rvbD9zlz/+w+1H/e3j4YivlEqlnVfp6ORz1fCih0Ph4jLyulzdeZWn qysXFpf8DdPVnSNh54ajkzuSpYXzQzJ54ZcWkIzVJ8VvvD4jL4h/zL0X9+D3WnX+3LmDr/XB73uC o5PV3Rf7mRHN5SuPV4fjMxw6Xzu3+PSh+3h2BOTAfeDvw8k7fB9Se+qZ/ccf5tp53ItXLoSf23ty Tz27/2DDEDsPMvQDcHv4EfCDqyvzS5cvzK8u7D3W7l3Id3j82epz09Xw/5WFs6v+lfEvyIJvAgtL 545O+vOVPxq+9/bvvv2+e089duxD55YfXziWvn+6iqPzwWO196dHJ++77wR64uryygeqD33o4yc/ dPxXqx9/cqH6EF7fq8srT13+1epvLJ19v9d9fHH1wsIHqic+kF1+cnEhO7u89MTCysLS2YXL2eLZ ixemG9nZC1fOrS6sLC6dzy4vXrx0YaG28/3i/GcW3r9w6fLewy0uL92P5/aB6nRtqj49lU5PN6rV WvqBxuwH0hmvun/57JWLC0urDy8snFs49+Dy0urlD1T3TjbDilMry0C4fGL5wvIKNCeenV8C/XlU 5qunFy5cWL5aPX5h/uxT/mdO4XZoav7Px5evLJ0D6vHlZz5Qxf/Qq6vSnKvSjKu++3rdA0vnTixf 9I8EfPlB3+jxqBeWz+Ov4tRveqfknIAnNr/yLNrg/s211nQ4XeycIe7zL/ji0qLXVpcvLax493Hf U4/7I+e5xxeXzoXWt/snlC5I6cLy/GjpGSntN8zHd25G20QUQ0H+vFu6b+9Bq/OPP76y8PSivB6e 4CwOdUn4hXAnZ8/i78tLZ+dX92655BXLlxekAezceH6liq+F8Ii7t13GbXI23bnh4mp1aqf57Nyy hPvabSW7d38Rd39lBYfV6sX51ZXFZ3YLl1G4vHDoxhXcuHLwPlcu4KadXrX7o9Xn/Ll8t2vtOXT5 LAoLq2cvPvvUWX/w7BdW5LFWzj8ut+/dz3m5+fzK/LO7N12VW/zDXV08t/8sPr1386eXF5f2ntv8 pb3b/Z93iYGxn98dhJUnwo1PLF64sP+6yXLk8qpvCnhVVnEI+hP4+ctPV9/7vp2j4fGwZAm37i5m 9u52IVSluPN67dUev3QBJ9iDr9mC3PbcZTzUUwvV84d+BAfSyqLPmRzJl68urp590gMtr6B5La4+ tHxuAb1J2uGF+aVz6FuXFuTGWrhxZdk/Fbklre7c49nl5RWkUvqeLA2rF+cvXUJM/eF57tJiegmO roa79fon4MOQGb5RPLb4mYVdwdSDDz0GN/Zu3kkF4ndOfvITQz8wXd37Szjn+YVhFY//1IJo8dL5 73uG/cZjj5zEU1+qPbB0dtl3kurVJxd85C4t44X+TTzE8tXLhxRps1WdX/FHULiT0fvI/7mzy5dw 7ljG2Xzq/VgH4Wf3vv/2leXVhcsQXXh8/vLC1BM4YheXwq3nHg+3oQsuXrq8eHnq3Pz58wsr3kn5 A+pTZxdX0EefuLDwzBS6wuqTy1cu47Waeuzs/Mry0tT5K4sX5L4vLDyxOvXIA4cfXeMRufxpZfH8 k6t7OPu34y+h9PgV3LrqD7alc/OXn5xauCjfVhdxDE7hQDqHs8fKU/4sfgAo/PDyQg7R6XOLeB3w fBUza+lstZ5Wz+NUsgTAp+cvVOfDcQGHsdx9TnnxHluFM/Mr5/ZueFyyFU65PrQLC76y23TPXbk0 fET45v3Z8MfDd7R/H9Xq3o/WERMc1Bf2D1Z/64WFpfNoNMPnkypWxbUqRAvPVKce/I37cfRWn/Nk 8lj+Ke2uCj6LuKzMX7ggS5q9Z7G3+t5pvHLfC+H8glYlDy33snPcL15eWb6Ik+xze2BTJ56cX3ls 1Z/4L3tbq1MfQpdEG3lqafnq0t4PIouPVp/be167zP5p7d3nroc7Tg3R75mDu9m7y7M4LzyHdrLX GPyi7djhG8KKyYd6b1WFexM7D/Slhd8Oj3qsjSWVNKfPir1DjcpL9qtA28Xbb4q+v1dXn73kVxqX ly8snvvV6rnlVRzx+I4jeyH83f8RT+CxR/DqfeKT6DZoL/626t5zu19K728OPRm/Ymwc/Hv+j37I /2jrdpT3386DVFuH696Xwnu/D83z7MryZX/YiS2Xq/C7uvz4p3Fq8w39JNY04VwdTjF7WA+dGjpQ ds7wYcX+3M7Z/bM4gV3CSnLvRz6EH1GLp+6/cFWOC/nbb+KUILfsHPH7J3B5Rz90Rt99U3sKd76/ 7FlYHj4vT91/oHj4meCQb1SPpQeeij/04JcczrurFfnL9P7br+rOWeiAaO9h9kOF+8djHHrUU+HW A5gna4twaOcpo7Lg32vg5+DV/LknF3Bq9AHeidxFLBT2fnL1obDs2ll97TwxnBcf8b3zud013K7s 4ALOB+rQm5mjk/5sW8NXHWvjeayAdn90b4m3//yQgr2HOvwsh2sHnuspHNYjL24qzWD/Qf1PeJ+r u0cHwi2bDPJ3rO2GjpoQ9R2kD/2TnnO1Gp51c+9oyH3mI0/ydh7tLfuP5h9k5zCLPvD+ETuKcMDL Bz+09D/v+S7dzhP+GR/vNp/x0u0/5aex7Fi8cnl39frkwb9eWHl2b315YeWZ3ZuvXHh2/497t+JH 98QPPYpY7t35Mwfu+5m9u9790zM7f/L3+8zuRoS/Z7n5kEHD7/VzzPI/tvtIfn8F9+nv7umhm/b9 3BXuVsPpMzham52upjMhP/4eRo304Lv34Bu2x899HP/jsYfyD6P2B//Dhx/Jm/f00N8LntH+YTJd bSsPM2LcoccQC27LODyC9059MnmdXZr4bm+vDjX3nYI/ZvdvfujCihxfsWNo6ACSI9Nrfc3blQ7v eBXnz9+Rv4udw2nvzwdeX3+j/zrkxO4RlH/0yA/svJ74s7+fgy9n3r36u2z/bP4Gv4JtIx4PFw/5 vPr4P8HncCi9mT77w03Nzz/d6J/9OA5+BdtGfB4uHvTZr0equ2+Xxclzh26RfaH51UOiQ7f49Qtu qh1U+ZvSQz+4+ATedA3/3LwsOKfwNmX4Mwr8dehDDL+Uw5d/6lLwb3+kKD/l/+bvRG7ZX0+tLM6t Ls/tvBLP+Z9N4as8nF/1yh8au7c0d26RJTcebXr/McLDzfg1SjjLhbex+2+Pli9eurK6MCevKBbh F8ODiUVi5kGS4NPi0tMLK7uHiRwTXn45X55Ww/7gzk17j3zusmwbLQ6/s96rYYl1TqvB8DRsd4D+ wu55cQXvxsMf/bsP70JYrcoRN7SiDYvlw+tZOXM+fnH+UriLvTeWu29owYM3rkuXgifzss3qRbtb izvbiTtL8EvVs/Kwnk64ZqujPsv7Rv+QO2vp6vmVz+4+3NCS+eAj7CyEgi7n3fjOdvTu3vPeZvRj C6v+CNzfct7ZGfDrbLy59X+QDwTkp3fEu9vhH/CdYn9bG7fsCEL59nbJRx5ZtiF3cffvc+9xHpG3 gB/wbwVrXn1hedVXh3b4ajtbggd3E9PpRnWm3mhU8cbf/wCAWg1Q1OstD9SoV6uN+vRMFe99zuIF PTo52xzaK6neJ58SrM77T4AP7Nafncbb4PBZyuX9jeCzNdla8f8fvjWVWw9r63IPtUO3Nnbu9+A9 NHfu96C2tXO/B7UzO/c7fOvZ6aOTteqn/X8uSwagkP+02rV6tVnDsz716NHJVvWqPwQa/vVq1pst ITkmf51tt6uNVq2BNvPQqaOTp/B1bE8S/YEmfmDnSD062fCPcP8jRycfe2Tv4YRoT34RzThtNsNf Th6qyF+qdRxUqCCC/q/+b3jsGU/R2n8hsWio48WvNRuzIkmHMXaQ69PTucW0Ua9X0+l6s5rWZmuH q7WZdoqfSqfVH23MtBs5P9pqtGGPZ67VwHugtuOeL47+YGN2xr+k/qmkrXb9YDGtTe/ebb15GKie NvyPyhNtT9f0Yq3VivxoHS/AwWo4kGqtuger4/gZfUqhmP/QtZY/gTVqjaZaawBolGroB0d9CkWJ tVZstOFzztORYi1N9Z9Mm2mqFmsztekc+3cswIPmHknH/M6yvKwt/3y0ovqj4ZVrp81IdcTB3ao/ gHEoz0Qed+RYG75nJCj/nkNw2o1ItT47+nz98dTGSrHRrFdP3Q/vpn0z8bc0RzzfK81Mt7XSbH1G K7VbTa2EMNXVWpqqILV6QyWpNWZUlFprWmeZqesssy2dpX3wOBx+rz+sS2s1lStNGypXWp9RudLm tMqVtlLVo3SmqbPMzqos9emaylKvNVSWetpSWer1tspSb6Y6S6ups8zoB3G9rR/FjWn9MG7U9OO4 kerHcaOhH8eNZv5xXKth9dbIf35SU0IqNSWlUlNiKjUlp1LTghqKSlJDsR55ilpWQ1EJaygqaQ1F Ja6h2J7VgbSQhqKS0lBUYhqKSk5DUQlqKCpJDUUlqlLUshqKSlhDUUlrKCpxDUUlr6HYihzQWmJD sR05pLXMhmItclBrqQ3FRuSgVnPrF8Fabn1Ny62vabn1NS23vqblVhbkWm6lqOVWivXIU1RzK0Ut t1LUcitFLbdS1HLri6O5xQJopr1TTHPWe3vFkdwOF0dyO1xs5SzJ9opabqWo5dYX1dxKUcutFLXc SlHLrRS13EqxFTmg1dxKsR05pNXcSrEWOajV3EqxETmoR3OLRS1esUZrRo2trzVn8w8DXxuJ7VBt JLb7DzeS2v3SaGiHaiOZHarV9Wc3mtih2khgh2ojeR2qjcR1qKal1RdH03qsNl3bq47E9UB1JK8H qiOBPVDVzrRS1BIrRS2xvqgmVopaYqWoJVaKWmKlqCVWiq3IsawmVopaYiUhWmKlqCVWilpipagl NqRSOQ35l1KLrK9pZ1pf0860vqadaX1NO9P6mnqmlaJ2ppViPfIU1TOtFLUzrRS1M60UtTOtFLXs SnxyzrQ+9FLLOdHu1XLOs3u1vNTuP6KWWrldS60UtdT6oppaKWqpleJeaoff6+cJtQRLUUuwFFuR Q1tNsBTbkYNbTbAUa5HDW02wFBuRw1tN8Ey7pibY17QE+5qWYF/TEuxrWoJ9TU2wFLUES7EeeYpq gqWoJViKWoKlqCVYilqCfTH37LtfzTv77lfzzr77VTXHUtVyLEUtx1LUcuyLao6lqOVYitrZV4pa dqXYTPPfUUhxJLvDxRllKSnFtrKWlKhM12f0Yq2lA41md7jYSHWgvOzKaqqNp5iT3b1aTnb3ajnZ 3avlZHevlpPdvVpedveLOdndL9YjTzEvu/vFnOzuF3Oyu1/Mye5+MSe7e0U9u1JVsytVNbtSVbMr 1Zz3uv4gkVpOdPeBR6K7c3j54mh0h4sj0R0upq22Xqy3Z/TiSHSHn+dodg9UR8J7oDqS3uHqaHwP VEfye6A6EuAD1ZEEH6gqp18cSS3t9Cs15fQrNeX0KzXl9Cs15fQrNe30G4rK6TcU65GnqJ1+Q1E5 /YaicvoNReX0G4rK6VeK2hZzKCpbzKGobDGHorLFHIrKqTcUlVNvKCqnXilqp95QVE69oaicekNR OfWGYs6pd6a9U9PSG6rKqTcUlVNviIly6g1F5dQbisqpNxSVU+9ONJWDOp2ZVXPra1pufU3Lra9p ufU1Lbe+puZWilpupViPPEU1t1LUcitFLbdS1HIrRS23vqjmVopabqWo5VaKWm6lqOVWilpupajl 1hfV3EpRy60UtdxKUcutFLXzbqiqyZWqdt4NVe28G6KinXdDVTvvhupweoff648qtXPwTlzz15hp ozGtLaOlpiyjpaYso6WmLKOlpiyjpaYto0NRWUaHYj3yFLVldCgqy+hQVJbRoagso0NRWUZLMS/L +8WcLO8Xc7K8X8zJ8n4xJ8v7RWUNHYo5Wd4r5mV5v5iT5f1iTpb3izlZ3i+qWZaqmmWpqlmWqppl iYqaZamqWZaqtoYOVTW/IaLKiao1XVfPxb6mnYt9TTsX+5p2LvY17Vzsa+q5WIrauViK9chTVM/F UtTOxVLUzsVS1M7FUtTOxb6ofdwbisrHvaGofNwbisrHvaGofNwbiiP5HS4q74GlqL0HDkXlPXAo Ku+BQ1F5DxyKan6lquZXqmp+parmV6Ki5leqan6lquZXqmp+Q0SVg3sGj6jl19e0/Pqall9f0/Lr a1p+fU3NrxS1/EqxHnmKan6lqOVXilp+pajlV4pafn1R3cYKVW0bK1S1baxQ1baxQnUkwweqIyE+ UB1J8XB1NMYHqiM5PlAdCfKB6kiSD1TVKEtVjbJU1ShLVY2ypEaNslTVKEtVjbJU1SiHtObOOKTt NGeCY7+UM8CxW8qZ39gt5Y9vSCl/fENKyvhGqOWPb4RaXX9qyvhGqOWPb4Sall8p5sxvHNv7yfZs Wy3q+ZWqml+pqvmVqppfqar5laqaX6mq+fVVPb9SVfMrVTW/UlXzK1U1v1JV8ytVNb9SVfMrWVHz K1U1v1JV8ytVNb8ho7kHeX26Pa3kV0r5+ZVSfn6llJ9fKeXnV0rI7/B7fZVJyXKo1fWnOZrlnRml UFVOx6GoxDkUldNxKCqnYylqW1uhqGxthaKytRWKytZWKCpbW6GobG2ForK1JUVtaysUla2tUFS2 tkIxZzm9+3oqO9Kh1tKP6bz96L1aWz+qc6O7d/xo29GhqGxHh6KyHb2TzUZ+N6mnzZxfVRguNpUu JcWZ6cjdzio9SoptpUVJcfQcfKCaphGk0eweqOald7+qpVeKWnqlqKVXilp6fVFNrxS19EpRS68U tfRKUUuvFLX0SlFLry+q6ZWill4paumVonYKDlXtFByqrdixrZ6CQ7UdO7rVU3Co1mLHt3oKDtVG 7PjWo9yoNfQo+6IaZV9Uo+yLapR9UY2yL+pRlqoaZanWY09Vj7JUtShLUYuyFLUoS1GLsi+qUZai FmUpalGWohZlKWpRlqIWZSlqUfZFNcpS1KIsRS3KUlSjLFU1ylJtxY5tPcpSbceObj3KUq3Fjm89 ylJtxI7v3CjLUqA5M5Mb5b1iXpT3inlR3ivOassPX8yL8l4xN8r71bwo71frsaeqr6mlqkVZilqU pZj/Kw6hlvMOebemBlmKWpClqAVZilqQpagFWYpakKWoBdkX1SBLUQuyFLUgSzEvyHuvWG6Q96ut 2JGdG+T9ajt2bOtra6nWYkd3bpD3q43Y0a2fk2fqbf2c7IvqOdkX1XOyL6rnZF9Uz8m+qJ+Tpaqe k6Vajz1V/ZwsVS3IUsz5VSV5/yJFZa+rPnPgvX7eLz8MA6i5lqKWaylquZailmsparmWopZrKWq5 9kU111LUci1FLddSVE/QUlVP0FJtxQ50/QQt1XbsUNdP0FKtxQ52/QQt1UbsYNc+ear767AonzxJ TfnkSWrKJ09SUz55kpryyZPUtE+eQlH55CkU65GnqH3yFIpamqWonZalqK2wpaitsOXaN1qApagF WIpagKWoBViKWoClqAVYilqAfVENsBS1AEtRC7AU1QBLVQ2wVNUAS1UNsFTVAEtW1ABLVQ2wVNUA S1UNcMho/lxTY7ql/gaE1JTRLakpo1tSU0a3pKaMbklNG90KRWV0KxTrkaeojW6FojK6FYrK6FYo KqNboaiMbklRG90KRWV0KxSV0a1QVEa3QlEZ3QpFZXQrFJXRLSlqo1uhqIxuhaIyuhWKWoBDVQtw qGoBDlUtwKGqBThkRQtwqGoBDlUtwKGqBXgno7mfxMjl6/I/cJJS/gdOUsr/wElKytlXasrZV2ra 2TcUlbNvKCpn31BUzr6hqJx9Q1E5+4aicvYNReXsK0VtbisUlbmtUFTmtkJRmdsKRWVuKxSVua1Q VOa2pKjNbYWiMrcVisrcViiq4ZWqGl6pquGVqhpeqarhlaCo4ZWqGl6pquGVqhreEFDlvWKjPjur vi2Wova2WIra22IpaiH2NS3EvqaGWIpaiKWohViKWoilqIVYilqIpaiFWIpaiH1RHf4IVW34I1S1 4Y9Q1YY/QlUb/ghVbfgjVEOUh9/rK0p1ECRUtUGQUNUGQUJVjbZU1WhLtRU71vVoS7UdO9r1aEtV jbZU1WhLVY12CK8W7WZzWo+2L6rR9kU12r6YPxAipfyBECkpA12hlj8EEmr5QyChlj/QFWpaqKWo hVqKWqilqIXaF/VQS1UNtVTVUEtVDbVU1VBLVQ21VLWJLqnqQZaqGmSpqkGWqhpkqapBlmordmTr QZZqO3Zs60GWqhpkqapBlqoa5BBV5YOZxkytrn4GJUXtMygpap9BSVH7DEqK2mdQUlQ/gwpV7TOo UK3Hnqr6GVSoaqGWohZqKWqhlqIWal9Ur7MVqtp1tkJVu85WqGrX2QrVvFDvV/NCvV/NC/VeNTfU +9W8UO9X80K9X9U+jwpV7fOoUG3FjnL186hQbceOc/XzqFCtxY509fOoUG3EjnT97Dw709TPzr6o np19UT07+6L2eZQUtc+jpKh+HhWq2udRoVqPPVX186hQ1UItRS3UUtRCLUUt1L6o7WCHorKDHYrK DnYoKjvYoajsYIeisoMdisoOthS1HexQVHawQ1HZwQ5F5feQQ1G57FYoKpfdCkXlsluhqFx2K+RF uexWKCqX3QpF5bJboahcdmsno8ph3fQwWoKlqCVYilqCpaglWIpagqWoJjhUtQSHaj32VNUEh6qS 4FBUEhyKSoJDUUmwFJVL6IVa/iX0Qi3/EnqhlrMFtlfLv3hPqOVsgO3Vcva/dmt52197tZzdr71a zubXXk3JbSgquQ3FnNwOv9ffFyoZDkUlwyExSoZDUclwKCoZDkUlwzspVdabzXR6Wl1aS1FbWktR W1pLUVtaS1FbWktRXVqHqra0DtV67KmqS+tQ1TIsRS3DUtQyLEUtw76oLq1DVVtah6q2tA5VbWkd qtrSOlS1pXWoaktrqapL61DVltahqi2tQ1XLtRS1XEtROR+HopZlKWpZluRoWZailmUpalmWopbl kFbtJFVv6b9AIUX1fOyL6vnYF9XzsS+q52Nf1M/HUlXPx1Ktx56qfj6WqpZlKWpZlqKWZSlqWfZF de8rVLW9r1DV9r5CVdv7ClVt7ytUtb2vUNX2vqSq7n2Fqrb3Fara3leoalmWopZlKWpZlqKWZSlq WZbkaFmWopZlKWpZlqKW5ZBW7QBvpvpvUEhRzbIvqln2RTXLvqhm2Rf1LEtVzbJU67GnqmdZqvm/ mBxq+deVD7X8oetQyx+6lpqeY6mqOZaqmmOpqjmWqppjqao5lqqaY1/VcyxVNcdSVXMsVS3HUtRy LEUtx1LUcixFLceSGi3HUtRyLEUtx1LUchySqh3crdmWnmNfVHPsi2qOfVHNsS+qOfZFPcdSVXMs 1Xrsqeo5lupIjg9UR5J8oDqS5QNVbdhaqnqeparmWapqnqWq5lmqap6lquZZqmqefVXPs1TVPEtV zbNUlet1hWJOnveLrfwpxlCcyR9jDMV2/hxjSM90XQfKy/N+MW3rQHl5rlUPvtffSa92wM829N+o kKKabV9Us+2LarZ9Uc22L+rZlqqabanWY09Vz7ZU1WxLVc22VNVsS1XNtq/q2Zaqmm2pqtmWqppt qarZlqqabamq2fZVPdtSVbMtVTXbUtXO1VLUztVS1M7VUtTO1VLUztWSHu1cLUXtXC1F7VwtRe1c HRKrHOSt6VpNzbMUtTxLUcuzFLU8S1HLsxTVPIeqludQrceeqprnUNXyHKpankNVy3OoanmWqprn UNXyHKpankNVy3OoankOVS3PoarlWapqnkNVy3OoankOVSXPoajkORSVPIeikudQVPIc0qPkORSV PIeikudQVPK8k1jtIMdBqufZF9U8+6KaZ19U8+yLap59Uc+zVNU8S7Uee6p6nqWq5lmqap6lquZZ qmqefVXPs1TVPEtVzbNU1TxLVc2zVNU8S1XNs6/qeZaqmmepqnmWav7FgUIt/+JAoZZ/caBQy784 UKjlXxwoBGe6rrLkXRpor5a2VZbRHO9kFIe1/7fDh/8d8WM4IP2oa9UvaRsz1eZMWq0fMO6QYMZv 1scEs/6fQI8J5N+Yjwlq07ONAkVaK8BEhgs4a41WAWit2S4inUmLSGebRaTyj1wfVrSHFCkMKVCk 9VaBot5qFCga7bRA0UqLSGcaRaSzMxrpbFDUp9FB44oa/hxXpOikcUUd3TSuaKKjxhWtRhHpzEwR qfwqZUzRmE4LSBu1ZgFpI50tIG00agWkaB81/UhOp/37gbwwDL/XHxJHG40XRBuNF0QbjRdEG40X xBuNKKKNRhTRRiOKaKMRRbTRiCLaaEQRbTSiiDYar4g3GlFEG40ooo1GFNFGI4pooxFFtNGIItpo vCLeaEQRbTSiiDYaUUQbjSiijUYU0UYjimijEUW00Uh8o41GFNFGI4poowlNolZAGm809Zma0miG BNHm4gXR5uIF0ebiBdHm4gXx5iKKaHMRRbS5iCLaXEQRbS6iiDYXUUSbiyiizcUrRpuLbAdU8d9j 8luzfkFbO9wacjT1w80hR9NoFD9W83CDyNHMTDcKnlW0zXhFvM2IItpmRBFtM6KIthlRRNuMKKJt RhTRNiOKaJuR8EbbjCiibUYU0TYTWkStgDSnzchLP1PFe8Xq7jW8m3kv3u6B36w3RluN/+H9o6iB NtE63CpGJTMjvWKnemxH25DLsdVmb0OFlnAbKpxBb0OFF+s2VM369G2oWs3boZ+ZvR36du026NPp RmSh4V+80Z6E2rF29Vj4Vqu3/ZKn2SjU1GfTQk2zNl2oadVnC4ijCx9RRDuSV8Q7kiiiHUkU0Y4k imhHEkW0I4ki2pFEEe1Iooh2JK+IdyRRRDuSKKIdSRTRjiSK6MKn1W7FFz5eEF34eEF04eMF0YWP F0QXPl4QX/iIIrrwEUV04SOK6MJHFNGFjyiiCx9RRBc+oogufLwi513VXrMKK46W/Ctm7dtRNdLW baia8ZbVlGHGWKsR7mirEcXhVjP8Xv/A4e3V8bYjimjbEUW07Ygi2nZEEW07ooi2HVFE244oom1H whxtO6KIth1RRNtOaBm1AtJ425ltzsbbjhdE244XRNuOF0TbjhdE244XxNuOKKJtRxTRtiOKaNsR RbTtiCLadkQRbTuiiLYdryha28zIjGW8UYimYG0jmoK1jWiiDUeIow1HFNG1jVfEm4wook1GFNEm I4pokxHFSJM5+O5yxr98renpQs1MOluomW22CjXtw+kY1TRqtbRQkzZymGeqO2+dgqbeahdqGu0Z 9QhPp9P4Jo8IYk1HBLGmI4JY0xFBrOmIINp0giLWdIIi1nSCItZ0giLWdIIi1nSCItZ0giLWdERR 1HRmZRA03nREU9B0RFPQdEQz2nTwQ8PrpXbLv2WqFT2vWWWzKNxZrd2oV9N2a7pIU5+enS3UpLXC x6rXG41CTWMmx8PZA/7UW61izUx7ulDTrs0WaRrThzcmcjS1mUahpj5dyNxo1AuZG81mJFW12Xq8 BXlBtAV5QbQFeUG0BXlBtAV5QbwFiSLagkQRbUGiiLYgUURbkCiiLUgU0RYkimgL8oroh1hBEfsQ KyhiH2IFRexDrKCIfYgVFLH1TlDE1juiiK53giK23gmK2HonKGLrnaAYfVN18I0pjkH/efh0rAWk 0zLm2o6tivDyNotWPEETX/GIpmDFEzTxFU/QxFc8QRNd8dQbzXi78YJou/GCaLvxgmi78YJou/GC vXZz4HN9VR1tPaKIth5RRFuPKKKtRxTR1iOKaOsRRbT1eEW89Ygi2npEEW09ooi2HlFEW48ooq1H FNHW4xXx1iOKaOsRRbT1iCLaekQx0noONpVUBnGj65ygia9zgia+zhFNwTonaOLrnKCJr3OCJr7O CZqcdU7DH9mNqvz2xUxe4zkoyGk8BwU5jeegIKfxHBTkNJ6Dgrx1ziFFTrM5pMhpNocUOc3mkCKn 2RxS5DSbQ4qcZnNIkdNsfHFfkddsDilyms0hRU6zOaTIaTaHFDnN5pAip9kcUuQ0mx3FbFDkNZtD ipxmc0iR02wOKXKazSFFQbOpy5xwvNmIpqDZiKag2dRlfjjebERT0GxEU9BsRFPQbEQTfVPVmpmO r3K8ILrK8YLoKscLoqscL4iucrwg/qZKFNGVjSiiKxtRRFc2ooiubEQRXdmIIrqyEUV0ZeMVBcM7 abOdFg3vBE18eCdo4sM7QRMf3gmakeGdHM1svZhZ29cZ0qj7OsMabV9nWKPt6wxr4vs6abNeuK8T NAUtSDQFLchrilqQaApakGgKWpBoClqQaKItyL/g0RbkBdEW5AXRFuQF0RbkBdEW5AXxFiSKaAsS RbQFiSLagkQRbUGiiLYgUURbkCiiLcgrCraW01Z7tmhrOWjiW8tBE99aDpqRreUczUyrVaiZbUej 4TVIRlqoqTWnCzXpzGyhpjG6R3Pgvf5hvQzZFNxnQTsSTUE7Ek1BO/KaonYkmoJ2JJqCdiSagnYk mmb+3NtM2BnD4reNA2l0zjBH1BpZ4R7aY/OimVZaLJptTxeLYvOGQ6rIvOGQKjJvOKSKzBsOqSLz hkOqyLzhkGp03nCoM7XbjZzedfCU3K75z1vqrSJNOt2K9qbZ1nRh/xJNQf8STUH/Ek1B/xJNQf8S TUH/8pqi/iWagv4lmoL+JZpGfI9ZNAU9SzQFPUs0BT1LNAU9y2uKepZoCnqWaAp6lmgKepZoIj1L ZqPr/tdpIz1rXxTpWTKpK6JIz9oXRXrWvqhwRjqoimakg6poRjqoimakg6pZ1xO5c4KvtfJGtXZ6 Ud3PgkVXWkERW2kFRWylJYroNnZQxLaxgyK2jR0UsW3soIhtYwdFbBs7KGLb2KKIbmMHRWwbOyhi 29hBMbqzlHOU1Jszt3H0jn6Klquard/G0Vtv3070GtPRPidH72h/OvBrDvVp+SfTRj5Lmx76nCxo Rj9LG9WMfJZ2ME1IQbVx+CUZlTQPvyajkpla4QPNHh4WHZW0D0+LjkpqtelC3lqaFgLX6s1C4lpj thA53ozqtYLfSQ2KaDMSRbQZeUW8GYki2oxEEW1Goog2I1FEm5Eoos1IFNFm5BXxZiSKaDMSRbQZ iaKed+Dvx9m//gVLpKCJL5GCJr5ECpr4EinEOb5ECpr4Eilo4kukoIkvkYJmZIl0MEN1/yvz8Rbk JQUtyEsKWpCXFLQgLyloQV4SWtCB9/pxfUE7Ek1BOxJNQTsSTbQdNWYLBhyDItqORBFtR14Rb0ei iLYjUUTbkSii7UgU0XYkimg7EkW0HXlFvB2JItqORBFtR6IoaEd1uRpHvB2JpqAdiaagHYmmoB1J tAvakWgK2pFoCtqRaArakWgK2lHT/2J9vB15SUE78pKCduQlBe3ISwrakZcUrYhEU9CCRFPQgkRT 0IJEE21BrUajoAWJYrQFHVpuN2dzf3E1T1WbibQA/2jxViSKaCsSRbQViSLaikQRbUWiiLYir4i3 IlFEW5Eooq1IFAWtqCnX64i3ItEUtCLRFLQi0RS0Iol1QSsSTUErEk1BKxJNQSsSTUEras2kRa3I SwpakZcUtCIvKWhFXlLQirykqBWJpqAViaagFYmmoBWJJtqKZqfzfrX1wI60/FpRbTb6oX7QjOxs H9gBFpq0lkbajKeJNyJRRBuRKKKNSBTRRiSKaCMSRbQReUW8EYki2ohEEW1EoihoRN73okYkmoJG JJqCRiSagkYkoS5oRKIpaESiKWhEoiloRKKJDQLU2634L7aKIDYIIILYIIAIYoMAIogNAoggOggQ FLFBgKCIDQIERWwQIChigwBBEV3/iCL6FkwU0bdgXhF/CyaKaLsRRbTdiCLabkQRbTeiiLYbUUTb jVfE240oou1GFFq7GX6vv68uaD2yr1zQekRT0HpEU9B6RFPQemblEh7x1iOagtYjmoLWI5qC1iOa WOtp1OrxX28VQaz1iCDWekQQaz0iiLUeEURbT1DEWk9QxFpPUMRaT1DEWk9QxFpPUMRaT1DEWo8o oq0nKGKtJyhirScoYq0nKGKtJyhirScoYq1HFNHWExSx1hMUsZVOUBS0m7ZcsiPebkRT0G5EU9Bu RFPQbtrtwpHHoCloN6IpaDeiKWg3oimaMWqk9ZnCGSMRFc0YiahoxkhEo5/XD+XNX10v2n68IN5+ RBFtP6KIth9RRNuPKKLtRxTR9iOKaPsRRbT9hMsjRt+wNdJas2gKO2jiU9hBE5/CDpr4FHbQxKew gyY+hR008Sls0RRMYQdNfAo7aEamsA9EroHjoaglBU28JQVNvCUFTbwliaagJQVNvCUFTbwlBU28 JQVNfBeoUfeX+IjuAokkvgskkvgukEhmG/pVHBsgjbciL4i3IlFEW5Eooq1IFNFWJIpoKxJFtBWJ ItqKRBFtRV5RMI0trhdMMwZNfJoxaOLTjEETn2YMmvg0Y9DEpxlFUzDNGDTxacagiU8zBk18mjFo ClqRaApakWgKWpFoClqRxLqgFYmmoBWJpqAViaagFYmmoBU1/KU+4q3ISwpakZcUtCIvGW1Fh5ZO zVatePI6qIomr4NqZ4px+L1+wU8UTWEHVdEUdlC1Ym+EW7WCmcagiPYrUUT7lVcU9auGXBck3q9E U9CvRFPQr0RT0K9EU9CvRFPQrxpyjZB4vxJNQb8STUG/Ek1BvxJNQb8STUG/Ek1BvxJNQb+S7Bf0 K9EU9CvRFPQr0RT0K9GM9qtGeN0b0kZawB3tVyOS0X41IhntVyOS0Q/QRiSjH6CNSHI+QBvVjH6A NqoZ/QBtVDP6AdqoJvYBWsNfoC7elkQRbUuiiLYlr8hpS/us7cCa05ZGNaNtaVQz2pZGNaNtaVQz 2pZGNaNtqREO831NTlsa1Yy2pVHNaFsa1Yy2pVHNaFsa1Yy2pVHNaFsa1Yy2pRFNTlsa1Yy2pVHN aFsa1Yy2pVFNwTLKf5pcsIzykoJllJcULKO8JP65vkjin+uLpOBz/aCJf64fNPHP9YMm/rl+0ETb UrtR8Lu2QRFtS6KItiWviO9ziyK6zy2K6D63KKL73KKI7nOLIrrPLYroPrdXxPe5RRHd5xZFdJ9b FLELGQVF7MLUQRG7MHVQxC5MHRSxC1OLInph6qCIXZg6KGIXpg6K2IWpg2LkwtQHE9P2VxNpFUoK Go6XFDQcLyloOF5S0HC8pKjhiKag4YimoOGIpqDhiCbWcLBeKJhpDIpYwwmKWMMRReE18pvTjdni a+QHVdE18oNq5Br5uarWbHobqtHrx+ap2vXboK9Pt26Dvl5rx+iH3+vv/kQ9vY1nUm80p9VQyisV bVJBEWtSQRFrUkERa1KiiDapoIg1qaCINamgiDWpoIg3qWatXjTtKJJ4kxJJvEmJJN6kRBJvUiIp aFJBE29SQRNvUkETb1JBE21SaatgDykook1KFNEm5RUFe0jCWrCHFDTxPaSgie8hBU18Dylo4ntI QRPfQxJNwR5S0MT3kIImvocUNCNv1oZD51+JeBMSRbQJiSLahEQRbUJeEW9Coog2IVFEm5Aook1I FAVNKG0X/T6sSAqakJcUNCEvKWhCXlLQhLykqAmJpqAJiaagCYmmoAmJJtqEGmnB78MGRbQJiSLa hLyiqAl51qImJJqCJiSagiYkmoImJJqCJiSagibkNUVNSDQFTUg0BU1INNEm5F+JeBMSRbQJiSLa hEQRbUJeEW9Coog2IVFEm5Aook1IFAVNqNEs+o1YkRQ0IS8paEJeUtCEvKSgCXlJURMSTUETEk1B ExJNQRMSzWgTOrSIb9ab0X9DcUgV+TcU91WxX0UbUo38KtrBgDfkmiHxpiSagqYkmoKmJJqCpiSa gqYkmoKm1JDrh8SbkmgKmpJoCpqSaKJNqdku+FdegyLalEQRbUqiiDYlr4g3JVFEm5Iook1JFNGm JIqCptSqFf1erEgKmpKXFDQlLyloSl5S0JS85FBTGn6vn68vaFCiKWhQoiloUKIZaVA5mpnil6Q2 2ypmbkcuAtMOmtFfTcvRFDQm0RQ0JtEUNCbRFDQm0RQ0JtEUNKZWrV3YmERT0JhEU9CYRBNtTDPN on0jUUQbkyiijUkU0cbkFfHGJIpoYxJFtDGJItqYRFHQmGZmC/eNvKSgMXlJQWPykoLG5CUFjclL ilZLoiloRqIpaEaiKWhGoiloRqKZKX4ZipqRaAqakdcUNSPRFDQj0RQ0I9EUNCPRFDQj0RQ0I9EU NCOvKWpGoiloRqIpaEaiiTYjfy3OeDMSRbQZiSLajEQRbUZeEW9Goog2I1FEm5Eoos1IFAXNqN1o FDUjLyloRl5S0Iy8pKAZeUlBM/KSomYkmoJmJJqCZiSagmYkmoJmJJqZ4pehqBmJpqAZeU1RMxJN QTMSTUEzEk1BMxJNQTMSTUEzEk1BM2rLtUTizUg0Bc1INAXNSDSxZtSanpmNN6OgiDWjoIg1o6CI NSNRRJtRUMSaUVDEmlFQxJpRUMSbUas23SpoRiKJNyORxJuRSOLNSCTxZiSSgmYUNPFmFDTxZhQ0 8WYUNPFmFDQzxS9DQTMKmngzEk1BMwqaeDMKmngzCpp4MwqaeDMKmngzCpp4MxJNQTMKmp1mdGk8 vNf/Tun/LUX18cYUNI3p+C5hq+Z/5SlyRdoh1cxt3VfkirRDqnbBL5eKavSKtLmqtHYb9I164zbo G43Do5atmvyDR4129dT9Ryeb9War6q9SLRtKOBENS6erxxo4gJXqsZ2fbaVN/29btg//rPxofvH8 ytHJo5P33fcbS2cvXDm38OjC5eUrK2cXPlB9YnlptfqRhQtPL6wunp0/Ojm19+fq1G8uLp1bvnr5 5Pzq4lLtgaWzy+cWl86jAU1XH3zosUf9HQLR7521a9WLq9Wjk+/98Puql49OVtuNmWq1CV249TG5 9ejkwtI5+XbpwvKqh3lg6dwjj3964exquHX+/MKe5vKTy1fDDV748ZX5xQsLK/Izjzx4dPIdY+H4 /hdj/3zs6OSlS3OX559eqK4sXF5dXsGPjB2R+i+PlcbGx47Jn98+dmrsocdO/dvjD5x6rHrikZOP PHr/Aw9WP4gaojL2b3a+194/PT122/+7C1/9UvjzT3/607EKvh8br/wcHzHB13YpPNrYzuNNjE3g +5FxD/J2fM1OjOOWc+Pf/fG58d/Z/sxd/sv/+bs/frOdOCLf/+FNd6Sy87j5zlwaf/3Hl8ZXthfv 8l/+z6+/Cc6UDjgzsUN4ZOxt+H5f2VOtbHsST+TJ/N9/FipA/ROpxvZ8O7Ln211CVRrb9enNJtn3 Z5jkZ32lfj6efKT0+o/x1bvznkxVXv/xVIXhOPm70od7f1diOE4+OPH6jz848eHenffk3x35cO/f HeHIzusdHLEdhn7y4d7iXR8myI5P8N/dgRTneXLPYKpyz4Chn3yps3jXlwiOE5B0QdKl6LF9nJP7 b/YKYdSTf1b62VP8P3etMnycLFS+1LnznvwfE1/qfHfiSwSe/OKRL3VPHvnZjtiflycf7n33ZzwD /nw8OZN8uPefkjefZNST//PI6/0fHHnzUzzqicMZ8J+VGDz5j+UP914uMxwnv1u6Z3C99LOdAX8+ nvw/lQ/3EorsfKT0YXT7ewh6LEh6IOkxnHde705VvknQY/36BO+8unfekw9OfLP7wQmG845fUX+k 9E2O4wQr6m8S9BO8L+7/XekegvMO3p0PQPMmvMsYU9b2P79H3t0DzFvJV4b2AO/e2QN8y9h/GE86 fzB+l3yty1eCryP4e/jy9Tf7vUdl5DX7g/G086f4ulPvxypDJBlIMgpPDEgMhScWJJbCEwcSR+FJ 0v1TfFFkByRplyI7IMkoPDEgMRSeWJBYCk8cSByFJ0kP2elRZAckaY8iOyDJKDwxIDEUnliQWApP HEgchSdJH9npU2QHJGmfIjsgySg8MSAxFJ5YkFgKTxxIHIUnyQDZGVBkByTpgCI7IMkoPDEgMRSe WJBYCk8cSByFJ8k2srNNkR2QpNsU2QFJRuDJfx1POt+7Azs5o5781/EUJGmHwZMMJBmFJwYkhsIT CxJL4YkDiaPwJOl+7w7s5ORmByRplyI7IMkoPDEgMRSeWJBYCk8cSByFJ0nve3dgJyc3OyBJexTZ AUlG4YkBiaHwxILEUnjiQOIoPEn637sDOzm52QFJ2qfIDkgyCk8MSAyFJxYklsITBxJH4Uky+N4d 2MnJzQ5I0gFFdkCSUXhiQGIoPLEgsRSeOJA4Ck+S7e/dgZ2c3OyAJKXYP8lAkhF40hlPOu8qMczk dMZTkDDM5HTGM5BkFJ4YkBgKTyxILIUnDiSOwpOki+x0KbIDEoaZHGQHJBmFJwYkhsITCxJL4YkD iaPwJOkhOz2K7ICEYSYH2el1xo9TeGJAcpHCEwuS5yk8cSB5icKTpN8Zf40jO31kp0+RHZAwzOQg OyAxFJ5YkFgKTxxIHIUnyQDnnQFFdkDCMJOD7IAko/DEgMRQeGJBYik8cSBxFJ4k28jONkV2QJJS 7J9kIGGYyTnmd3FK9xK8Lz5WSjsPlxhmco75XZzScQpPDEguUnhiQXKTwhMHkg6FJ0kX2elSZAck J7sU2QHJVQpPDEiuUXhiQXKTwhMHkg6FJ0kP2elRZKf3cIlhJgfZ6f1CKaPwxIDEUHhiQWIpPHEg cRSeJP2HSwwzOcgOSBhmcpAdkGQUnhiQGApPLEgshScOJI7Ck2TwcIlhJgfZAQnDTA6yA5KMwhMD EkPhiQWJpfDEgcRReJJsP1ximMlBdkCSUuyfZCBhmMk5W0o6axQzOWdLKUgYZnLOljKQZBSeGJAY Ck8sSCyFJw4kjsKTpLtGMZOD7ICEYSYH2QFJRuGJAYmh8MSCxFJ44kDiKDxJemsUMznIDkgYZnKQ nd7Z0lUKTwxIrlF4YkFyk8IT1ztechSeJP21EsN1cpAdkDDM5CA7IMkoPDEgMRSeWJBYCk8cSByF J8lgjWImB9kBCcNMDrIDkozCEwMSQ+GJBYml8MSBxFF4kmyvUczkIDsgSSn2TzKQMMzkvFBKOl8t MVwn54VSChKGmZwXShlIMgpPDEgMhScWJJbCEwcSR+FJ0v1qieE6OcgOSBiuk4PsgCSj8MSAxFB4 YkFiKTxxIHEUniS9r5YYrpOD7ICEYSYH2ek9QzGTg+yAxFB4YkFiKTxxIHEUniT9r1LM5CA7IGGY yUF2QJJReGJAYig8sSCxFJ44kDgKT5LBVylmcpCdwQuluwcU2QHJcQpPDEguUnhiQfI8hScOJC9R eJJsv1B6jSM7ILmbYv8kA8lxgj2lb5WSzg8pZnK+VUpBwjCT861SBpKMwhMDEkPhiQWJpfDEgcRR eJJ0f0gxk4PsgIRhJgfZAUlG4YkBiaHwxILEUnjiQOIoPEl6P6SYyUF2QMIwk4Ps9L5Syig8MSAx FJ5YkFgKTxxIHIUnSf9bpXv7FNnp/5BiJgfZgSfHKTwxILlI4YkFyfMUnjiQvEThSTL4Vuk1juwM vkIxk4PsgIRhJgfZAYmh8MSCxFJ44kDiKDxJtr9CMZOD7IAkpdg/yUDCMJMzUU46/6rMMJMzUU5B wjCTM1HOQJJReGJAYig8sSCxFJ44kDgKT5IustOlyA5IGGZykB2QZBSeGJAYCk8sSCyFJw4kjsKT pIfs9CiyAxKGmRxkByQZhScGJIbCEwsSS+GJA4mj8CTpIzt9iuyAhGEmB9kBSUbhiQGJofDEgsRS eOJA4ig8SQbIzoAiO4OJ8skBRXYG/43iOjnIDjy5SOGJBcnzFJ44kLxE4UmyPVF+jSM7ILmbYv8k AwnDTM6vlZPOJ8sMMzm/Vk5BwjCT82vlDCQZhScGJIbCEwsSS+GJA4mj8CTpfrLMMJOD7ICEYSYH 2QFJRuGJAYmh8MSCxFJ44kDiKDxJep8sM8zkIDsgYZjJQXZAklF4YkBiKDyxILEUnjiQOApPkv4n ywzXyUF2QMIwk4PsgCSj8MSAxFB4YkFiKTxxIHEUniSDT5YZrpOD7Ax+qcwwk4PsgCSj8MSAxFB4 YkFiKTxxIHEUniTbv1RmmMlBdkCSUuyfZCBhmMn57XLS+TzFTM5vl1OQMMzk/HY5A0lG4YkBiaHw xILEUnjiQOIoPEm6n6eYyUF2QMIwk4PsgCSj8MSAxFB4YkFiKTxxIHEUniS9z1PM5CA7IGGYyUF2 QJJReGJAYig8sSCxFJ44kDgKT5L+5ylmcpAdkDDM5CA7IMkoPDEgMRSeWJBYCk8cSByFJ8ng8xQz OcjO4FSZ4d+uQnZAklF4YkBiKDyxILEUnjiQOApPku1TZYZ/uwrZAUlKsX+SgYTh3676cjnp/BXF TM6XyylIGGZyvlzOQJJReGJAYig8sSCxFJ44kDgKT5LuX1HM5CA7IGGYyUF2QJJReGJAYig8sSCx FJ44kDgKT5LeX1HM5CA7IGGYyUF2el8uH6fwxIDkIoUnFiTPU3jiQPIShSdJ/8vl1ziyA5K7+xTZ AQnDdXKQHZBcpPDEguR5Ck8cSF6i8CQZIDsUnqSD36eYyUF2QMIwk4PsgMRQeGJBYik8cSBxFJ4k 279PMZOD7IAkpdg/yUDCMJPzN+Wk81OKmZy/KacgYZjJ+ZtyBpKMwhMDEkPhiQWJpfDEgcRReJJ0 f0oxk4PsgIRhJgfZAUlG4YkBiaHwxILEUnjiQOIoPEl6P6WYyUF2QMIwk4Ps9L5GMZOD7IDEUHhi QWIpPHEgcRSeJP2vUczkIDsgYZjJQXZAklF4YkBiKDyxILEUnjiQOApPksHXKGZykB2QMMzkIDsg ySg8MSAxFJ5YkFgKTxxIHIUnyfbXKGZykB2QpBT7JxlIGGZy3llJOu+s3EvwvvidlRQkJzsMnmQg uUrhiQHJNQpPLEhuUnjiQNKh8CTpIjtdiuyA5GSXIjsguUrhiQHJNQpPLEhuUnjius2Ko/Ak6b2z 8hpHdkByd48iO73XKa6Tg+yAxFB4YkFiKTxxIHEUniT91ymuk4PsgIThOjnIDkgyCk8MSAyFJxYk lsITBxJH4UkyeJ3iOjnIDkgYZnKQHZBkFJ4YkBgKTyxILIUnDiSOwpNk+3WKmRxkByQpxf5JBhKG mZyPVpLOR+/ATs6oJx+tpCA52WHwJAPJVQpPDEiuUXhiQXKTwhMHkg6FJ0n3o3dgJyc3OyA52aXI DkiuUnhiuu+tGApPbPepiqXwxIHEUXiS9J6qMMzkIDsgYZjJQXZAklF4YkBiKDyxILEUnjiQOApP kj6y06fIDkgYZnKQHZBkFJ4YkBgKTyxILIUnDiSOwpNkgDXbgCI7IGH4t6uQHZBcpfDEgOQahScW JDcpPHGD91YchSfJ9kcrr3FkByR3U+yfZCBh+Lerfq+SdOYqDNfJ+b1KChKG6+T8XiUDSUbhiQGJ ofDEgsRSeOJA4ig8SbrITpciOyBhuE4OsgOSjMITAxJD4YntXqtYCk8cSByFJ0nvWoXhOjnIDkgY rpOD7IAko/DEgMRQeGJBYik8cSBxFJ4k/WsVhpkcZAckDDM5yA5IMgpPDEgMhScWJJbCEwcSR+FJ MsCabUCRHZAwzOQgOyDJKDwxIDEUnliQWApPHEgchSfJNrKzTZEdkKQU+ycZSBhmcv68knS+UGG4 Ts6fV1KQMFwn588rGUgyCk8MSAyFJxYklsITBxJH4UnS/UKF4To5yA5IGK6Tg+yAJKPwxIDEUHhi u69SzOQgOyBxFJ4kvVcpZnKQHZAwzOQgOyDJKDwxIDEUnliQWApPHEgchSdJ/1WKmRxkByQMMznI DkgyCk8MSAyFJxYklsITBxJH4Uky+EKF4To5yA5IGK6Tg+yAJKPwxIDEUHhiQWIpPHEgcRSeJNtf qDBcJwfZAUlKsX+SgYThOjl/X0k6r1DM5Px9JQUJw0zO31cykGQUnhiQGApPLEgshScOJI7Ck6T7 CsVMDrIDEoaZHGQHJBmFJwYkhsIT233bhKXwxIHEUXiS9P6+cm+PIjsgOdmjyE7vbRMZhScGJIbC EwsSS+GJA4mj8CTpv22CYSYH2QEJw0wOsgOSjMITAxJD4YkFiaXwxIHEUXiSDF6hmMlBdkDCMJOD 7IAko/DEgMRQeGJBYik8cSBxFJ4k269QzOQgOyBJKfZPMpAwzOS8ZyLpjE8wzOS8ZyIFCcNMznsm MpBkFJ4YkBgKTyxILIUnDiSOwpOki+x0KbIDEoaZHGQHJBmFJwYkhsIT271/wlJ44rrvmXiJwpOk 956J1ziy00N2ehTZ6d0/wTCTg+yAxFB4YkFiKTxxIHEUniT9+ycYZnKQHZAwzOQgOyDJKDwxIDEU nliQWApPHEgchSfJAGu2AUV2QMIwk4PsgCSj8MSAxFB4YkFiKTxxIHEUniTbyM42RXZAklLsn2Qg YZjJ+a2JpPOZCYaZnN+aSEHCMJPzWxMZSDIKTwxIDIUnFiSWwhMHEkfhSdL9zATDTA6yAxKGmRxk ByQZhScGJIbCEwsSS+GJA4mj8CTpITs9iuyAhOE6OcgOSDIKTwxIDIUnFiSWwhMHEkfhSdL/DMVM DrIDEoaZHGQHJBmFJwYkhsITCxJL4YkDiaPwJBnMTDDM5CA7IGGYyUF2QJJReGJAYig8sSCxFJ44 kDgKT5LtmQmGmRxkByQpxf5JBhKGmZw/nEg6f0Yxk/OHEylIGGZy/nAiA0lG4YkBiaHwxILEUnji QOIoPEm6f0Yxk4PsgIRhJgfZAUlG4YkBiaHwxILEUnjiQOIoPEl6fzbBcJ0cZAckDDM5yA5IMgpP DEgMhScWJJbCEwcSR+FJ0v8zipkcZAckDDM5yA5IMgpPDEgMhScWJJbCEwcSR+FJMrhIMZOD7ICE YSYH2QFJRuGJAYmh8MSCxFJ44kDiKDxJti9SzOQgOyBJKfZPMpAwzOR8YyLpfGPiXoL3xd+YSEFy ssPgSQaSqxSeGJBco/DEdr5PMZPzjQkHEkfhSdL9PsVMDrIDEoaZHGQHJBmFJwYkhsITCxJL4YkD iaPwJOl9n2ImB9kBCcNMDrLT+8bEcQpPDEguUnhiQfI8hScOJC9ReJL0vzHxGkd2QHJ3nyI7IDne p8gOSC5SeGJB8jyFJw4kL1F4kgz+M8VMDrIDEoaZHGQHJBmFJwYkhsITCxJL4YkDiaPwJNn+zxQz OcgOSFKK/ZMMJAwzOYOJpLNJMZMzmEhBwjCTM5jIQJJReGJAYig8sZ17jlgKTxxIHIUnSfeeIwwz OcgOSBhmcpAdkGQUnhiQGApPLEgshScOJI7Ck6SH7PQosgMShpkcZKe3STGTg+yAxFB4YkFiKTxx IHEUniT9TYqZHGQHJAwzOcgOSDIKTwxIDIUnFiSWwhMHEkfhSTLYpJjJQXZAwjCTg+yAJKPwxIDE UHhiQWIpPHEgcRSeJNubFDM5yA5IUor9kwwkDDM5tSNJ5+1HGK6TUzuSgoThOjm1IxlIMgpPDEgM hSe2Uztyk8ITB5IOhSdJt3bk3i5FdkByskuRHZBcpfDEgOQahSe2++gRS+GJA4mj8CTpPXqEYSYH 2QEJw0wOstPDeYfCEwMSQ+GJBYml8MSBxFF4kvSxZutTZAckDNfJQXZAklF4YkBiKDyxILEUnjiQ OApPkgGyM6DIDkgYZnKQHZBkFJ4YkBgKTyxILIUnDiSOwpNkG9nZpsgOSFKK/ZMMJAwzOeePJJ0H jjDM5Jw/koKEYSbn/JEMJBmFJwYkhsITCxJL4YkDiaPwJOk+QDGTg+yAhGEmB9kBSUbhiQGJofDE dj9HMZOD7IDEUXiS9D5HMZOD7ICEYSYH2ek9cCSj8MSAxFB4YkFiKTxxIHEUniT9B44wzOQgOyBh mMlBdkCSUXhiQGIoPLEgsRSeOJA4Ck+SwQNHGGZykB2QMMzkIDsgySg8MSAxFJ5YkFgKTxxIHIUn yfYDRxhmcpAdkKQU+ycZSBhmcv74SNL5HYqZnD8+koKEYSbnj49kIMkoPDEgMRSeWJBYCk8cSByF J0n3d44wXCcH2QEJw3VykB2QZBSeGJAYCk9s9+sUMznIDkgchSdJ7+sUMznIDkgYZnKQHZBkFJ4Y kBgKTyxILIUnDiSOwpOk/3WKmRxkByQMMznIDkgyCk8MSAyFJxYklsITBxJH4Uky+DrFTA6yAxKG mRxkByQZhScGJIbCEwsSS+GJA4mj8CTZ/jrFTA6yA5KUYv8kAwnDTM63jyQdSzGT8+0jKUgYZnK+ fSQDSUbhiQGJofDEgsRSeOJA4ig8SbqWYiYH2QEJw0wOsgOSjMITAxJD4Ynt/ohiJgfZAYmj8CTp /YhiJgfZAQnDTA6y0/v2kasUnhiQXKPwxILkJoUnDiQdCk+S/reP3NunyA5ITvYpsgOSqxSeGJBc o/DEguQmhScOJB0KT5LBjyhmcpAdkDDM5CA7IMkoPDEgMRSeWJBYCk8cSByFJ8n2jyhmcpAdkKQU +ycZSBhmct6SJJ2/pZjJeUuSgoRhJuctSQaSjMITAxJD4YkFiaXwxIHEUXiSdP+WYiYH2QEJw0wO sgOSjMITAxJD4Ynt/kpiKTxxIHEUniS9X0kYZnKQHZAwzOQgO72/pZjJQXZAYig8sSCxFJ44kDgK T5L+31LM5CA7IGGYyUF2QJJReGJAYig8sSCxFJ44kDgKT5IBzjsDiuyAhGEmB9kBSUbhiQGJofDE gsRSeOJA4ig8SbaRnW2K7IAkpdg/yUDCMJPz60nS+V8ThpmcX09SkDDM5Px6koEko/DEgMRQeGJB Yik8cSBxFJ4kXWSnS5EdkDDM5CA7IMkoPDEgMRSe2O58Yik8cSBxFJ4kvfmEYSYH2QEJw0wOstND dig8MSAxFJ5YkFgKTxxIHIUnSR/nnT5FdkDCcJ0cZAckGYUnBiSGwhMLEkvhiQOJo/AkGcwnDDM5 yA5IGGZykB2QZBSeGJAYCk8sSCyFJw4kjsKTZHs+YZjJQXZAklLsn2QgYZjJuZIknY8nDDM5V5IU JAwzOVeSDCQZhScGJIbCEwsSS+GJA4mj8CTpfjxhmMlBdkDCMJOD7IAko/DEgMRQeGK7V5KbFJ44 kHQoPEl6V5J7exTZAcnJHkV2eh9PMgpPDEgMhScWJJbCEwcSR+FJ0v94wjCTg+yAhGEmB9kBSUbh iQGJofDEgsRSeOJA4ig8SQbPU8zkIDsgYZjJQXZAklF4YkBiKDyxILEUnjiQOApPku3nKWZykB2Q pBT7JxlIGGZy/iRJOusUMzl/kqQgYZjJ+ZMkA0lG4YkBiaHwxILEUnjiQOIoPEm66xQzOcgOSBhm cpAdkGQUnhiQGApPLEgshScOJI7Ck6S3TjGTg+yAhGEmB9kBSUbhiQGJofDEgsRSeOJA4ig8Sfrr FDM5yA5IGGZykB2QZBSeGJAYCk8sSCyFJw4kjsKTZPDXFDM5yA5IGGZykB2QZBSeGJAYCk8sSCyF Jw4kjsKTZPuvKWZykB2QpBT7JxlIGGZyvpcknb+kmMn5XpKChGEm53tJBpKMwhMDEkPhiQWJpfDE gcRReJJ0/5JiJgfZAQnDTA6yA5KMwhMDEkPhiQWJpfDEgcRReJL0/pLiOjnIDkgYrpOD7IAko/DE gMRQeGJBYik8cSBxFJ4k/b+kmMlBdkDCMJOD7IAko/DEgMRQeGJBYik8cSBxFJ4kg/JdDDM5yA5I GGZykB2QZBSeGJAYCk8sSCyFJw4kjsKTZBvZ2abIDkhSiv2TDCQMMznvuivp/JhiJuddd6UgYZjJ edddGUgyCk8MSAyFJxYklsITBxJH4UnS/THFTA6yAxKGmRxkByQZhScGJIbCEwsSS+GJA4mj8CTp /ZhiJgfZAQnDTA6yA5KMwhMDEkPhiQWJpfDEgcRReJL0f0wxk4PsgIRhJgfZAUlG4YkBiaHwxILE UnjiQOIoPEkG7bsYZnKQHZAwzOQgOyDJKDwxIDEUnliQWApPHEgchSfJdvsuhpkcZAckKcX+SQYS hpmch+9KOr96F8NMzsN3pSBhmMl5+K4MJBmFJwYkhsITCxJL4YkDiaPwJOkiO12K7ICEYSYH2QFJ RuGJAYmh8MSCxFJ44kDiKDxJeshOjyI7IGGYyUF2QJJReGJAYig8sSCxFJ44kDgKT5I+stOnyA5I GGZykB2QZBSeGJAYCk8sSCyFJw4kjsKTZPDwXfcOKLIDkpMDiuyA5CqFJwYk1yg8sSC5SeGJA0mH wpNkG9nZpsgOSE6+CSRjiic/v0dO8LVdGst1YGJsAt+PjPvy3fianRgfe8vYI3dlcOOT8vWQfGXy 9fBdn8BXtv3Im7LTdJB4Yu81e3Od2n+F8pz6D+NJ5w/G75KvdflK8HUEfw9fvn6nju6JsUk5ustj 18anOn86nnaujTfedJqJPf/2aU6DJgPNHIU3a6AxoFmn8GYDNBY01ym82QSNA80WhTeV7p+OJ91r 45NdikyBJgVNo0uRKdBkoJmj8GYNNAY06xTebIDGguY6hTeboHGg2aLwptJDpnrIVI8iU6BJQdPs UWQKNBlo5ii8WQONAc06hTcboLGguUHhzSZoHGi2KLyp9JGpPjLVp8gUaFLQNPsM3pwBTQaaOQpv 1kBjQLNO4c0GaCxoblB4cws0DjRbFN5UBsjUAJkaUGQKNClomgOKTIEmA80chTdroDGgWafwZgM0 FjQ3KLy5BRoHmi0KbyrbyNQ2MrVNkSnQpKBpblNkCjQZaOYIvHl1vNL53njSeXV8kuB9+KvjU6BJ QdPoMHhzGjQZaOYovFkDjQHNOoU3G6CxoLlO4c0maBxotii8qXSRqe6rd2A3KTdToElB0+hSZAo0 GWjmKLxZA40BzTqFNxugsaC5TuHNJmgcaLYovKn0kKneq3dgNyk3U6BJQdPsUWQKNBlo5ii8WQON Ac06hTcboLGguUHhzSZoHGi2KLyp9JGp/qt3YDcpN1OgSUHT7DN4cwY0GWjmKLxZA40BzTqFNxug saC5QeHNLdA40GxReFMZIFODV+/AblJupkCTgqY5oMgUaDLQzFF4swYaA5p1Cm82QGNBc4PCm1ug caDZovCmso1Mbb96B3aTcjMFmhQ0TYr9mzOgyUAzR+DN0VKl865S0jlamiR4H360NAWaFDQMs0lH S6dBk4FmjsKbNdAY0KxTeLMBGgua6xTebILGgWaLwptKF5nqIlNdikyBJgUNw2wSMgWaDDRzFN6s gcaAZp3Cmw3QWNBcp/BmEzQONFsU3lR6yFQPmepRZAo0KWgYZpP+cfxjvc748V4f3++8N/84fgU0 F0FzhcKbF0HzPGhepPDmFdC8BJpXKLx5AzSvgeYNjkz1kak+MtVn6DdnQJOBhmE2Cecp0BjQrFN4 swEaC5obFN7cAo0DzRaFN5UBzlMDnKcGFJkCTQoahtkkZAo0GWjmKLxZA40BzTqFNxugsaC5QeHN LdA40GxReFPZRqa2kaltikyBJgVNk2L/5gxoMtAwzCZNl97VOVa6t/M+//2Oe3OiNNV5uJR2TpQY ZpPeV/oYvDnemfbf77g37ytdAc1F0FzpMBw3L4PmJo6blwm8mS79ADQd0PyAwpt3d5Gp7vv8dwJv ToDmJGhOdBm8WQLNVdAsUXjzAmiugeYFCm9eBs1N0LxM4c0PQNMBzQ8ovHl3D5nqvc9/ZzhP9XCe 6p0oMcwm/WJprvcLpaz3i6XTBN78YmkdNAY0axTe3ACNBc0GhTdboHGg2STw5kSp0n+4lPRPlBhm k5Ap0KSgYZhNOlE6A5oMNHMU3qyBxoBmncKbDdBY0Nyg8OYWaBxotii8qQyQqcGJEsNsEjIFmhQ0 DLNJyBRoMtDMUXizBhoDmnUKbzZAY0Fzg8KbW6BxoNmi8KayjUxtnygxzCYhU6BJQcMwm4RMgSYD DcNs0rOlSmetlHSepZhNerY0BZoUNAyzSc+WToMmA80chTdroDGgWafwZgM0FjTXKbzZBI0DzRaF N5UuMtV9lmI2CZkCTQoahtkkZAo0GWjmKLxZA40BzTqFNxugsaC5TuHNJmgcaLYovKn0kKnesxSz ScgUaFLQMMwmnS8t9c6Wrvbm8P3Oe3O+9AJoroHmBQpvXgbNTdC8TODNQ6Wt3vGS6z10B3aTcjPV R6b6z5YYrpuETIEmBQ3DbNKzpTOgyUAzR+HNGmgMaNYpvNkAjQXNDQpvboHGgWaLwpvKAJkaPEsx m4RMgSYFDcNsEjIFmgw0cxTerIHGgGadwpsN0FjQ3KDw5hZoHGi2KLypbCNT289SzCYhU6BJQdOk 2L85A5oMNAyzSRulSuerpaSzUWK4btJGaQo0KWgYZpM2SqdBk4FmjsKbNdAY0KxTeLMBGgua6xTe bILGgWaLwptKF5nqbpQYrpuETIEmBQ3DdZOQKdBkoJmj8GYNNAY06xTebIDGguY6hTeboHGg2aLw ptJDpnobJYbrJiFToElBwzCb9O9Lc71nSlnv31PMJv370jpoDGjWKLy5ARoLmg0Kb7ZA40DDMJuE TPWRqf4GxWwSMgWaFDQMs0kbpTOgyUAzR+HNGmgMaNYpvNkAjQXNDQpvboHGgWaLwpvKAJkabFDM Jn2hVB28ULp78Mf4fue9+ULpY6A5DpqPUXhzBTQXQXOFwpsXQfM8aF6k8OYV0LwEmlcovHkDNK+B 5g2OTG0jU9vI1DZFpkBzHDQfI9jb+n6p0vlhKel8n2I26fulKdCkoGGYTfp+6TRoMtDMUXizBhoD mnUKbzZAY0FzncKbTdA40GxReFPpIlPd71PMJiFToElBwzCbhEyBJgPNHIU3a6AxoFmn8GYDNBY0 1ym82QSNA80WhTeVHjLV+z7FbBIyBZoUNAyzSX9Rmut9pZT1/uIO7CaNevMXpXXQGNCsUXhzAzQW NBsU3myBxoGGYTbp26V3979Vurd/E98pMtVHpvrfp5hNuln6GLw53v82vt95b26WroDmImiuUHjz ImieB82LFN68ApqXQPMKhTdvgOY10LzRZ+g3zcFXSungL0pTA4rzFGgy0JwZUJynQGNAs0bhzQ3Q WNBsUHizBRoHmlsU3kxuf6WUbP9FqbJNkSnQpKCZ2qbIFGgy0Jwh8OaecqXzr8pJ554yw2zSPeUp 0KSgYZhNuqd8GjQZaOYovFkDjQHNOoU3G6CxoLlO4c0maBxotii8qXSRqS4y1aXIFGhS0DDMJiFT oMlAM0fhzRpoDGjWKbzZAI0FzXUKbzZB40CzReFNpYdM9ZCpHkWmQJOChmE2CZkCTQaaOQpv1kBj QLNO4c0GaCxoblB4swkaB5otCm8qfWSqj0z1KTIFmhQ0DLNJ95TPgCYDzRyFN2ugMaBZp/BmAzQW NDcovLkFGgeaLQpvKgNkaoBMEbwPf0v5xGCifHJQwvc7781/L80N/lspG/z3O7CbNOpNqXwF3lwc vKXMMJtUKr8ImudB8yKFN6+A5iXQvELhzRugeQ00b1B4U92eKN+9/ZYyw2xSqfwx0BwHDcNs0qPl SueT5aTzaJlhNunR8hRoUtAwzCY9Wj4Nmgw0cxTerIHGgGadwpsN0FjQXKfwZhM0DjRbFN5UushU 99Eyw2wSMgWaFDQMs0nIFGgy0MxReLMGGgOadQpvNkBjQXOdwptN0DjQbFF4U+khU71HywyzScgU aFLQMMwmIVOgyUAzR+HNGmgMaNYpvNkAjQXNDQpvNkHjQLNF4U2lj0z1Hy0zXDcJmQJNChqG2aRH y2dAk4FmjsKbNdAY0KxTeLMBGguaGxTe3AKNA80WhTeVATI1eLTMcN2k+8rNwS+V08F9ZYbZpPvK c6DJQHOGwpt10BjQrFF4cwM0FjQbFN5sgcaB5haFN5Pbv1ROtu8rM8wmIVOgSUHDMJuETIEmAw3D bNLnypXO58tJ53MUs0mfK0+BJgUNw2zS58qnQZOBZo7CmzXQGNCsU3izARoLmusU3myCxoFmi8Kb SheZ6n6OYjYJmQJNChqG2SRkCjQZaOYovFkDjQHNOoU3G6CxoLlO4c0maBxotii8qfSQqd7nKGaT kCnQpKBhmE1CpkCTgWaOwps10BjQrFN4swEaC5obFN5sgsaBZovCm0ofmep/jmI2CZkCTQoahtmk z5XPgCYDzRyFN2ugMaBZp/BmAzQWNDcovLkFGgeaLQpvKgNkavA5itmkT5Sbg1PldPCJO7CbNOrN J8pzoMlAc4bCm3XQGNCsUXhzAzQWNBsU3myBxoHmFoU3k9unysn2J+7AblJupkCTgmZqmyJToMlA c4bAm6+XK52/Kiedr1PMJn29PAWaFDQMs0lfL58GTQaaOQpv1kBjQLNO4c0GaCxorlN4swkaB5ot Cm8qXWSq+3WK2SRkCjQpaBhmk5Ap0GSgmaPwZg00BjTrFN5sgMaC5jqFN5ugcaDZovCm0kOmel+n mE1CpkCTgoZhNum/lD/W+3L5eO9P8P3Oe/NfyldAcxE0Vyi8eRE0z4PmRQpvXgHNS6B5hcKbN0Dz GmjeoPCm2v9y+e7+n+A7RaZAcxw0DNdNQqZAcxE0Vyi8eRE0z4PmRQpvXgHNS6B5hcKbN0DzGmgY rpv0R+Xm4PfL6eCPKGaT/qg8B5oMNAyzSX9UXgeNAc0ahTc3QGNBs0HhzRZoHGhuUXgzuf375WT7 jyhmk5Ap0KSgYZhNQqZAk4GGYTbpR+VK56flpPMjitmkH5WnQJOChmE26Ufl06DJQDNH4c0aaAxo 1im82QCNBc11Cm82QeNAs0XhTaWLTHV/RDGbhEyBJgUNw2wSMgWaDDRzFN6sgcaAZp3Cmw3QWNBc p/BmEzQONFsU3lR6yFTvRxSzScgUaFLQMMwmfbM81/taOet9s8zwb7p9s7wOGgOaNQpvboDGgmaD wpst0DjQbFJ4M9n/Wjnpf7Nc6TN40wRNCpqpPkWmQJOB5gyFN+ugMaBZo/DmBmgsaDYovNkCjQPN LQpvJgfI1ACZGlBkCjQpaBhmk5Ap0GSgOUPhzTpoDGjWKLy5ARoLmg0Kb7ZA40Bzi8KbyW1kavub FLNJyBRoUtAwzCYhU6DJQMMwm3R35V2dd1bu7bwD3++8N3dXToDmJGhOdBi8WQLNVdAsUXjzAmiu geYFCm9eBs1N0LxM4c0PQNMBzQ8ovHl3F5nqvgPfKTIFmpOgOdGlyBRoroJmicKbF0BzDTQvUHjz MmhuguZlAm/+dWWz26y47r+uMMwmvaPyBrx5rXs3vt95b95RqfbeWbm7dze+33lvflKe671ezno/ KTP8m24/Ka+DxoBmjcKbG6CxoNmg8GYLNA40mxTeTPZfLyf9n9yB3aQ8b5qgSUEz1afIFGgy0Jyh 8GYdNAY0axTe3ACNBc0GhTdboHGguUXhzeQAmRr85A7sJuVmCjQpaBhmk5Ap0GSgOUPhzTpoDGjW KLy5ARoLmg0Kb7ZA40Bzi8KbyW1kavsnFLNJyBRoUtAwzCYhU6DJQMMwm/RQ5V2dj1bu7Tx4B3aT Rr15qHICNCdBc6LD4M0SaK6CZonCmxdAcw00L1B48zJoboLmZQpvfgCaDmh+QOHNu7vIVPfBO7Cb lJsp0JwEzYkuRaZAcxU0SwTe1Cvr3fdWTLdeWSPw5lOVje5TFdv9VIVhNulTlU3QONBsUXhT6T1V SXqfqjDMJn2qMgWaFDQMs0mfqpwGTQaaOQpv1kBjQLNO4c0GaCxoblB4swkaB5otCm8qfWSqj0z1 KTIFmhQ0DNdN+lTlDGgy0MxReLMGGgOadQpvNkBjQXODwptboHGgYbhuEtZ+A6z9Blj7DSjWfqA5 CRqGf9MNaz/QXAXNEoU3L4DmGmheoPDmZdDcBM3LA4Z18dbgvRU3qFcYZpMerLwBb14bPFRh+Dfd HqxUtz9auXv7oQrDv+n2YOVjoDkOGoZ/0+3TlcnOXCXpfLpSIXgf/ulKAzQpaKY6DN7MgSYDzWkK b9ZBY0CzRuHNddBY0GxQeLMFGgeaTQpvJrvIVBeZ6lJkCjQpaKa6FJkCTQaa0xTerIPGgGaNwJsv Vja61yq2+8UKw3WTvljZBI0DzRaFN5XetUrS+2KF4bpJX6xMgSYFDcN1k75YOQ2aDDRzFN6sgcaA Zp3Cmw3QWNDcoPBmEzQONFsU3lT6yFT/ixWGf9MNmQJNChqGf9Pti5UzoMlAM0fhzRpoDGjWKbzZ AI0FzQ0Kb26BxoGG4d90w9pvgLXfAGu/AcP6pgmaFDQMs0lY+4EmA80ZCm/WQWNAs0bhzQ3QWNBs UHizBRoHmlsU3kxuI1PbyNQ2RaZAk4JmimL/Zg40GWgYZpNerEx2vlBJOi/egd2kUW9erDRAk4Jm qsPgzRxoMtCcpvBmHTQGNGsU3lwHjQXNBoU3W6BxoNmk8Gayi0x1X7wDu0m5mQJNCpqpLkWmQJOB 5jSFN+ugMaBhmE26Wdnovlqx3ZsUs0k3K5ugcaDZovCm0nu1kvRuUswm3axMgSYFDcNs0s3KadBk oJmj8GYNNAY06xTebIDGguYGhTeboHGg2aLwptJHpvo3KWaTkCnQpKBhmE26WTkDmgw0cxTerIHG gGadwpsN0FjQ3KDw5hZoHGgYZpOw9htg7Td4scJw3aQXK03QpKBhuG4S1n6gyUBzhsKbddAY0KxR eHMDNBY0GxTebIHGgeYWhTeT28jU9osVhusmIVOgSUEzRbF/MweaDDQM1036TmWy80ol6XyHYjbp O5UGaFLQMMwmfacyB5oMNKcpvFkHjQHNGoU310FjQbNB4c0WaBxoNim8mewiU93vUMwmIVOgSUHD MJuETIEmA81pCm/WQWNAwzCbVJ7Y6L5twnbLEwyzSeWJTdA40DDMJv1D5d29v6/c2/v/8P3Oe/MP lROgOQmaEz2G4+Z0720TWa88wTCbVJ5YA40BzTqFNxugsaC5QeHNJmgcaLYovKn03zaR9MsTDLNJ 5Ykp0KSgYZhNKk+cAU0GmjkKb9ZAY0CzTuHNBmgsaG5QeHMLNA40DLNJWPsNsPYbfIdiNuk7lSZo UtAwzCZh7QeaDDRnKLxZB40BzRqFNzdAY0GzQeHNFmgcaG5ReDO5jUxtf4diNgmZAk0KmimK/Zs5 0GSgYZhNeuvEZGd8Ium8dYJhNumtEw3QpKBhmE1668QcaDLQnKbwZh00BjRrFN5cB40FzQaFN1ug caDZpPBmsotMdZGpLkWmQJOChmE2CZkCTQaa0xTerIPGgIZhNqk9sdG9f8J22xMMs0n/cuKV7nsm Xur+Mr7feW/+5cQboHkNNG9QHDfNHjLVQ6Z6DMfN6d79E1mvPcEwm9SeWAONAc06hTcboLGguUHh zSZoHGj+/+buJkZu8zDj+I495Euz+VLT1pESB9KlgJHEg7zYheqrto6QWLa7trrJTF6HSdHGdQtL gQFBSlFdUrZ0WbVNW6vy3iwUuooBtTwqqCUYMLAqcsgh4IgAL0JQHbqHKDE/D/17E1goohyrfQ7E eHZmZ396xo84GjyYrSWymfZ/6Jn+aU9hm/S0N0Nj0Shsk572FmgcmkgimxhNgiaVyCZDk6MpJLJZ oqnQKGyTeO038Npv4LXfIHGeQmPRKGyTeO2HxqFZSGSToknQxBLZFGhyNJlENjWaCs1SIptwpFMj nRolOoXGoplJvH8ToXFoFLZJp71p81eeaU57ocC/w097MzQWzWqjkM0cjUMTSWQTo0nQpBLZZGhy NNsS2ZRoKjS1RDbTlk61dKqV6BQai2a1legUGocmksgmRpOgSSWyydDkaLYlsinRVGhqiWymHZ3q 6FQn0Sk0Fo3C5ybRKTQOTSSRTYwmQZNKZJOhydEUEtmUaCo0tUQ2055O9acltkl0Co1Fo7BNOu0t 0Dg0kUQ2MZoETSqRTYYmR1NIZLNEU6FR2CYd88LhqGeGY57CNumYt4bGolHYJh3zIjQOzUIimxRN giaWyKZAk6PJJLKp0VRolhLZhCOdGo95CtskOoXGolHYJtEpNA6NwjbpijdtfuCZ5so+vJv069lc 8WZoLJrVRiGbORqHJpLIJkaToEklssnQ5Gi2JbIp0VRoaolspi2daq/sw7tJD+wUGotmtZXoFBqH JpLIJkaToEklssnQ5Gi2JbIp0VRoaolsph2d6q54Cp+bRKfQWDQKn5tEp9A4NJFENjGaBE0qkU2G JkdTSGRToqnQ1BLZTHs61V+R2CbRKTQWjcI26Yq3QOPQRBLZxGgSNKlENhmaHE0hkc0STYVGYZt0 zguH1zwznJPYJp3z1tBYNArbpHNehMahWUhkk6JJ0MQS2RRocjSZRDY1mgrNUiKbcKRT4zmJbRKd QmPRKGyT6BQah0Zhm/Se96nmXe9I8w6X+5/Ne946mhNo1huFbE6hOYvmlEQ2W2guo9kSyOa2lzU/ 9fLmtsQ26bZXoqnQ1BLZTNufeqa9LbFNuu3N0Fg0Ctuk294cjUMTSWQTo0nQpBLZZGhyNNsS2ZRo KjS1RDbTjk51tyW2SXQKjUWjsE16x3u2e9c71r3H5f5n8453Bs1raM5IZPM2mkto3pbIZgfNDTQ7 Etn8As3/oPmFRDaH+3e9x/v3uJToFJpjaJ7tJTqF5jU0ZySyeRvNJTRvS2Szg+YGmh2BbK564fAf nhmuSmyTrnpraCwahW3SVS9C49AsJLJJ0SRoYolsCjQ5mkwimxpNhWYpkU040qnxqsQ2iU6hsWgU tkl0Co1Do7BNuuOFTemZ5o7E5ybd8VbRWDQKn5t0x4vQODRziWxSNAkahc9NOuBnzWf8vDngK2yT DvglmgpNLZHNtP2Mb9oDvsI26YA/Q2PRKGyTDvhzNA5NJJFNjCZBk0pkk6HJ0WxLZFOiqdDUEtlM OzrV0alOolNoLBqFbRLnqY7zVMd5qpM4T6FJ0MQS2RRocjSZRDY1mgpNKZFN2PPar+e1X6+QzRoa i2bWS3QKjUOzkMgmRZOgiSWyKdDkaDKJbGo0FZqlRDbhQKeGOxLbJDqFxqJR2CbRKTQOzUIimxRN giaWyKZAk6PJJLKp0VRolhLZhCOdGu9IbJPoFBqLZibx/k2ExqFR2CYd8sPm475pDvkKv9PtkL+K xqJR+J1uh/wIjUMzl8gmRZOgUfidbqv+zeaL/q3mC1zufzar/i6aBs2uRDYH2y/6R9ovcKmQzTqa E2jWW4VsTqE5i+aURDZbaC6j2RLI5riftS/6eXvcV9gmHfdLNBWaWiKbafeib7rjvsI26bg/Q2PR KGyTOE91nKc6zlOdxHkKTYImlsimQJOjySSyqdFUaEqJbMKe1349r/16hWzW0Fg0s16iU2gcmoVE NimaBE0skU2BJkeTSWRTo6nQLCWyCQc6NdCpQaJTaCwahW0SnULj0CwksknRJGhiiWwKNDmaTCKb Gk2FZimRTTjSqZFOjRKdQmPRzCTev4nQODQK26QNP2ye8U2z4Stskzb8VTQWjcI2acOP0Dg0c4ls UjQJmlgim200OZpMIpsaTYWmlMgmbOlUS6daiU6hsWgUfqcbnULj0MwlsknRJGgUfqfbeT9r3/Dz 9rzENum8X6Kp0NQS2Uy7N3zTnZfYJp33Z2gsGoVtEp3q6FS34Stsk+gUmgRNLJFNgSZHk0lkU6Op 0JQS2YQ956l+w1fYJm34a2gsGoVtEp1C49AsJLJJ0SRoYolsCjQ5mkwimxpNhWYpkU040Klhw1fY JtEpNBaNwjaJTqFxaBYS2aRoEjSxRDYFmhxNJpFNjaZCs5TIJhzp1LjhK2yT6BQai2Ym8f5NhMah UdgmJX7Y/LVvmkRim5T4q2gsGoVtUuJHaByauUQ2KZoETSyRzTaaHE0mkU2NpkJTSmQTtnSqTfbh 3aQHdgqNRTNrJTqFxqGZS2SToknQxALZXPOz9od+3l6T2CZd80s0FZpaIptp90PfdNcktknX/Bka i0Zhm3TNn6NxaCKJbGI0CZpUIpsMTY6mkMimRFOhqSWymfZ0qqdTvUSn0Fg0Cr/T7Zq/QOPQRBLZ xGgSNKlENhmaHE0hkc0STYWmlshmOtCpgU4NEp1CY9GsDRKdQuPQRBLZxGgSNKlENhmaHE0hkc0S TYWmlshmOtKpkU6NEp1CY9GsjRKdQuPQRALZXPfDJvdNc11im3TdX0Vj0Shsk677ERqHZi6RTYom QRNLZLONJkeTSWRTo6nQlBLZhC2daq9LbJPoFBqLRmGbRKfQODRziWxSNAkahW3SXT9rf+7n7V2J bdJdv0RToaklspl2P/dNd1dim3TXn6GxaBS2ST/xT3U/9s92P+Jy/7P5ib+F5jKaLYlsbqK5heam RDa7aBo0uxLZHOx/7B/pf8SlQjbraE6gWe8lOoXmLJpTEtlsobmMZksim5tobqG5KZHNLpoGzW4v cZ4aOE8Nd/fh3aQHnqfQWDRrg0I2CzQOTSSRTYwmQZNKZJOhydEUEtks0VRoaolspiOdGu/uw7tJ D+wUGotmbZToFBqHJhLI5p4fNv/tm+aexDbpnr+KxqJR2Cbd8yM0Ds1cIpsUTYImlshmG02OJpPI pkZToSklsglbOtXek9gm0Sk0Fo3CNolOoXFo5hLZpGgSNArbpMMmaz9n8vawUdgmHTYlmgpNLZHN tPucMd1ho7BNOmxmaCwahW0SneroVHdP4nOT6BSaBE0skU2BJkeTSWRTo6nQlBLZhD3nqf6exOcm 3fPX0Fg0Cp+bRKfQODQLiWxSNAmaWCKbAk2OJpPIpkZToVH43CTOUwPnqYHz1CBxnkJj0Shskw6b BRqHJpLIJkaToEklssnQ5GgKiWyWaCo0tUQ205FOjXRqlOgUGotGYZtEp9A4NArbpCdN2HzWmOZJ o7BNetKsorFoFLZJT5oIjUMzl8gmRZOgiSWy2UaTo8kksqnRVGhKiWzClk61dKqV6BQai0Zhm0Sn 0Dg0c4lsUjQJGoVt0qbJ2m+ZvN00CtukTVOiqdDUEtlMu28Z020ahW3SppmhsWgUtkl0qqNTHZ3q JDqFJkETS2RToMnRZBLZ1GgqNKVENmHPearnPNUrZLOGxqJR+NwkOoXGoVlIZJOiSdDEEtkUaHI0 mUQ2NZoKjcLnJnGeGjhPDZtGYZvEeQqNRaOwTdo0CzQOTSSRTYwmQZNKZJOhydEUEtks0VRoaols piOdGjeNwjaJTqGxaBS2SXQKjUOjsE2KTNicNKaJjMI2KTKraCwahW1SZCI0Ds1cIpsUTYImlshm G02OJpPIpkZToSklsglbOtVGRmGbRKfQWDQK2yQ6hcahmUtkk6JJ0Chsk75rbrZnzK32dS73P5vv ml00DZpdiWwOdmfMke51LhWyWUdzAs16J9Gpjk51kVHYJtEpNAmaWCKbAk2OJpPIpkZToSklsgl7 zlN9ZBS2SZFZQ2PRKGyT6BQah2YhkU2KJkETS2RToMnRZBLZ1GgqNArbpAtmOlwyZrggsU26YGZo LBqFbdIFs0Dj0EQS2cRoEjSpRDYZmhxNIZHNEk2FppbIZjrSqfGCxDaJTqGxaBS2SXQKjUOjsE26 aMImNaa5KLFNumhW0Vg0CtukiyZC49DMJbJJ0SRoYolsttHkaDKJbGo0FZpSIpuwpVPtRYltEp1C Y9EobJPoFBqHZi6RTYomQRNLZLONJkeTSWRTo6nQlBLZhB2d6uhUp5DNGhqLZtZJdAqNQzOXyCZF k6CJJbIp0ORoMolsajQVmlIim7CnU/1FiW0SnUJj0Shsk+gUGodmIZFNiiZBE0tkU6DJ0WQS2dRo KjQK26QbZjr8lzHDDYlt0g0zQ2PRKGyTbpgFGocmksgmRpOgSSWyydDkaAqJbJZoKjS1RDbTkU6N NyS2SXQKjUWjsE2iU2gcGoVt0o4Jm/80ptmR2CbtmFU0Fo3CNmnHRGgcmrlENimaBE0skc02mhxN JpFNjaZCU0pkE7Z0qt2R2CbRKTQWjcI2iU6hcWjmEtmkaBI0sUQ222hyNJlENjWaCk0pkU3Y0alu Zx/eTXpQNmtoLJpZJ9EpNA7NXCKbFE2CJpbIpkCTo8kksqnRVGhKiWzCnk71OxLbJDqFxqJR2CbR KTQOzUIimxRNgiaWyKZAk6PJJLKp0VRoFLZJrZkOjwZmaCW2Sa2ZobFoFLZJrVmgcWgiiWxiNAma VCKbDE2OppDIZommQlNLZDMd6dTYSmyT6BQai0Zhm0Sn0Dg0CtukSRA27xvTTAKFbdIkWEVj0Shs kyZBhMahmUtkk6JJ0MQS2WyjydFkEtnUaCo0pUQ2YUunWjrVSnQKjUWjsE2iU2gcmrlENimaBE0s kc02mhxNJpFNjaZCU0pkE3Z0qqNTnUI2a2gsGoVtEp1C49DMJbJJ0SRoYolsCjQ5mkwimxpNhaaU yCbs6VRPp3qJTqGxaBS2SXQKjUOzkMgmRZOgiSWyKdDkaDKJbGo0FRqFbdJTwXR4OjDDU4HCNump YIbGolHYJj0VLNA4NJFENjGaBE0qkU2GJkdTSGSzRFOhqSWymY50aqRTo0Sn0Fg0CtskOoXGoVHY Jh0NwubzgWmOBgrbpKPBKhqLRmGbdDSI0Dg0c4lsUjQJmlgim200OZpMIpsaTYWmlMgmbOlUezRQ 2CbRKTQWjcI2iU6hcWjmEtmkaBI0sUQ222hyNJlENjWaCk0pkU3Y0anuaKCwTToarKGxaBS2SXQK jUMzl8gmRZOgiSWyKdDkaDKJbGo0FZpSIpuwp1P90UBhm0Sn0Fg0CtskOoXGoVlIZJOiSdDEEtkU aHI0mUQ2NZoKjcI2aSM4ODwfHBm+wuX+Z7MRrKM5gWZ9UMjmFJqzaE5JZLOF5jKaLYlsbqK5heam RDa7aBo0uxLZHBzp1EinRolOoTmBZv0haFZ+Qzb/fz/ZcIyPrDwwBX/F49KffHDz4xx/4E1WHlt5 IXAk8vLe8dze4faO54Ovc7jxg9sf9vPmf/i8BXvP2yMr35m8Pr4avL5v77n5/0fyncnP3t//TF4N fvb+fkh+PZMPJA/n2Vn5DZk83Ebdb/L9Rp3+VaPWVk4G0XiS1nxz7/j2+HLw5+Mi+Ivxqxwng78c /4jLE8Gr45eDV8bjwZ+OXwq+OT7Dfb8UfI3rJ/n6Bre/QPOe5/7PjZscc/7763ztZW77RvDSGAWb owu+ytf+mMd+idtf5PE3uO/G+BLHBscLHM/x9RPc/iz3+wrf8+VgzuE4Iq7/Cbf9GS3/Nvd/hZ/1 Ct//yvg1ri/4usPm9v4s0UN/Zu//nfVwn9n7/0fdf2b/+VfP7DMr35t8svm7veN3m2TyePPG5FPN 308ONf/A8U8c3+f4F45/mxxs3uS2N7nPRe775uR3OD7J13+b2w9wvwPNhckn+N6P8zgfa/528tEm nnyk+ZtJyBFw+Fyf8nMe4WdMmnSy0vwjx/c5/pXrb04ebf6d2y9xv0uTx7j+GI9tuM3jPlPu+yjf M+V7PR7f5wg4HuN6yNd/i9s/guOjfM/HeJxP8BgHmrfwvYX1LcyX8F+cfJrbn+B+T/Bn+yzf9wSP 8QTWT+M72Hxv8nvNB3n8MreVDxvxy+uPfPj8+XvP3+/v/R3y6F7S/wtRmu5FTv0FAP== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0022_image017.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhgwEyAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAACC ATIBgQAAAAAAAP///wECAwL/lI+py+0Po5y02ouz3rz7D4biSJbmiabqyrbuC8fyTNf2jef6zgT+ DwwKh8Si8YhMKpfMpvMJjUqn1Kr16ksEYlse19v1indhQ7l1HqvSN7b6DTuHgWs4y13D2/co+SHY M/Qn+KeFQHRIKFA0KNToaKa4qIhYaHk5CZkJ+EgXqWmmRSk5RzoKWrnp+cmp+pOomeqaxdoqu2hY O+JHogfh+wCcG3qooOd77CA8wdtwnOwMHaisLL1A20NdrCvSzBdS5j2zLEH+Dceb9do9fP4R3j4O Yu4+lt46H1+/AZ9oQx9sn0Bil1bNEzfwAjZcg9q8A5gQB8ISECNqI4hR3juL/+4mcuRTMUXIjxE9 kkQH5qTKCCYXGsPjUlS0ajNrXntm8+bFbM4C9tz5Mqcxnz95Fr1GFClQfUwxKcUkK1XUU/ikWo11 tSpVg/2gbl1XC+wtOV8XlkJVNlxaWLZMYUX7Nq5Bhk5XpnQ6Elw+u2pM8nWIF7CHvH/7NN1V2ELX SP/2JiZzGBxhkmb9NR48+THibX38Pl5Ml0bmjJov16UYufRp06pPenZMunVqGaNl5/A2VS7Y0KFz t10bVqzb31U5Y8zKFThe5XR9r3IuHC7x5LrNpj2Lj3ftCtuNK9ZAD3Qaa0/LD93o/fy0o0HZy3Rv 2ahO+IHj2Db+ekXt7vcVzv9G3V9v/9WBXoBfpAfgaqWJx59/BRr4Qn4duNTgHpU1xBo/FUL4y4Ac UGigeIJNyKF9CPbi4Wd3lbiPhCSeeN+G/LAoUEs4rVfffDi2t+N7Pcano3nqCRkZMjT9CGNsSRpJ n4K8DQlllElqxxZ10+2G23DPIRdddWp5iV87XFoHpndjdgUdNmliZ+WW0rmplZds7kalif2JmOGM NN6Rol4B4pkHbHsaNqUJMqK0XJ7gDapfn4K2FhOGnWBBqSSTiFJppppueoSjD9oGqGiPMopioQly eGgGqZLqp5MnrGrhiqwiqmCkkvIon49B6oorr77e2mtq5Enp6o27/srYUlP/DltkUsQ6yaRXb3bJ lZhaUgtnttqSeeWFGX3ZLbjbinNmldh6u+ak3I5rrprXustcnSJ5SpmZI2o0K1CwumhXqPiKmq++ jZr6F4OK2hmweauCeOditkb4D6wr8aunq4n5S9u9CduLMIQSc7dxX/S+WOLHFJgcMsgwPnwpkisb GyyQyCZ7rMzCKrukszOz/DDLTy6LM7Q6A93kt+2Ke26o6QYHb7hzrpt0mPgxV66ZVFOtLtJNtxl1 1l5/PcvWdFK8KMF8YXxgoCnHs+/IHKHdccZrczyv2f3SDbDac3+7KZQo62Cw3nnvTbaqblsEN8QH b1z4YDT+zdLekNndKqqy/0qeIUwwz6xkyy57XjPNoYMec7FBd/7zkpofOTrq0RKZ8+mvP2s02GHb nuW0Yncd9tNa+y61tOEebbt2V5dJ0NK3g61881XnSKjFhk+PQXh4uz600JkzC2yR3ItOO88CEyh9 wdfLLfjcjTtY/tnnpz24+oefPH9CgcePPuH1l7N/je/HDT/5UW5Ud7oc5uIHOcgZUGMHzF/qSLez 2dlsguCLIOtKV0Gi5Sp82cugBz/os+9hT3YdfOAI2fM74jENb85TDiNWyDxSsPB4w7Ma8paHw+a8 C3gx1B0PYdi74q3vOwN0zf8Ul761DVFl7bsb9P4FRSX2r0NFrNcTHYjFlP8tkX5V/Mj9ohhAKXaR O1sciLcgGMY4KLAkU3TWGllTRsctLmFxZJ/lePDGBl6xexRE48suyLmeba6PfkSdCSWowdHxbHUk bGTREGm6oskrh8+bZAuR90JKYg16layk8W7YSdzRkF09jJPThNjG8W2memVjxvlEeMjToSGQgPSe 7EQoPhQOrIkTO6IL8mjIfNWRirxUSeJ+ycCQDdONIbLaHGeJuWUKBZi0cRg1gekzYaZSl3+SJZ+e qc0xVkyPryJn+hJozuils2MpLGW1pubDrP0wk5d8pyZBGS+y4BOUotxnDYFYT1LusVTFdCX17Bi5 RNIulpL8Jh+hhcuI3tL/kcH8FDWzOLkkanGbubpoGjOKv42Kk4sFtaLIwDkrafbIoxHSJ6demile wXSmNM1O20ZqP5yaL5ocTWjJ1jkolfKPRSwF6jlVCFAZwtOU45pnJzfJt3gCEYcBhdru2nnPf1JV Kv085VQnWU6dwm6orCSpCYmJ0AaSS6KP7KnfmllSo7Ijc7us6E7tKtewHpV8eHViX/VKzmzy9awP cSs6QMOForbomob1pYoAJ1TNMFasRMxrK/9q2Q81tqGZVWVn1dmbHeLOWlINYlJHybutlrZ5bEPt VZFaVazSs6tNRSVlmUnQtJq1sro1KGeJ2tqJtvW2nv3sZQW417rF1bjE/1XNZJfLRMwaEaSETSk2 N0s6xd6hqAwj3HOlW1bo2g+xasSuXxH40QU1N7zgZe5u0dlN8VrUvYAN7WqVKjyuXdWprv2hgJCq 2uJw051Y6i9tU3umORx4v7aVL1nbi1beSji6Ey7hQve0VuFu0LGC0m56V+lT/dF1sB4G4KkiLMbk grbEfJWYOVjcl+uut1cwfhV5e/HiaH63unIs2Y1zi2KRqpi+/6UukTt8ZKgsMMma/aNQCCnIWj4U hIOccnZV92QolzDKWf5gIRfZZSpvecyejS1sRcs0/vpToEWeaii/6lI2P3XNCL5OQdBs2hyClck8 7nGIkYtePs94rIQO2P+OR1PjuZJMuX8e52dkLE8lyAwJvaspFHRlaSdgmlMyzTRNO+3psWGGwgTc BaIdioabapSdnwqyn0F73F4O1mP9sx6Q6xtrY9b1jrWCmWdgydbWdfeswcbgl6s85WFDUnufI7Ys gf1s0jJVLABusyfjnDT/qlnAC9a2gd3c7QYD9Mz3LfcN91zX/XRknTkuTGTRqu5zJNoO7b4rhPuk llYbSt/TZfQv5n3QPgfD1oXsran57UVV//ti+xNs8BZtY4S/TeGuNum9KVchF0v8bbumtVgJrmh/ 57rfsIZrry/462g/kpErVagtm81lmLP8OBYu9kKhLUmc57fAsai2Nc//jW0Gg/u0qMDUtF9LYKtq O9xwNvpzyH10b8tJ3AKvXLy/AfA31Nt9g07K1UHC7t3K2sFV/7HBQezosZf8t2ZsOMgd3miKnLpg FB+4u90e3a9H/NXGrLs32fhxUm+8cmnX9axNfnFeZhzVhSc5rhH/QDAPlNkbhmiYrxzJyos5zJL/ nLJrruFC61z0Kt+50k2PG6jPJei+WzrRue3mb2dVzkzXM+tRX/vY+hfdEkcssqt+uN+PZPgYtbzm hfT7Q9q8WV3fid7pHXbfcp3sby812kdu8cez3X86hftx5i4S8PfL75dHHN4D3viD89015D8+xwOP 0sFL/7zaLyD8V31r//QbXuSQnzmuUp5zoRdIqUdRGNR5xnaAMuVyzAdzZDZcD9hRAMZVSSdblnJ7 aTZar2dPsUdnSKeB6IJnY+F01EaBqsdzXmV75sV40JdOW0d/WRd3Whd9D/aC8nd4J4Z9XtRxf5dT 3bdjhiJ+E9N+rWN+PqgYfbMGQdh3O9hs70d9yZR/7EV/+zZ/agd821d/16d/jkeFwOVkO/Ji/kdL nLc5ZsdQnid8yRdzNFd6hJZLEOiGDhiBuTdwHahgNoSCUUGBDGJu+iV1WvWHbPZJtmCCp8eBeWiB KnhixJeDPZODFcd/WOZ++qCGUmZ8x/duuwaDV8gqLmiFm8g2K7iF2f/XhTGYfaAIJPuihOzHhE0I eGRXfJ1hgzUyhJzjhIkXUq1YYVyIgzzYds1nb+mngzdof7AoG6vIi1FYjN/XhycoZ3b2gR6oQ1O3 HM1oiLOHYLmXiLBViBVojdLIe9wnhbsIQo9oYVmofNs3HpboOssXO8bIUy0odvuHi3qEing0j453 j/V4doKXjFpYfq/Ijx+2dxC3hMRobLfIibehgshIir2IhfWQiSQmiuNIj+h4jIoIZBq3fv8YcrzG jFGnjavVWtRISXuIh344kiI5dNgIjhdoX4Aoe/S0jfCIf6ZIgztjjr4IkJd4YelRiZcXlHM4jFg3 gzipkPtYaEsmOZ7/eJFKOZH3N31r54o9aJPIBIxe5o8Jp4u2WIRXCU1guZRVeJGlCIlJaZUDeY5k 6ZGE53FiqYkV2Y9fSZUxIm0qeYjP2HMpmW0ZGGCw55KBWGf8lJepNYhPV4KJGY27F5VYaZH5WI6j iJToKHxiN5TCxo5nqJY3KQZKyZBseYptyZQilpNPmZYLOZegWZR1mZACiZr3ooqzuFhdSUho+Ytw WZuQOXG0mZWBJo6bGZHBKW+86VwaiXYcaZBlCZHLGJN4GZh0yJdCd5KLGZ3eyJLP2ZJmVp24V5jS mGBU55lukxeVKZlnSZns6ImXiYCZSR7h2ZtueUBOqY+iiY/yqJq2//mbr7mV5LiaGJmbpwmV5uV9 a2lGtfifAIqgSSibEmmg7+lAAfqeDtmfZvmWwBmWFjqWpUmfHcmcWPWXzjiYhLidGJid0BhV16md 0cgZKXqiiKmiivmhhtmYF8qfOOmI5UmgytiO6hg+7uiTHOSgCNmZR2merumeuKmjgFakCZqfyNlk 5hSGdxekx6kfiBZqV4qlaDGCWVopxvmRscmhpDmZRsqkkZiayoRhXkp4TiqMSlqhkZeG6OlrLyds BAiHQHqnDIiJoLdyQimGfZSAtZk7KBqCIViSiFh01MlJM1mo39iNh6qXiCqpzkkuGwo4akqXoBKK DymDUzqhkLKpJ/+loaKqlSkVInzIpRbIeqlaBQdKKkeqEJUaY7o5pJ6qkBnJYfGHoRQan2lqq/v5 k0bWq49zGIFqZWsYjs4WgMvqfqN3c234rAVoQWQop9VaZo1KqImagiuKrf+USZbEqI4Ko9e2pYIp nd/ZnXc4eS2Cqe81rB3arhtJK7sqpfD6q7QaXKRacOHUMLkqGls3oPATsMTar/46Doyor7DqmHZp sGDEqU3pq0ianA8rpiAJp9aKhqFKesy6sc0KrTfDsdGap8kWp9RqsgOGnSmrh+m6kt66stypst0K ot6ZT+XasgJFkyTZsJk1o1aIq+tKXz2rnMW5swSZo1A0sD/1J8f/1AYhkbQL+7QWCymyag8AW2NC a6k1eK/3qbG1CqURS69h+lad+rUEe3Inq0hUqzB6GofSyrZ4uqduu4A/CmV/SrIYS5QxK67JU5N8 +43btoHNGaKRSriDK6M2y7IkGqM0u5dAu27DubW3ybCOC7lGGbn5+bO1o6+4lbDvOrma25lWO6/7 aqpLW7QtZVZRy3gKS6OZm6wS4a5kG1hgq59C2rWj6ab956e723LvCLoi27HBC7zDC7IeK7ebl7Fo a4Cz46FzBojgebO8862DWriMK5N2CJ7hiknO66K/S28wYXd9Kr5r+7bl67ttO77B2pNjirKv+7iW K7FTCKqni0R//3a18Sq5mkq/qZa69+uqQXWq+7td/StvrCuXj3a79UmDm2jAxDmV7qurECxoSeq6 /rvAE2xi9pq4esttIsitf4uSG8yi3Gu4L4m4HOyyM5nAWge+7OnCQimH6Bu3I0u38au+GZq3Egx2 GJyL6iXAuMvDZipZK3yp88fAngsqTBsxBIx1DWy7dEfEn3nBslu7zFnB+Mu+3qvAYfvAb2q3yJu2 Uay2xDu3v+usxSu8aCzDy4u3bKy8YyvCMguCoCAvuhfCoEvC1lu95sqYJ2zHiqqyURy6OeejOzqJ Z1zGymq8nme0PAnHWpy/LFi2GszFrTu6aqW7l7ukkFzENjy0C/+ixHkgurKrupA3tYIMheTbufFI yVXMmqg8YhBrtpXcprA8ZCmmtDC7uI96Z3TcDH+8uEqjvZQqx9a5en5czIBbwurKyVVLyAJIpxeG yJmXxol8o/Vbo5zbzBL5vrMbX5QLvzs8yZ+rww6bw5fjxEKMwOB8sExMqqX8zUlsy+0sffDMaPZM zse4Dlj7mOe8xa6svzECCPx8ptsMm0jsXKUArnu7y71sU8MAzOiKx8M8s3wcvWuFvYor0emiDq9A 0B/SwjAs0owcw3C7xibdo8I5saPqUx2tDlgMn1SMyf3q0TBtfV7rzQK9zza9oKmcu/Fcaawq1ENN 1EVt1EdNx0EFrNSPUQAAOw== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0055.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Performance
 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0055_image018.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC+1dDXAcVR1/73KX5C4nRjhjSnF6ZtqQUsDLJ7EEaJKmaZqPS2iFioiJMTQd0zRN r03aSElEYwcZYRjUWtSgRtSKttTRiqJltGV0MkxEZ/yATsNYEQp2kkIbkdSJv7f79m7vZffu3e6m 5aPX+d3tx/3f+73f+//23u6+Tf/6zNN7ifIKuUZpp7J0ah0lXkLS/ldOiJ/kZLNt7D0//b0kL3hr 46pqqnyPkg2H2GcaewsScuAqYA3bTsh9JPYKuQgZpbH1A2WERPIRUtO4in2bxWSgkEzsey/QBVSy L3oANyErEDtFhNc/3GTwh2mkEd9cRxrwXgUKq8hGsoFsI72kg3/tfYS1QS0bbSKsMa0cWcK6uP9S Hsuo3wz0UHV5enZ2Vk/lgVjb8laj5i6yHe8RcGknbSTV19k+Rc7nspS1z546rizn4nPsxJadbFsn ll8A2Gts5X/79vzOR5750tm+1Rluct0DZ/sCvKRHsJ0c8JJ+6laQ/f2fU0Y1F+WxdcKX4193vP/H 0L3zGkJ+7VM/y7d4d7HPwiwKLqwIN/nnv48r5V9x+jiZ2v1m3+4rzvbNKrqc7Xulx7uLfcetpgX5 rjvnQIDzGcygSqcuBv9WlCnGf3q70lVkiU57bfk9hAwWmPTJ6+exT5jObPm7bt/jrE0MmZPW+oRp MfOh/573PskP5Fz1TuyT/IBv2du1T+CTg070ycXlt8dyIVCCBZfyI1onmff3AyPAfuAwMA5MAJMo E2mVDQSBELACaAZagR5gELgfGAH2A4eBcWACmASQltmzkwifQPg4wg8jfD/CRxB+P8IHEd6D8FaE NyN8BcJDCA8iPJuFE4QThBOEE4QThBOEE4QThBOEE4QThBOEE4QThBOEM/Zohdqacd66/by1rNWD bCzB9WOalfDf9XpgBd/n4V6iHNk65WbvViGuG5W5JDZuGLyS+V8b57C42VmX3nts+zL+He1ViG98 j8a+k4FytGWiW6YmdblN6mLbbwGuBup4W2dQyHaq7qvDUh22zlCG7dSs/k/irZzrVc7rKtfxp7H6 47JPK8MjtP9Wrr/IyWPAia2b8SrQlV+ga7NeIx2HwYW8j1m92xPorbXXnaS94muJAQfW3jJer9ZO Nm4N6NZZOiXre6ttScZJ6wONk7Yuw4k6oLUMvy00nt8Wm5rhRIIMJ+CUG1t2LeDLzFs5se04j1DB ytqBshrwuZtqufH7SjL4q0ptH9vWIOTKLH81cy9kSGgZsallRNAy4oCWmxzScj0vi+m100BLbZ8r iZZFUDNDUs8Rm3qOCHqOOKDnHgf13MM1+56Bnnsk9SxJQc8xm3qOCXqOOaDnEQf1PMI1e9ZAzyOS epaloOe0TT2nBT2nHdBz0kE9J7lm5wz0nJTUszwFPfOoPT1ZvF5Pbd2OnrnUGT1beVlMs6vpXD21 fWkJ9FynnGeEor9Hsro22tS1UdC10QFdqx3UtZprd6uBrtXSuhalrGu/TV37BV37HdC120Fdu7l2 nzPQtVta15KUdR21qeuooOuoA7rudVDXvVy7nxjoulda17KUdR23qeu4oOu4A7oedVDXo1y7vxvo elRa1/KUdZ2xqeuMoOuMA7qedlDX01w7j2uurqcldS1K8Lt1Ps5TG0n8ub22bkfjYmDAIY2L+fF2 JT7vEjTW9iXP3WtJKdfWLZm7TNcam7rWCLrW2NT1RmCNQ+f/rKwbuK4Ngq7aPpr0nFVey4BNLQOC lgEHcnSBgznq51ouNMhRv9RxwFqOniD2dD1B4nU9Qezn6MvEuRx9nqi6vkLm5ujzJHmOFqeg5cEU tDSeK6C+onMFfOpcASPdDwq6HyT28/kQcS6f93HdnyBz83kfSZ7PxRbzechmPg8Jug45kM/DDubz Dq7rboN83iGRzyUpaNlsU8tmQctmB3L0ow7m6Gqu5a0GObpaIkdLLOZo0KauQUHXoAM5usTBHM3l ul5pkKO5EjmqaZrsnqWm5Tls6KNz1/X3xhLe23JR9f5yCve2WM5c4VLrvIHlFOoLU/X+az25Cf1y EwkDLcBa4BZgPXAbcDtwB9CKLGsnbJbHarIB2Ah8BtgE9AC9QATYDrX6gZ3AXcDdwBBwD/AF4IvA vcB95BPky8ADwIPAQ8DXgK8DDwPfBEaAb6P2UeBR4AfAPuAx4CfA48BPgZ8Bh1DbE8CTwG+Ap4Df AkeAp4E/AGPAM6h1HHgW+DPwF+BvwHPAMeA48AJwgvyIvAi8BJwEXgVOAVPAa8AZYBp4g/yLvAn8 D5gFKP0XSQM8QAbgA/zAJfQymg1cCgSADwALgIXAB4FFQB6wmBbTfKAAuAq4GvgwUAiwf6XAdcBH 6DJaAdwIrACqgJXAKqAOqAcagTDeW4C1wC3AeuA24HbgDoDlBpuPdztQT9XzznU8P9n+VqAd6KA3 0w3ARuAzwCagB+gFIsB2oB/YidruAu4GhoB7gC8AXwTuBe4Dvgw8gO8+CHwF+BrwdeBh4JvACPAd YBR4FN/9AbAPeAzYDzwO/BT4GXAIeAJ4Et/9DfAU8FvgCPA08AdgDHgGGAeeBf4M/AX4G/AccAw4 DrwAnABeBF4CTgKvAqeAKeA14AwwDbwBzADJ7nun4udZC35ezv38KtH8XAE/V8DPFfBzBfxcAT9X wM8V8HMF/FwBt1Sk4Oda+LkWfq6Fe2rh51r4uRZ+roWfa+HnWvi5Fn4OwzVh+DkMP4fh5zD8HIaf w/BzGH4Ow89h+LkFfm6Bn1vg5xb4uQV+boGfW+DnFvi5BX5ugZ/b4Oc2+LkNfm6Dn9vg5zb4uQ1+ boOf2+DnNvi5HX5uh5/b4ed2+Lkdfm6Hn9vh53b4uR1+boefB+DnAfh5AH4egF4D8PMA/DwAPw/A zwPw8wD8PAw/D8PPw+iPYfh5GH4ehp+H4edh+HkYfh6Gn79BsoFLgQDwAWABsBD4ILAIyAMW0z+S fKAAuAq4GvgwUAgUA6XAdcBH6HOkArgRWAFUASuBVUAdUA80AmF6krQAa4FbgPXAbcDtwB2A3s8v 8etIa6N+PolcOUnagQ7qg5998LMPfvbBzz742Qc/++BnH/zsg5998HMQfg7Cz0H4OQg/B+HnIPwc hJ+D8HMQfg7Cz2Xwcxn8XAY/l8HPZfBzGfxcBj+Xwc9l8HMZ/FwFP1fBz1XwcxX8XAU/V8HPVfBz FfxcBT9Xwc9N8HMT/NwEPzfBz03wcxP83AQ/N8HPTfBzE/wchp/D8HMYfg7Dz2H4OQw/h+HnMPwc hp/D8DNUA14FTgFTwGvAGWAaeAOYAcz8rM2DZL/tU+dhHmRnjT96nsWWJzq9uwqz6K8O+I/R5lWZ 5JFfTkTnQfb3eQ3nQQ5iu34eZPCP/l1v3L+CtKI8bR5ka41fqH/lOm0e5MueufMgO2vUeZCjT0wo 5UcOTySdB1m5NKcwwPn8vledB/lPxLN5kGK8OA/So5uDms7Ojc/jHNQSvrzssHcX027oqY/tDDVN kMzGmPYFrxvPQQ29Pn9zUP0NE0r5/eHk2hcW5PQEOB9tDmpDg6q9GC9qn24yLzj9AvaJNi+4sMC3 hbVJwdusTx4tyDnnRJ8kWrY7h1JfltG8PbO6zc7bXgS+JXFdNJ2fd8mcwz3Ey70X5X4c+I5wPVLb 509wDvdppaxe5GYb8jRIusk2jIE+hXztJZfwczuPrv3p/DmTLN3zJH7hfFrURvMKiz9zHr3yVP2E srwi3U2KXlaPYb8Y9pLH1sS80jmdbuiVHmzXe+XOiSwW74hXDtRNKOVLHr96A5yP5pWpOjmvuE2O X26Tea9my2a57knRl3bKNPPVIjZPx6av9Brpc1U7rn8+bT5y9cIez9nrQh7PjXLV6NofY389ilzM +1l/fMvi25Jd++sim3GOFzuO+Uny47tZvi3lfJLlm8dCvk3Na75d2PE0e13I8bRMvrH5fpezPqbq 8wjXC/l2Od+WbL4fm5/m5c9fLrE4/38R53HxuPbOO64d4yqxZ4IX834Wj2tsW26CPPuswlE/Wgvi KHcn3rei5zpIDz6DWNqM9w14Z6O7dN3xTxy7pevySRvj6b+fpdvnF/YZ/W4b5bqdZ4OWIzifjR+p OrZ3K9vqgXxsY6hP+qxMhsVngzJMng0SOXkMOC1PwKtAV34qxwat3vP9bNAZXq92P0tbt3M/aznX zuo9Nk17Pad6m5zO2ODU7VHnCvzJZzxXgM0TUP5mwJq3rqbLBU7L36H9/FbVKd+mThc9+vbUyU4+ OfUct1nb2D22UiL3/K/s9atjvFz223cSn9cJcxC0fcnGQS0Y43RhzNOGUVA3Rq3dGPEESR+WIqQT S9XYu42Pi3rx6eN9kKabh6vNCRG3efk2t8E2MVZ2XglrV43DWu7m5Zbz64y1Blqyfb4EWrZyLRsE LX26tiXSK9k1Vy2fz/JnUcX1+Z4DwsbaN6Gua/F5G1XntLJyKmmIrAes+p49CxRK0J/zfS9ca9dN vF3V0XZVol2VttpVKdkufc4my4NzfKCtXWsX1+c7Ptm1/0v4c2qVcd9/d/+dilx+7SCNZiqaZimJ qlxRc43SkMutpC8lV7B77q7pWeHKQt7qjq7tHZGN7WbXE1j8Nfi2Gp+tHE8IeVKphbpceWt3bI10 bDK/6qDGU/U6WrSc1y3zmJ/yLn6afT6Mz/8gveolrzK9u92oXalg6pW4XMovQamyxp6rZd7JcYe2 hbZpZ52lykWpN12e6HUOVfc05bvnXNq1F3WrW9m60P0oLVL8p8aKOe3Rx0a3pgs9a1xDhvJbsTjN PedvqWhla6Xp647xzowyZcfqJWkew7/JIpalejmTf2YIDGlc6zKUY1vAxUpKxCkWEePEIi9D5Axl JS9SDo1szYgPqz95jcY1BNHqWA1szTiO2K5pC9XXtMU0TqxhmK6a8022x8vUU7KRulhesH3XUnVk XeTbgSjWG6HyDFPmEWnmkTjmEWnmm1Ji7uPMNykjr6JQBjHnPiLNfSSO+4g09z2WuO9R/2JUQu5j 0tzH4riPSXM/Yon7EYV7WULu09Lcp+O4T0tzn7TEfVLhXp6Qex6V5Z5H9dzzqCz3XJoKdz/nnkvZ 70thiP2Fi0T8G6X5N8bxb5TmX22Jf7XKvygZ/35p/v1x/Pul+Xdb4t+t8i9Jxn9Umv9oHP9Raf57 LfHfq/IvS8Z/XJr/eBz/cWn+Ry3xP6ryL0/Gf0aa/0wc/xlp/qct8T+t8C8S/Gt/xADHE/3op1Hy GDpAi0nq7ehGlNIP15YqbTDrhxpp/jVx/Guk+a+hN1oY8dyAKDbiKTId8QSkmQfimAekmS+wpLyf K1+URPkTRJb/CaLnz9bk+L9MrCj/PFGVLzZV/qA084NxzA9KMz9ErCi/j6jKFydRfkia/1Ac/yFp /sOWlN/BlS8xVb5ZmnlzHPNmaeYftaT8aq58SRLlg0T+TK4x7kxOlv8SS8rncuVLufKxawZ99BzV WLMzZe1M12VyputmM5TSBi8+Rfkuf4ryq8pVmLqLT0xKPDFpdlVK77XZhF67+ITju/kJx5jXLj7N mOxpRtFr6l2cKYt3cbw8/hWL8T7hrpLPofKMP+feCdDvnYoeWcRocYzxLfqiyWjEl3CssUAZa9yL aDZqWte7sa0r2L1t06c6ei8hrGw/SkknWfh3KZb90VFUjI+qzhmL6mQKamdGe9H4U9TBm0Rl2ThR z/V0Ucp6XhO9s8pyhc3btqZGbAy7nmYpY9iuzRuUvvCbZIfI/nq61IS9V4L9lG326jXT6+nlyu9k YchL4vvRnPnSC6z7MoX5UujO5vRwKwQ33xncGuno2RqMbA5u2NzWla70heYItpauuETdnqWs+fma n3hMei3x3bfM6N03cbay3N23rDl334xmPRuPc7KS9FI+Yq3db2M1x86h2JrcOVS9QY3GNdTH1VAv UYNcufPPnKmqr2H5W0Cb+W/DW0X9d37epDrfQ2RUipGsWcmJxhjqMbUA0eyY2hLs6mjr7d7YvSHY tzHSGazu2oZja+9WdnRM41e3M6JLXpLGr914dXsTXDu2yDFH4ViOaJ/CsYFz9Cl1iaz047YGcpZq mstcjVH+n580V3Q+o1w2hEij5XPSWI3qTEO5GisT1KjPE3EEe25Wfqv2vxv9H2WHDTsQcgAA ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0055_image019.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhgAEOAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAEAAQB+ AQ0BgQAAAAAAAP8AAP///wL/nI+py+0Po5y02ouz3rz7D4biSJbmiabqyrbuC8fyTNf2jef6zvf+ DwwKh8Si8YhMKpfMpvMJjUqn1Kr1is1qt9yu9wsOi8fksvmMTqvX7Lb7DY/L5/Q6OYC36/elAP8P +IHnF1hoaEFoMJh42BhIyMjoOAmYhyBJmVmIqdm5x+nZsThKWmp6ipqqusrayuoFGroRy0dLZSt7 gWu3G9WbO/E7J+xEDPxgzDC4YJzsB8lsOUCbPA17LCh6gPm8vZwXae2d2A35rUCuOE3+DH7ZrnyN rc2RnuB+KS4Nrz/eb34P3TpF6bqpkyawS7V5Acetaiju3T58+wb2I2gwoyRu/xYjtrMXa2ESkQzV 1TN5L5w1gAeheaSob+O2lgdr5jOYkAvJkjsxcrS0KObMieEK8pM4UyjBjkdzbuk5D6oKqSaoCrGa C+uIZUG0/vDqCWwYsTzIZjIrT6eAtWzbti3ZACpXiIiOoM2xSwDceCdRJn0JFOdKfz7u4mCld6/T DPbyDXbMcmXGr2kdJFbseNbfpHO5RobnsofhG3kx03WL2m1DTqEvfh7YesdoZKdC7Apw2bTcuf4C A11Kc93sugol5MY8vK/oypaNp05daQjvsjpK2Xb+XHWE47wmJXfgEivVXtzrfLdynm/jCBsRWi39 x5X8+fTr25ePF6UwlXRFIf82DSBfF7EXCUIRsYBbgApCZCBtzEg0Gm7ZabdghdGAlBmCE0z4loWP jFKBUf1NBUJ5HtLBEQjJUWXiiXKkmM0LLLr4yWLKtTAjjd2NqJmMJeqoBykifJcjkAoS+aORR8Jw n3UtKpkGKjEq4ceTUKLBn1xMVMkhW1eKkaKWSwjT5VpfThGmB+mpGYKVZ9qVkJhUtvmmFEIe2OOY SdY5TBNF8hnHmmzuCegbgtLzgZuFlnHojR0ouugYjebpAaSRigbidVvSeSkSMA6qJ6GdEvEponOK OupVNlI6EqepkrqqBpOyyoGlr9pwp4rFuHoro7ui2it1PMr6a6JlBktDqcT/bUblQ8AiO1VtxGZ4 hVW2QjukbTJpYS22TxgYFBbderulezrxSm6r1FaLbrp+nltimde6K8OsKCQ3L73qNegouy7kq+9q zPbrLwsAB8zZwLQWvMLBCA/4nkLOJvqwrOuBesa4FcumhsYb5xelCA4jnOuUGYv8cYixMraGxylD oOzCd7T7clwrY2AvrjTXLCCepp78LM8pSUvwzPEeK3RVLb8wctLILP0vz55l2zHTNYcWcdVRv4y1 piFvnXLXukJt8NX6ef112VdnOrbWaod9sy5sDNd0pzFP2/HElYZNNG3umVtj0PrudLFf5u3sNMzs bBsk4gGb04xT4aLoOL1i/8N89oE5S1e5u5crPhTjO1LMdeYmH955uhWhjbrgibMcuOuvyx076bOf TnnqtwdTC8q7Y9y67Q8LRsLmquqOLPFb9Y588qawnrvs5JplfFe+/150oNd/XDLwwT9a+rCw1w4+ 3OLjHF/zyce9bOPSe3v3+O4b2+X0bOMefcPYY858C3Vznz6w7e9B/XvbANnnBnxhi3oBFOCtGFhA gyGNT2SpnvV48D+4PA9nvLEgEKqXQbutK386COGiAEe+HJgwKr+50eRIqEIKmo44/JnfDlZ4jM8h QnS8eAUMcJiVGf6MEucBoix0mL1DFHGCFereEL3jAyPqyIPv6oEUaUTFTf9Z8UROTKIhLHjFCPpM Zo0Ao4fih76w/CCMNjScFzexxjOOkIxKjOOR+vZER3iQjRbKYqi2OLhQ7PGMXcSbJgbZxHsdMYp9 VKQgGdnECi4SkJG8Xx71aMclncCPf8RgIjc5SUpqsg+h9OSCNog/KELSc6W8Yf1SxclTJYGPWLJO 2zoRywXQsg3iaaURdsnLYPythodkAjDJ9jSPEFCNSzim2ySwuPMp0YemBIYteSe56TDEjEraFjH3 ws0cFhIdUwNXgMIZRFACCJ2+pJpuMtlO6GkQnvFM5TbpGZZrem+eUWTiFM9pTCjlEj0B7eYokeBM MYKzoEAaKEGbaVCAQhT/ly+k3ToZSkRpJvOiBFLFD3GpUfAcVKQb9R9I5wjNkfbMZlajqDZVJtGS stSBLnJoFXDRiyIayaa30BtJW0rEcbKnV99JKLfsWaii+hOG8hsVCN/5wFXODY+GdKpU55YZOcHy qs80iVatKkpkelUXKNwqV7EUJxqiNKlrXKpOMiWSb16KpwcwKpqGRtdgKsGuvkBgnfI6AL7Wy5Ih 6mCwACtYJvkVm1GdKHPGWFWw7hUMaLToq6gZ1qOGlAKAfURbX+kLwr4RWp2VV2JJw70mqXa1rZCX Zg/oUn1eErZX6ixtjXZbGeZWt7t9k217+4XfArc4w/2ScIv7WuTuVLkC/2VuRJ37T+hGV7o1pS4W rVtd7N4TKQprKh0tO1vwfhemSBXveL8i1/HcMrztYy9n1zva98LXCAXi4XkZu8/I3jel5W1vfPHr qWVCNo359e5++Vtg8+p3wQYugkrqy9oIS3jCFK6whS+MYdVqUTLa7bCHQ8svzoLjpe+t6A5FC81v kHiowvkIMaJpYv7FJMTJVDGNWRqUGINnxDw+bmxgOhli/Vi+A1LrkAk01iMrIzxFRvJYBxwXwSh5 yUl2I0knM2U4QZl3WXZQl3ccZA4+mXbQ+DK1wozNFjpDOGNmrJrlVmYrn4pwbd5hk4FsZgKiGcg2 eTFk5HxlJtclzlvWc9+f1foYHwMawXm+UKPj9Og/RzphiSYvhvC855QSOnL/WPQyNw3iuNqYJC6e RakZA7mFpLrFmD41kVdNQ1i/WtYdLbWOP4zrXOt617zuta9/DWxeFpIapAz2TRn8X2PTVyD4YLUb P+KXMkM7dBC+yXGty41L15kpmgNOC7ndbeAo+4OSMzRdpGy69bSmcOMWVoaq7bOi/DndkGHbtbX7 qcb8WN6UVma05Xxv7Ob73/G+ybz9HW5QBBzb5X7Jsw3u7b+wo98Ib3dhGs7mfdc7zPpms0PWavGQ i3zkJC95oQoAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0056.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Performance
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0056_image020.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC+2dD2xcRX7HZ9Ze/wcSMJER0JocyQGByAlJnHMLJQSjcImvhoMLxy6qI3Ced71r O4ljx5w3bABBxEUCegXucu2lqtJURseBfFXTa9OiqIKgIJpIzVFZ3CUVOZUU5USiIznAqO5n3ptn v33ZP+Pd5/Cn2ei779/OvN98PzPvz8x7zn++/cZOYX+aQrtllz3323ulqBai7I9ahKgTc2apdep7 fsUlYm7j2rY7V0r7d1Ic/ZqalqmvRiH6FqDVar0QO8TUpykkxG45tfzqMiH655Okte1O9WuVppxM qth2CUqgFeqHYVQuxG2kPSV8n/fKRXqkTLTxy3vFGr5vJ4Q7RUxYYrPYKDr1z2YLVQYnb8okVGE6 tGp9y/7tl+q0KvR7UJ905s9OTEx4Q3l2qmxzV7HnhBjgu59YHhLrxHQ/ZwbV96xHau2l4d8eFYdv uEg0MH3r+IZH1Lou5v8Lqc9bd3wyuCpZJd7+/pnBVZXlovnZM4P1Oqd21r9J2i2y3Nbhf3j8chVq FXmrZaHnMz/3Xf4yvnfdJMRrFc50+YbqrWraer0kLikVlN+cPGrnf9Xpo+LU9k8Ht191ZnDC9uXM 4Ad91VvVb8qdaiHuvn7O1nodT9/fSRvqtcTfQZ7+9A8P2KjEPI/37vxFQqSvy8Hkd+eRifJZzd99 fc2jqkxKVR8Wx0R5MX7NJ+edyZV1c/Z9FZlcWVfzL19WJrSTdBBMLsx/OeYXoSXMhOyT6F2G9f4Z tAu9gl5Dh9Ax9CF5Uq1moUbUhG5D7agD9aE0egbtQq+g19AhdAx9iKiWsyY+JPkxkh8i+Wskf4Xk u0j+DMnTJO8jeQfJ20l+G8mbSN5I8lkquSC5ILkguSC5ILkguSC5ILkguSC5ILkguSC5ILkguYqe UjilOaRL94ourSp1Wl1LaP+UZ0v0eZ3LB3Gb3hbWbUlqzfI4N/GoI/9ytjznTV03pL/OtMG9zlHp JiZC3ran1i/Qv3E/i/jF38qp31SSjzsvPPMyx77Kc+xLrf8OuhHdpcs6TiYD0tl2F3N3sXZcKg3I XPv/M76Wa7+W630t98Qvp/afUfvcPMK+8q/V/vtjCmeJSS3nius6T/7Xecrs9cgTQ/pKzVjtdyCP 3255ywuU1/+ZlyUGVd5ler9uOdV1a71nWVWnQuyLLUuhmFwGbkzusklMMgCvTeLbIDPj21CiZ9xI iCfzxNQwNR+6Qs+rtjVnaj33EY5UXkPktYbpdunWjTdXiPQ/r3C3qXVrfHVlQn/adVuoNPByqEQv h3xeDgXg5UCAXg5ov4azeDlg6OUiQy9HSvRyxOflSABe7gnQyz3ar59m8XKPoZeLDb0cK9HLMZ+X YwF4+U6AXr6j/fpVFi/fMfTyZkMvq2VpXqr0Xi/d5VK8rJDBeanyUn7VyXO9dLcV8nKJoZctJXrZ 4vOyJQAvmwP0sln7dUsWL5sNvVxq6GW8RC/jPi/jAXhpBeilpf1KZvHSMvRymaGXO0v0cqfPy50B ePligF6+qP36qyxevmjoZbOhlwdL9PKgz8uDAXh5IEAvD2i/3s7i5QFDL5cbejleopfjPi/HA/Dy 4wC9/Fj79b9ZvPzY0Mtv+Lw0uQf6NT/6SQBleFjn9Rw6g/7aVwZ3W3meMqh8tnCHt4i7jkrNslLL 7Y9V99OfnMf+2HI9H11zTKw4Vru19Xp5v5q/qu3YZH/slsHqrP2xadZn9seKyT7YUvtjG4lB5S9+ dMykj/yteh2P2x/bQXrVH+tP7++PDefoIw/7mPz+PDDpaq2bXKPmj3VV2zwOy5OhHx6oEq++YMZE vFo9yaTxcN3WVffMFx3k5zLp8OzH+dxxr8vkRPhcJl2tDpO/f95hcuTHhZl8Z/acXpfJmxsdJh89 7zDxp8/HJGzQxs+S/fMBHadUXjvQp9I5l3rbuLvN5FokbNC3PFP9Nm0is6/LXS7lXHAr+l5AHqu8 esjrDqYpn8futqD6bZSXrSV62erzsrVEL9W6bwbgZULndYv2co3PS3dbOIeX2Z8VcD6TzwrUOs8K qN/fb/u+EDVN9k94z2GmPOpL5FHv41EfAI8rAuRRp3lcmYVHXR4efo+XlODxcVGax8dFpsfHReke nxDBefyucDz+QJzr8bvC3ONlJXg8WqLHoz6PRwPweG+AHr+kPf5FFo9fmobHy0vweFuJHm/zebyt RI/VeN6TIpj7iRv1OIDyeLvPY3dbofsJx+NF5/hr4m17id62+7xtD6D+3hdg/V2lvV2bpf6uMq6/ i0o61zWW6HGjz+PGADyeF6DHDdrjr2fxuGEaHuc61xV6ZsD1+DNWDMpzl71j03nHlkPSeb5jGmPL ql49ofe5XvdxqOvsB/RYehu6B61FUbFZrhOW7BR3yLgYF70cVfvFX9DGt4mtYoN4TPSIp0RS7MDZ 50S3eIHpTqa70G40IuLiZ0x/znQv2of2ozfQW+gwOoLG0FF0HJ3gbvQk09NMz6JxJGRMhFE1uhhd hhrQ1Wgumo8WoCa0FLWgW9FKtAq1oXvQWhRBHahTdok46kWbWN7CNIUeQ0+hHehZ9DzbdjLdhdQz rCPoZTSK9qJ9aD96HR1Eh9ARNIaOovfQ++gkOo3OokLPJxTiHsp4pmBi2txv1/eM+bnfDPffQOJx uH8L7rfDvRXuD8G9E+4Pwz0O9zjc43CPwz0G9xjcY3CPwT0G9xjcY3CPwT0G9xjc8QiNoaPoOOtO MD2JTqOzaBwJvAqjanQxugw1oKvRXDQfLUBNaClqQbeilWiVtODeBfcuuHfB3YJ7F9wtuFtwt+Bu wd2CuwV3C+4W3C24W3C34G7B3YK7BXcL7hbcLbhbcLfgbsHdgrsFdwvuFtwtuFtwt+Buwd2Cu4Xf 1rSPqWv12JV7bx3V04qszJQGZTH7SJBnnGnMUy/U8TJOfnFc6hVfk/3iv+G/D/6Pwn8L/Afh3wP/ JPyT8E/APwH/BPwT8E/Avxv+CVJ1w78b/t3w74Z/N/y74d8N/274d8O/G/7d8I/DXx1tiIi9h1E1 uhhdhhrQ1Wgumo8WoCa0FDntPg7/OPzj8I/DPw7/OPzj8I/DPwb/OPxjlCwO/xj84/CPwT8G/xj8 Y/CnTqNdzO9mOsL0ZTSK9qJ9aD96HR1Eh9ARNIaOovfQ++gkOo3OolLbvfd4P1HE8T5e8Hg/QLsf oN0PUOIBuA/AfUAOoa3oMfQU2oGe4xjxguiWHO/lLvGQ3C2+K0fEXfJnYrH8ubhU7hVS7uNOZ7/4 D7iPwv0v4f4s3B+Fewruw3B/BO5b4D4A9w1w74F7D9yTcE/CPQn3Hrgn4d4D9x6498A9Cfck3JNw T8I9Cfck3HvgnoR7D9yTcE/CPQn3BNwTcE/APQH3BNwTcE/APQH3BNwTcE/AnXMY2oV2oxH0Mhol n71M96H96HV0EB1CR9AYOoreQ++jk+g0Oos+7+N9c8Hj/XS5t8C9Ge7XwL0S7ido7/9Ge/8B7X0r 7Z1zIkeQN8QKuC+HezPcb4H7H8P9VrjfTooVcG+Fezvco3CPUMgI3CNwj8A9AvcI3CNwj8A9AvcI 3CNwj8A9QhQRuEfgHoF7BO4RuEfgHoF7FO5RuEfhHoV7FO5RuEfhHoV7FO5RuEfhHoV7FO5RuEfh HoV7FO5RuEfhHoV7FO5RuEfhHoV7FO5RuEfhHoV7FO5RuEfhHsXvaFHH+w7P8f45PZ1dMjOl4s4N 17H/fxLOM1ZuHVop1LpB2ndC7sfxN8S18i3a+2Ha+xHIj4l/hfMv4DwK51fhvAfOP4BzGs5pOKfh nIZzGs5pOKfhnIZzGs5pOKfhnIZzGs5pOKfhnIZzGs5pOKfZaxrOaTin4DwM5xSch+GcgnMKzik4 p+A8DOdhOA/DOQXnFJxTcE7BOQXnFJyH4ZyCcwrOw3AehnMKzsNwTsF5GM7DcB6G8zCcU3BOwTmF J8M5ObtjMep8feo8jo99V8+rtq7GlJ5FfajKMz4m9tdkHYupYr13fOzBG65puwKGQYyP1akYyH/L nxYei7nohjnr63U87vjYGj0+5k/vH4upyDE+VuEZH1PzH3wO75BcdEONpcqk1PSt4pgE+Q7JdJj8 csGcx76KTH65oObxLysT2klXEEzyzZf6DoQ3r2zP3efad67+oNMi//MR3rxM+4ae0Pk+Tb7t6G98 4yDutuo8fUMddl4x7oY68WgTtbRT9LHuYu3bpZ64avW0Tmu6TIr1TnmwsETvvO3Ybbvlnrb7RNlM tN3Pt82qz+fZZv1tV+Xlr8OHdT1T/Zx/oDl763BYr5uTpw5vUe8KipsgtY58Nts1uVes53s93Dba tdqp2+vs+j1b93nWep59qPPUIXfqKuz5fUWWtlCRpT1kO2aU8m6Tei53NfoI9eq62yJ7ZYtcLT+y 1Vvw3abKIt9tqszxbpM/pnCWmFryxHWdJ//pvNvk7vd8v9v0kd6vO07gLpcyTtCivXO3/XvYeVbg z2uzPyvQp5u6elZguuMcLidv/L0lxv+RL/5iYpppT4t5Jn6175n4zzummWD3RfVp9VesPl1odzNf n4J6Dz1X2TpUP6Th+8um1++/0vm2of9RYyW+ax93W0Oea59h+9qnkeueTq5uNooernh6hMWaQeb6 RRdzK+2rIufaXl0N1WgGZZ7xX3c82L+uWq8rz7LOn7bccLxelSsesJfbdb4P6ufQE1m8VNtqCtwL KS/X+LysyTJens2vQvecbn1Wz7Wr+w//8kyPoV+L/oT9fFvX5bX6by2skPfJ+1Gx7V7V0fvy8Jzp sSK3XA/pclmT5eqkXJ0llavTsFzeOluoHnymL7Tdvgb/8kynL9T3oe79G6TzN6dmXfg7G/bf2WjQ 91xlssr2tNauqPadY2i3bAqV29VXiqvUO0GhsxO+3oS5qzoTA539sYdy9SGo9Dfxayf9LPt4IsQ+ ey8yFJr77aFN/Z3J3D0NTnppT09N5vO7ouOYmfwuTHNNf6zefaF6rTbsWfr/3Rrdngrl3pJQyD4T LLWXbpTOuXhOedPmps3uXfNSuyPq01B4sp/D8b3M/u1nIbfvxVlbbq+9snyPXGy3Pyetv06HvWkn 11b4yGbfQ6V9rri2rPycvwXj5u3m5t33VNxVk5GqY/W8snDWvynjz8tpy1V6WumLUGaUrtI+ttWH VE75YppKMRWTSnkZKcelyvkP7UOjWsoWj9p/4T1m30MjpZ7ag1rKnk6UvKcN0runDTnT+ffwpLzz nF+qLdXKPbs2ypCqF+W6l1NdWS+uGSKVotG0vDJn5EPGkQ9lRD5kHPlAUZEPOH/JK0/kI8aRj2RE PmIc+Z6iIt9jR744T+RjxpGPZUQ+Zhz5O0VF/o4d+c15Iq+WppFXS2/k1dI08gpZTOQVaiBCLMkT eYtx5C0ZkbcYR95cVOTNduRL80QeN448nhF53Dhyq6jILTvyZXki32kc+c6MyHcaR/5iUZG/aEfe nCfyg8aRH8yI/KBx5AeKivyAHfnyPJGPG0c+nhH5uHHkHxcV+cd25N+YjDzXufQn8tfTyv8iO//n dKotjYua1B4q+efcaXxS5J1Gte/OpVrn9/sS86vOUe7n5dkifN1BKqcVhnPcERR/raKeufNed7UZ nn2+J28t4uzTQ6r81yqtxpG3ZkTeahz5N+W8aUV+iR35LaRS25oWNjUtFm7ty1WKeuNS1GeUot64 FFcUVYo6TymWFCzFcWFaiuPCWwq1ZFaKE6KYUrwrpkqxrGApRo1LMZpRilHjUuwtqhQveUqxvGAp thmXYltGKbYZl+JJcaOY/jF5SKeiFIvcMmTPv924BO0ZJWg3LsF9RXFYNcVhUeGW3SjM7zfbMu43 TUsxr6hSNHhK4W/ZU70cg/Iz6ZZA3du79+ahHPfm/2j3Fzxw4b258/TeXK4+GIfT1GjH9DhdeM8t 6PfcCh0B7Hd6yypyvsNWKP1P7fT3Xng/LeD30/K3L+c4ODGt4+CF98lm4n2y4I+DF97/mon3vwod x+bbPGYH8m5XoX09be9r5YX3tjzvbfk9c3o3ThXZu1Hj6y2p0fl98AXLL/v03LE+79ZTkzXJn/rc vqvTOep9Td7r9Mvt6/SnSa36Qu6NJTsbN/V39omLSXWpqBMVohbV8c+kNIXjXCivmHacN00+k6AY qLccpsvAcXGhfj5pcc1C9q6evrr7poF1ic2djb3rG9fHNm7qp+zr+jtnQ6SWEoftclfYUv/C9vqK SU8qfK5M0co/clo1OXLqf9LcbOS09pyR02xPrGc/W9X6IvXzWZ1l1MFsrFTteerOUi2Z3Vn2Go+P 9GbsoddgD2b5znzkqzNGdlYbj+zMpDczX4Yvivtf/Xoz3Wd1/BE9IDty5pzv7LHAPpq2kVo943p3 Y6Jz3caeWI/VOBjr72pcmdjMuWTjJnV0LNP9PpWTc9WizF525tyt5Tn7uOJFxjjHjvFBUtfYMa7R MdZ4eqOmovKekfvkGel6btIvZf8fU2WhyWdRDfsHZZss9g5wao/OU6Jme+zMs0dvPfFfm3w2Yb7W /Z+1/g+mHYvwjHQAAH== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0056_image021.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhkgEbAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAEAAQCQ ARkBggAAAAAAAAAA//8AAP///wECAwECAwECAwP/SLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bz+i0es1uu9/wuHxO r9vv+Lx+z+/7/18BAQ6CDIOGgImKF4eHCo2PjgSFi5WWDZALmZOanJeflZuejpmSgpSgqXeim42n pKqxeKyPnYi2srlyp528uLW/usKXksPGisXHyn3Jy86zz9F6zdLVb9TW2WrY2t1l3N7hYODi5Vvk 5ulW6OrtUezu8UzwVQIL9vLO9FP4Cv35x/ZJ+UeAIEBdArUIWHgwV8KGEHc89GIwIqCJTCpKYGgR /xkYjRc4dtyDceMPkBhQjoxT0kE/lStjVmjJ4B9MmTgh0Lz3MeecnUVupvSZBZUmU0cxUTQhlGgS WqNK3ULRlEVVp9+AAYM11ROIilexRoPKVatZD2DFxiP7q9irD2EHqn1ntlUwrwpTxJ3bwxcnv1GN AhWy10JhvlMGPzg8gjHinyFAOn6cjibKyZTFWUaDOTMSxSdb2BS58CXp06ZT30O9WrU/zxpA9+hM gbYM1v5wF9RdurXv3K53B+8N/Ldw48SzrrCdgTnsNJtNPo8ZPYLz6cuqYxcre8L1xdtDwRApOvyi 7jm+J2lKMLl73vCHx0c+v7j93crRdoBZ//j99//y4RMAgKeUVuBCBwqQ4IKCGNgggg8qGOGBrzg4 IIQPVihhhgxeuKGFGL5SoIggYljihyamiOKKBea3H1rDPfBWJCLWaOONOJpSCID0Bdjjj//191pN cBmHnzct7ZWjhis+wmOQPkKZxQBUVmnllVhmqeWWXFq5QJdghtklD76gd1YLgAX3og1jKuBlBGJm +WWVc2o5BJ3QaWWmmYQUQ15jMWA5AZ5ubtkBoQQgGpFdKfCZ1AvqNUCleTEwikKSFDhawwCUymDp CZhOoOkMnHbqqYgrhBpBM5G2UKqpWjjKqgqRTgprpYSowOestKKg6K0uJLNnqjy5wJytwFaKo67/ RUrgSKu+vprsDBMZFUlXeG2A0rNJvPmRbrWB29yfNZFbLJBdsDWJW83q1O2vbDxZgbzJuWTufUfO K64ONs5Ul55KXUpDXMiKca85UvmL16frWhvCqC5tIC1c/dFbccUNMezuwgBjSwK3xnIwsbbTgtqx qP96xa5+Gxd051clm7zVTKgMAtjKa+Z6LhAjawAtrKTs2m7ARvQcsxjdQUykxEf7ICyzMsewl9FN 1/B0o1G7HDLTLFcNQo5Q59zVz4cO7fXXuHY99stqn/2U2WcKQbXbnj6KdQlKL53B3HTDwNWweA/M dd83/B32CH5ujQHf8xKOtsqH+yxjA2RrwHht/45/oPHHbZtV+d4wZ655WQKLfRfPooehWN47V3B5 6lSsLvjisI8TuN7LgQ537UBgZFDiPrzuHe9F+K5z6zoILx3xGFxdOge/F6E884gfn/XjM1Q1PfWj 3xi55rNbsD332AfLOe69XjB+xOQzYvfzJGP7ufiRtW+B4XfHL/8P66Nv/6q2AJz+TteD/v0vNie7 ngdYhzw4HdB8kNOJsIL2vmwtsAZNMeADm0e6gCHlL7WAysPYZxX17W6DHWALJEQoOY/Nb1AUQyHK 7pcym23FLYAxDNEayCYZEouGHAOhED84QAsGr4g+dBYQP2jDIXqMAwwckuuQmEQZ9cuKsChTzf9w Br7wUWB7L6xipkSQjJ8JRYNilIjYyljANJ7Pe/lr4ZnC6MCQuPEHE/lHFGmnwzsC0Ijfg6IXYdg4 P/4xbzuJYgYN+bBlxfGClCsP/azDSPf5DXo7lGIO+EbHSs5Qjmfp5AM46Uk8YpKAPEBjKetWPqlN cpVHoMkezwjLDYgCTSOMZAlfCQFReu2WLHiIHkmYvOXVEhOOhN9Q7uJLSRnzmJkM5injlkpoGkGW g6yjNWODqkAiMHu83CYQn1i9VpoxnMQUpwUF2Jz9VfOZ4nSeAvsYynfCc5vyNIEwyVlMdWagRj/s oiun6M8gtESR6CyoGgXKwxLc5HLNtGZChqn/y00WUqEc7JPdrDVRF+4AohhtXg1HQVJUGnOPBA1p OZlI0hweFJxfVOn9KIENFYawLjkU6W1SKtNDEtCmTjTpODWpF55SEqObK+nMVpiydgYjoom6aEGT 6heb0XSL0ZQq8HDwOqieLalkHCBKJaBKaALUmzodKFl76rRWSpKQbF0oJBsKqJhiLqRwVCY98RLR rsaVmuV0arb6ule2snOvY9XmX4eqT4Yqbq2LtaRe/7lTu0a2T9185DfpKgKY+PWyaD1qPblaWNDe bpmjvQFITavZuwK2h6xtWF7nmdbHKja2rEyhWm+L27TplrP1g+s9FXrFyQ7vF81crUwT1trj/3rO oq5VKVhziVpAbqq3t+jgaW25WwiUFZ8BDO0Yx2NZ7KpMaL8F7gkZoFzzBtaOqaXBZ3vqMPESc6vy La10R7UPin7UvRp11gQrSETuQqq8AK5geMPbigJv9q3eTTAylRjEJjJVqJ/UWlEhG13DkgOoFuZn L9cG2w7vd4nZbemMZItY/wVKv70FcVf6K+IXS7jGVW0YjXZYLfJy+MbWPZ+BbeuA7yIVl4J0cV15 O2LQaorGzy2xcz1cJvtWVBKinO9ycRhQeib2SxJuBnqHvEsmx1bMQXarZMscYQCj+bC1/Y5nYTzV PxpXtFiW8nDr3DLaNjnKpIKvYSdn5Z19Of+qYeZyoa2Xvh8nOLOLzuoJ5gzk3FJWyaFztGgrrWY8 V1QGWub0XC+KXxgIz6tHFvShjSxqhT14w5pudaRr3FkEy3rWyJ20Ufd8a1eTGtMrUB6qiYvaPAda 0L1+tXPRs8hkS1PZJGi2s3Gd5vUWedpdXrOGTV1dIDsMq9ei9blK7QJhi1rGLFz2p19gbk7LuC2q brRwse3BFDNsxb52qAnpTWiW3pvM+t41v2fGYIKLm9Hzm96wUZhjLbpCMPEGtq8EO/DxFpKNNqZz 7JbE8Y57/OMgD7nIQ24GeAhI3mY+w6FjVXJoLznWaFh5UVp+aaoIPOaykLmlGfs5MKbrYxz/BXht a1oIONuuj+QONihnHtamXhKQFKRwkslg8viyYHwLb2zTg7guGlm1yjre8U0f/nXrhVi2V91xTl/L BXb0w0wqYXXsDAHy47GUYyGWil3ybvB/XZjgZ7fzGPbBp7izJLBMdLgT9Q4wvivV7ze8YdEVjGGW 15Yp824DUG5p0wt7vhct7ft5Hx/4bFFD575l7LaVvnSmk/FmR/m6DT8fGERksfRqV3vadfxwwSMN 6elUAda9gHoKF9/3qgP+ulMwfOI/5fjIP/pd4Q7zNUA/MTTXdsBHSYfrS8H72ba4rtu8C6Dn8/JL AH+keSV801meulz8vYtMwn7mT/PnW+fi/1WL3s2yi50I6tdcEVN/J9A/WQc/cVIwCRRBSMF4tAeA 2Tc8hXdtdrB5wRA0+1dSs3dT1cYvESh+48deq7BSdmdvSxV6HZgDAXhnjBaCbgINiONSvDCDM3iC pVBfpjR/UidxEgMv3dZ2SgBMvfOBI0aAIVBWB8iCBrWCK+hnNUENtvFdSeiEHqGDi+Eo9iB3gTBy XNiFXviFYIhv5xB/0Zdh8yI86JGG1KZ977VdWncF6VaGO3iGbLh1b0iFjiVkd7iHc8d1c4hp+6A8 4FAYaiiA1MWHbeiGfYhlC1ElpeGIjUgljyiJkTgAk2iJlXiJkxiGI8KJnviJoBiKojiKYf/IdE1Y cVBziqh4KTi4ijFXiKCSNEbXSLLRit1Ti2moir+Fi6UDGkLYSEq4i2tofDKjiwaWNOGXiCOEi8w4 jMSodcb4TchoiHmYS7XIdsdIjSk0jdF4adMYjKP2emtnh033jbcji92oU+aoiOWojdl4juD4jvCY LrZYjbTIi+5oS7KYj/rYKOnoigAZkAI5kARZkKYyjtPiSP94DfVYMgy0kG8DkdlBLQHRkAlZgooX e6fncBh4exIJQV0HO84DbqRndk7nCpW3DcyVOSM5Y04XNwUWhzi3TqJHOC0Jb2gnZvrXfwenOmj2 eI5zkyNFMzjJYwzZb9jYNEJZYer4RA4uppJImYIxs5R3dyZVWWHaNZNXg5BuQ5VTIYYaGXthGUIW aZDfZ5byh5aBEBEJAAA7 ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0057.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Car-Hill Problem
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0057_image022.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC+y9D1CUd77u2Z1kkpnMiTFWbhLLk4pxSI5HzTpq9ESjGeOyOhmGMFwZDqMdBA0y fTgQposlLCynQ5GJFHawiwuheqkuBsNyvCzCoQhjKBnsYrsoLaqL7UvJobpIh0CYDttLweX2UJ75 c9nnlYcJYTDBPM0M99Y19Y1Cdz+/7/P+3v79ns/br/ivvt7/w3T713fvaTD/9PafJpPMpm+ZTPf+ 59A9pm+bHltvfM/4/7P3P2zasvnv44/+z+bbzzObxkqM3580/rfZZPos+17TZ+/eazIedZg+//Xd e0ymBvPnX//LPpMp71m85H+JP2qoGK/5z/tNpgfw2MOof0S9bDzxG6j7TKbDeO20acmv0ftMb/3H e03xeGaS6Yf4/xG0cNRkNWWa/leTzZTBpz2C+ivUN1HwZDLMpLG+veTrpY9v4GuN1nehXjDP/9l0 7w9MK/91AVWHakVdQ/WjPkZNQWsOA6I2o76LOoz6ESoNlYt6C3UBVYdqRV1D9aM+Rk2h5ubw8im8 /GO8vB8vv4aXt+LldXj5Bbz8Lbw8Fy9Pw8t/hJcfxsu/i5dvxsvXGy834eUmvNyEl5vwchNebsLL TXi5CS834eUmvNyEl5vwchNebsLLTXi50T1czLvpp7tWujVcv2XM/aJj+DznIQ51mI99Y/7Uuf24 mcd/4decfb6Wfr2c5jOfz/Nbf4Pfn0Dds/C6ubnbf1x4jvHFdj5n4dcufPP/NH/+HJyLby382bDy 7/G/XDx+L36fnTOO3Oe/nJ+f21u+jzPvH035+H8ezsUzpnTT3f76zZvG/1O+8dDtr8YbQ28af/5F 9QOGjzn/3zz97kbTv5i3vvrx7cf7Yv/tzX/5dw+YfOd/8+b3H7jP9HfO37z5KJU+xPcPP/VXpgLz fbdrtuXBf3qh9UHTj6BnfG38+hHH+fxXh/kyJuWnz5lMyd+e//2Fn33rn4zfI9vM6MtsNt6UO+I/ vq1vcn1smi777Ztlf/2bN+duH5ffvPn/5n7rn4zn3Me5Tfzbx/75Ufbz4W7z7Tf1P+L1adBc+vrX 829P5x/n4d5Ff0anb21d9P3Fc/Jf/oxzYhxn48+Jf/vgpdueUD9K+Nj0debEOBa/e/rf/uxz8u2/ esz/3+OcfPuvHvxP/63OCd4n/zEac3LvHdbD++6wHhrfT0btQL3CXn6H4fPN84+9gj+9gu/+zmxU vvlOa+QpY4/kmv4Cx3ph0REzfz7+F47kgsY3lqzRf889YmlP31imJ+PrO/W1dZH+1kWeFx+jRT28 tYn7kDFu/pfsCQt+71vG7z2L/C799cwyPRjP37do3M3MQo8u+tp42p16WU3NOx2fvK84Pl/VU96S nvKi4FPRvJPPOtFn3ZKe6qLgU9G8k88+0Wffkp76ouBT0byTz1nR5+ySnmaj4FPRvJPPLWbNp/H6 xT0tfK34VDTv5DNe9Bm/pKf4KPhUNO/ks0D0WbCkp4Io+FQ07+SzQfTZsKSnhij4VDTv5LNf9Nm/ pKf+KPhUNO/k83eiz98t6el3UfCpaH7dvLiSnozXL+4p36z7/LqaX+bzkOjz0JKeDkXB59fV/DKf 94k+71vS031R8Pl1Nb/MZ59J89m35L3UF4X359fV/DKflaLPyiU9VUbB59fV/DKfPxF9/mRJTz+J gs+vq/llPjeb/tvfV+72msSXef6yaxIm4ZrEf2/75VrMKqt1nFbj2s9anLu/5HG6NwrXE1fS38/M X+zvZ+a/7BpqvD5myTGLEc+tI6h3vuQ4PfH5n+/ZyD8b17Qf+/z7piTjswFq/W/Q+iF+L1v4/NZ0 /WXTW1dfXnjM8PPDJevhHH/9CH9+zrTrtv8HVjDH0bgmuniO86Iwx3niHOctmeO8r5jj5e8VmP/1 x3sF4ufvFVhJ/5lROh9ep5Yx5//7MufDwmP3fcn5kHT7fPiu6X/CrrpwTqz03IjGdeTF50ZdFM6N OvHcqFtybtRF4f1fGcX5ruSc/vMy8115V/O9767nOxrX0xfPd18U5rtPnO++JfPdF4X57o7ifHdz TgeWme/uu5rv5+96vqPxucLi+Z6NwnzPivM9u2S+Z6Mw3xNRnO8Jzul/XWa+J+5qvnff9XxH4/OV xfO98LX6mY8y38brY5Z8PqPO91Pm6Mx3ErWMOX3G/KfzvfCY+Svy3HdXOL/R+Fxp8fzGR2F+48X5 jV8yv/FRmN/YKM1vGrWMOfzxMvO78Ni9X/F+nn833917ORqfrS2e64IozHWBONcFS+a6IApz/UYU 5/oNzudby8z1G3cx18/f5VxH4/PFxXPdEIW5bhDnumHJXDdEYa5rozjXtZzP5mXmuvYu5nrfXc51 ND5jXTzX/VGY635xrvuXzHV/FOa6N4pz3cv5/Ndl5rr3Lub6hbuc62hcu/3ZkmuSP4vC9WT1M4qY JdeT1bm+FcU8dovz+V+XmetbK8xju5bM75/jWmq86YvXnxe+/kt+djC75Jr4bBSuiR9AFUXpWqqh lQOtWPxevGSuFx776mupe1b8flbvM4g3ffGegIWv1Tk+JM7xoSVzfCgKc3wkinO8l3Mcu8wc713R HO++qzm+T5zj+5bM8X1RmOP7xDm+b8kc3xeFOf5WFOf496b5Of72MnP8e1N0PxNR7y+JN33xXpCF r9U5VnqaXXJ/ymwU7k95CfX/mKKzLxta/zfn+D+ZvjjHC49F6zqJel9NvOmL98AsfK3Ob6U4v5VL 5rcyCvNbE8X5Lef8upaZ33LT3eeu1bqfKN70xXt/Fr5W5/cn4vz+ZMn8/iQK85saxflN5PymLTO/ iSuY3913Mb+bxfndvGR+o3HP8GZxfjcvmd/NUZjfZ6I4v09wfv9mmfl9YgXzu+fPNL+reU/Tiu4V eG7+XoH/cZ/Y2rv/6av+Lv+Cxu/xjTfNf/r14nsfV/r3Ke/090fvW+bvj/71PfNjPrvo3py3/vj3 R99A/RSVjjqBOo76Ieoo6nuoXahtqEt4fi2qCvUu6h2UHVWAsqGyUK+jXkMloxJQxn9HUC+inkc9 h3oW63a+ORHjJmLcRIybiHETMW4ixk3EuIkYNxHjJmLcRNPDqCq85l3UOyg7qgBlQ2WhXke9hkpG JaBeQR1BvYh6HvUc6lnTCYx5AmOewJgnMOYJjHkCY57AmCegewK6J6B7AronoHsCuiegewK6J6B7 AronoHsCuiegewK6J6B7BrpnoHsGumegewa6Z+DhDDycgd4Z6J2B3hnonYHeGeidgd4Z6J2B3hno nYHeGeidgV4W9LKglwW9LOhlob8sk/Hf66jXUMmohNtHOwuvz8Lrs/D6LLw+C6/Pw+vz8Po8vD4P r89DP3nQyENPeeglD8/Kg04edBbOjTyeG5OLzg3j8TyMk4dx8jBOHsbJwzh5GCcPzy7GM4sxRjHG KIZ+MZ5fjOcX4/nFeH4xnl+M5xfj+cV4/s/x/J/j8Z/jsZ/jsZ/jsZ/jsZ/jsQo8VgGdCmhU4DkV eE4FnlOB51TgORV4Tg2eU4PxavC8GniqwfGtwVg1eG4NnluD59bguTV4bh2eW4e5rcP36/D9Ony/ Dt+/hO9fgsYlHItLGOsSxrqE11/C45fweBsea4N2G77Xhu+14XudOJad+H4nxu2EXie+34nv90Cr B4/1QKMHz+/BYz0YqweP9+DxPjzeh8f78FgfHuvDY334/gC+P4DvD0BzAJoD8DCA5wzg8QG8dgDP Gcbjw+hjGNrDeGwYrx/G48N4fBiPj0NjHM8Zh8Y4NMbx3HH4Hcc8jGNux+FtHLrjeO04XjeO143j dZN43SReN4nXTeJ1k3jdJF43iddN4phM4vyYxOsn8dqFc2OS58bC5/xv3T5PXkEdQb2Ieh71HOpZ 0yyeMQv9WejPQn8W2rPQnoX2LOZrFtqz6G0WvmahMQuNWWjMQmMWGrPQmIPGHDTmoDEHjTlozEFj Dq+fw5owhx7n8F6Yg84cdOagMwedOejMQWcOOnPQecCcb37A/Abqp6h01AnUcdQPUUdR30O9gNqF 2oZ6DFWA172Oeg2VjEpAvYI6gnoR9TzqOdSzpvXQXw/99dBfD/310F8P/fXQXw/99dBfD/315idR j6Gq8JoC1Ouo11DJqATUK6gjqBdRz6OeQz1r2gj9jdDfCP2N0N8I/Y3Q3gjtjdDeCO2N6H0jet9o /g7qMZQNr8tCvYZKRiWgXkEdQb2Ieh71HOpZzGe+eQv0t0B/C/S3QH8L+t+CMbZgjC0YYwvG2IIx tmCMLRhji9mMqv3juWGcE8/yGtnCuWE8vgVet5jtqAKUDZWFeh31GioZlYB6BXUE9SLqedRzqGex 4udjtDdQP0Wlo06gjqN+iDqK+h7qBdQulPHfd1BPoh5G/RavH0K9i7KjClA2VBbqddRrqGRUwu0d bRvG34bxt2H8bRh/G8bfg/H3YPw9GH8Pxt+D8fdg/D0Yfw/G34Px92D8PRh/D0bcg/H3YPw9OP57 0MMe8zdRZtRvoTWJ+hj1S1Q9qgr1LsqOKkDZUFmo11GvoZJRCahXUEdQL6KeRz2HetZ0EL0dRG8H 0dtB9HYQvR1EbwfR20H0dhC9HURvB9HbQfR2EL0dRG8H0dtB9HYQfR00R6Dza9QQyo/6FeqXqFZU Pepd1DsoO6oAlYV6HfUaKhmVgHoFdQT1Iup51HOoZ3Ee5JsXzo1YnhvJi84N4/FY9B+L/mPRfyz6 j0X/seg/Fv3Hov9Y9B+L3mPReyx6j0Xvseg9Fsc1Fv3H4rjGwkMsjm0sfMTi+MbCSyxW2lh4iYWP WKzusfASi+MdCz+x8BMLP7HwE4tjHgtPsfAUC0+x8BQLT7HwFAtPsfAUC0+x8BQLT/HoOR49x6Pn ePQcj57j0XM8eo5Hz/HoOR49x+OYx6PvePQdj77j0Xc8+o5H3/HoOx59x6PnePQcj37jcezj0XO8 2Yv6FaoVVYuqQr2LegdlRxWgbKgs1Ouo11DJqATUK6gjqBdRz6OeQz2LY55vTkbPyeg5GT0no+dk 9JyMnpPRczJ6TkbPyeg5GT0no+dk9JyMnpPRczJ6TkbPyeg5Gcc6GX0no+9kHOtk9J2MfpNxziSj 52Qc52T0nYy+k9F3MvpORt/J6DsZfSej72T0nYy+F86NZJ4buYvODePxZPhKhq9k+EqGr2T4Soav ZPhKhq9U+EqFr1T4SoWvVPhKha9U+EqFr1T4SoWvVPhKha9U+EqFr1T4SoWvVPhKha9U+EqFr1T4 SoWvVPhKxZykwlsq5iQVc5IKj6nwmIpzKRUeU+ExFR5T4TEVHlPhMRUeU+EvFf5S0X8q+k9F/6no PxX9p6L/VPSfiv5T0X8m+s9E/5noPxP9Z6L/TPSfif4z0X8m+s9E/5noPxP9Z6L3TPSeid4z0Xsm es9E75noPRO9Z6L3TPSeid4z0Xsmes9E75noPRO9Z2J+MtF/JvrPRP+Z6D8T/Wei/0z0n4n+MzFH mfCQCQ+Z8JAJD5nwkAkPmfCQCQ+Z8JAJD5nwkAsPufCQCw+58JALD7nwkAsPufCQCw+58JALD7nw kIs5yIWPXPjIhY9c+MiFj1z4yIWPXPjIhY+FcyOX50b5onPDeDwXPnPhMxc+c+EzFz5z4TMXPnPh MRcec+ExFx5z4TEXHnPhMRcec+ExFx5z4TEXHnPhMRcec+ExFx5z4TEXHnPhMRceC+CxAB4L4LEA HgvgsQAeC+CxAB4L4LEAHgvgsQAeC+CxAB4L4LEAHgvgsQAeC+CvAP4K0H8B+i9A/wXovwD9F6D/ AvRfgP4L0H8B+i9A/wXovwD9F6D/AvRfgP4L0H8B+i9A/wXovwD9F6D/AvRfgP4L0H8B+i9A/wXo vwT9l6D/EvRfgv5L0H8J+i9B/yXovwT9l6D/EvRfgv5L0H8J+i9B/yXovwT9l6D/EsxRCTyUwEMJ PJTAQwk8lMBDCTyUwEMJPJTAQwk8lMBDCTyUwEMJPJTAQwk8lMBDCTyUoP8S9F+C/kvQfwn6L0H/ Jei/BP2XoP8S9F+O/svRfzn6L0f/C+dGOc+N2kXnhvF4OfyVw185/JXDXzn8lcNfOfyVw185/JXD Xzn8lcNfOfyVw185/JXDXzm8lcNbObyVw1s5vJXDWzl8leM9VA5v5fBWDm/l8FYOb+XwVg5v5fBW Dm/lmJ9y+CuHv3L4K4e/cvgrh79y+CuHv3L4K4e/SvirhL9K+KtE/5XovxL9V6L/SvRfif4r0X8l +q9E/5XovxL9V6L/SvRfif4r0X8l+q9E/5WYn0p4qISHSniohIdK9F+JuamEh0p4qISHSniohIdK eKiEh0p4qISHSniohIdKeKiEh0p4qISHSniohIdKeKiEh0p4qISHSniohYdaeKiFh1p4qIWHWnio hYdaeKiFh1p4qIWHWniohYdaeKiFh1p4qIWHWniohYdaeKiFh1p4qIWHWniohYdazEMtfNTCRy18 1MJHLXzUwkctfNTCx8K5Uctzo2PRuWE8XguftfBZC5+18FkLn7XwWQuftfBZC5+18FkLn7XwWQuf tfBZC5+18NkAnw3w2QCfDfDZAJ8N8NkAnw3w2QCfDfDZAJ8N8NkAnw3w2QCfDfDZAJ8N8NkAnw3w 2QCfDfDZAJ8N8NkAnw3w2QCfDfDZAJ8N8NkAnw3w2QCfDfDRAB8N8NAADw3w0AAPDfDQAA8N8NAA Dw3w0AAPDfDQAA8N8NAAD83w0AwPzfDQDA/N8NAMD83w0AwPzfDQDA/N8NAMD83w0AwPzfDQDA/N 8NAMD83w0AwPzfDQDA/N8NAMD83w0AwPzfDQDA/N8NAMD83w0AwPzfDQDA/N8NCMuWiGj2b4aIaP Zvhoho9m+GiGj2b4aIaPZvhoho9m+GiGj2b46ICPDvjogI8O+OiAjw746ICPDvjogI8O+OiAjw74 WDg3Onhu9C46N4zHO+CzAz474LMDPjvgswM+O+CzAz474LMDPjvgswM+O+CzAz474LMDPjvgswM+ O+CzAz474LMDPjvgswM+O+CzAz474LMDPjvgswM+O+CzAz474LMDPjvgswM+O+CzAz674bMbPrvh sxs+u+GzGz674bMbPrvhsxs+u+GzGz664aMbPrrhoxs+uuGjGz664aMbPrrhoxs+uuGjGz664aMb Prrhoxs+uuGjGz664aMbPrrhoxs+uuGjGz664aMbHrrhoRseuuGhGx664aEbHrrhoRseuuGhGx66 4aEXHnrhoRceeuGhFx564aEXHnrhoRceeuGhFx564aEXHnrhoRceeuGhFx564aEXHnrhoRceeuGh Fx564aEXHnrhoRceeuGhFx564aEXHnrhoRceeuGhFx564aEXHnoxF73w0QsfC+dGL8+NkUXnhvF4 L3z2wmcvfPbCZy989sJnL3z2wmcvfPbCZz989sNnP3z2w2c/fPbDZz989sNnP3z2w2c/fPbDZz98 9sNnP3z2w2c/fPbDZz989sNnP3z2w2c/fPbDZz989sNnP3z2w2c/fPbDZz989sNnP3z2w2c/fPbD Zz989sNnP3z2w0c/fPTDRz989MNHP3z0w0c/fPTDRz989MPHEHwMwccQfAzBxxB8DMHHEHwMwccQ fAzBxxB8DMHHEHwMwccQfAzBxxB8DMHHEHwMwccQfAzBxxB8DMHHEHwMwccQfAzBxxB8DMHHEHwM wccQfAzBwxA8DMHDEDwMwcMQPAzBwxA8DMHDEDwMwcMQPAzBwxA8DMHDEDyMwMMIPIzAwwg8jMDD CDyMwMMIPIzAwwg8jMDDCDyMwMMIPIzAwwg8jMDDCDwsnBsjPDdmFp0bxuMj8DgCjyPwOAKPI/A4 Ao8j8DgCjyPwOAKPI/A4Ao8j8DgCjyPwOIK5GoHPEfgcgc8R+ByBzxH4HIHPEfgcgc8R+ByBzxH4 HIHPEfgcgc8R+JyAzwn4nIDPCficgM8J+JyAzwn4nIDPCficgM8J+JyAzwn4nIDPCficgM8J+JiA jwn4mICPCfiYgI8J+JiAjwn4mICPCfiYgI8J+JiAjwn4mICPCfiYgI8J+JiAjwn4mICPCfiYgI8J +JiAjwn4mICPCfiYgI8J+JiAjwn4mIGPGfiYgY8Z+JiBjxn4mIGPGfiYgY8Z+JiBjxn4mIGPGfiY gY8Z+JiBjxn4mIGPGfiYgY8Z+JiBjxn4mIGPGfiYgY8Z+JiBjxn4mIGPGfiYgYcZeJiBhxl4mIGH GXiYgYcZeJiBhxl4mIGHGXiYgYeFn8c5s+jcOMLPfYzHZ+BxBh5n4HEGHr/uZ0Z//KzzDj//+U4/ L2MlP+d0G59zzLjOb57/WdQFK/jM7JtLPj9brJNNnZ+JOmep8w+izinqnBF1UqhjEXUSqfNjUecH 1HlV1ImlzvdFnUPUeVnU2UmdPaLOVursEHUaTfM6TSZNx0Udt6hTSZ0V3Svw0Py9Au+JY57nmBdE nbepUybqFFOnRNTJp06hqJNLnTxRx0qdbFHnNHXOijonqXNK1EmiToqoE0+dRFHnGHXiRJ3D1IgV dfZT55Cos5s6+0Sd7dTZKerEUGeroJPAbHJczCYJzCbHxWySwGxyXMwmCcwmx8VsksBsclzMJgnM JsfFbJLAbHJczCYJzCbHxWySwGxyXMwmCcwmx8VsksBsclzMJobOQ9R5RNQxMoWh8574Pj1PnQui ztvUKRN1iqlTIurkU6dQ1MmlTp6oY6VOtqhzmjpnRZ2T1Dkl6iRRJ0XUiadOoqhzjDpxos5h6sSK Ovupc0jU2U2dfaLOdursFHViqKPkhBTmBIuYE1KYEyxiTkhhTrCIOSGFOcEi5oQU5gSLmBNSmBMs Yk5IYU6wiDkhhfuFRdwvUrhfWMT9IoX7hUXcL1K4X1jE/SKF+4VF3C9SuF9YxP0ihfuFRdwvUrhf WMT9IoX7hUXcL1K4X1jE/SKF+4VF3C9SuF9YxP0ihfuFRdwvUrhfWMT9IoX7hUXcL9K5X2SI+0U6 94sMcb9I536RIe4X6dwvMsT9Ip37RYa4X6ST4zJEjksnx2WIHJfO9TlDXJ/TuT5niOtzOtfnDHF9 Tuf6nCGuz+lcnzPE9Tmd63OGuD6nc33OENfndK7PGeL6nM71OUNcn9O5PmeI63M61+cMcX1O5/qc Ia7P6VyfM8T12cr1OVtcn61cn7PF9dnK9TlbXJ+tXJ+zxfXZyvycLeZn68IxEtcfK9efbHH9sXL9 yRbXHyvXn2xx/bFy/ckW1x/rwufj4vpj5fqTLa4/Vq4/2eL6Y+X6ky2uP1auP9ni+mPl+pMtrj82 rj/54vpj4/qTL64/Nq4/+eL6Y+P6ky+uPzbmw3wxH9q4juWL65ht4e+fiTnTxlyXL+Y6G8/FfHFd tXE9zBfWwxXdKzB5z+17BRbGTOKYKWLv8dRJFHUWf16v6BymTqyos586h0Sd3dTZJ+psp85OUSeG OsraWcS10y6unUVc8+zimlfENc8urnlFXKvs4lpVxPeXXXx/FfH9ZRffX0V8f9nF91cR31928f1V xPeXXXx/FfH9ZRffX0V8f9nF91cR31928f1VyvfXOfH9Vcrz55x4/pRy3s+J817KeT8nznsp5/2c OO+lnPdz4ryXct7PifPu4Lw7xXl3cD10iuuhg+uYU1zHHDwPneJ56OB56BTPQwfPQ6d4Hjp4HjrF 89DB89ApnocOnodO8Tys5nnoEs/Dau7vLnF/r+b57BLP52oyjUtkmmpeq3aJ16qrl9w7rOgcpk6s qLOfOodEnd3U2SfqbKfOTlEnhjrK+8LN90W9+L5w8zP2evEzdjfnq16cLzfnq16cLzfnq16cLzfn q16cr0bOV5M4X41cx5rEdayR1zCaxGsYjdyXm8R9uZH7cpO4Lzdy/WkS159Gnj9N4vnTyPOnSTx/ Wjnv7eK8t3LfaRf3nVYe53bxOLfyOLeLx7mVx7ldPM5XeA31qngN9Qrn66o4X1eYN66KeeMK1+er 4vp8hfN1VZyvK5yvq+J8ebiuesV11cN594rz7uE65hXXMQ/fX17x/eXhvHvFefdwX/aK+7KH549X PH88PH+84vlzg+ePTzx/bvD88Ynnzw3Ou0+c9xucd5847zc47z5x3m9wvnzifPk5XzfF+fJzvm6K 8+XnOn9TXOf9XOdviuu8nxx3U+Q4P8/Dm+J56Of5c1M8f/xcN26K64af5+FN8TwM8PwJiudPgHks KOaxAPedoLjvBDjvQXHeA1x/guL6E+D5ExTPnwDPn6B4/gR4/gTF82eM61hIXMfGeB6GxPNwjOtY SFzHxriOhcR1bIzvi5D4vhjj9Y2QeH1jjJ/HhcTP48b4GXtIvOdojNwdErl7jPtFSNwvxrhuhO5i 3VjRvQK+e77wbxAsHXM3x9wn9r6dOjtFnRjqKGtDmGvDlLg2hLk2TIlrQ5hrw5S4NoS5NkyJa0OY a8OUuDaEuTZMiWtDmGvDlLg2hHlNbkq8JhfmfUVT4n1FYa5VU+JaFeYaMyWuMWGuMVNiNgkzm0yJ 2STM9WdKXH/CXH+mxPUnzPVnSlx/Ilx/bonrT4Trzy1x/Ylw/bklrj8Rrj+3xPUnwnXjlrhuRLhu 3BLXjQjXjVviuhHhZ4m3xM8SI1w3bonrRoQZ55aYcSJko1siG0W4jt0S17EI17Fb4joW4Tp2S1zH IlzHbonrWITr2C1xHYtwHbslrmN/4DpmNmvr2B+4jpnN2jr2B65jhs4/iDqnqHNG1Emkzo9FnR9Q 51VR5xB1XhZ1dlJnj6hj5ChDJ088D09T56yok0SdFFEnnjqJos4x6sSJOoepEyvq7KfOIVFnN3X2 iTrbqbNT1ImhjrIefsM8vx5+S1wPDZ1s6vxM1DlLnX8QdU5R54yok0Idi6iTSJ0fizo/oM6rok4s db4v6hyizsuizj7qHBB1dlJnj6izlTo7RJ1HqfOEqGP8vQhDp1B8v5+mzllR5yR1Tok6SdRJEXXi qZMo6hyjTpyoc5g6saLOfuocEnV2U2efqLOdOjtFnRjqKPvXOu5fG8T9ax33rw3i/rWO+9cGcf9a x/1rg7h/reP+tUHcv9Zx/9og7l/ruH9tEPevddy/Noj71zruXxvE/Wsd968N4v5l6GyizlOizqPU eULUMX4+o6Hznvg+zadOoahzmjpnRZ2T1Dkl6iRRJ0XUiadOoqhzjDpxos5h6sSKOvupc0jU2U2d faLOdursFHViqKPsX49z/9ok7l+Pc//aJO5fj3P/2iTuX49z/9ok7l+Pc//aJO5fj3Pf2STuO49z 39kk7juPc9/ZJO47j3Pf2STuO4+TmzaJ3PQ4uWmTyE2GztPUeeYOOiu6V+Dy/L0CKx3zUY75hNi7 cY1wk3iN0NCxUidb1DlJnVOiThJ1UkSdeOokijrHqBMn6hymTqyos586h0Sd3dTZJ+psp85OUSeG Ospet5l7XYy4123mXhcj7nWbudfFiHvdZu51MeJet5l7XYy4120mq8WIrLaZe2aMuGdu5p4ZI+6Z m7lnxoh75mbumTHinrmZe2aMuGdu5p4ZI+6Zm7lnxnzJnrlSnTnTvM69oo6LOm5x3aikznuiTjF1 SkSdfOoUijq51MkTdazUyRZ1TlPnrKhzkjqnRJ0k6qSIOvHUSRR1jlEnTtQ5TJ1YUWc/dQ6JOrup s0/U2U6dnaJODHWUvLGVeWOHmDe2Mm/sEPPGVuaNHWLe2Mq8sUPMG1uZN3aIeWMr88YOMW9sZd7Y IeaNrcwbO8S8sZV5Y4eYN7Yyb+wQ88ZW5o0dYt7YyryxQ8wbW5k3doh5Y/H1gqdEnYeo84ioY9xn Zuj8Xlx/BqkTEHXOU+eCqFNMnRJRJ586haJOLnXyRB0rdbJFndPUOSvqnKTOKVEniTopok48dRJF nWPUiRN1DlMnVtTZT51Dos5u6uwTdbZTZ6eoE0MdJf/sYv7ZK+afXcw/e8X8s4v5Z6+Yf3Yx/+wV 888u5p+9Yv7ZxfyzV8w/u5h/9or5Zxfzz14x/+xi/tkr5p9dzD97xfyzi/lnr5h/djH/7BXzzy7m n71i/tnF/LNXzD+7+PnFXvHzi13MUXvFHGXo3E+dB0Ud43rUXvF61C7mur1irjN0wtSZEnU+os4n os4H1PlQ1KmjzvuiTiV13hN1zlPngqhTTJ0SUSefOoWiTi518kQdK3WyRZ3T1Dkr6pykzilRJ4k6 KaJOPHUSRZ1j1IkTdQ5TJ1bU2U+dQ6LObursE3W2U2enqBNDHSX3HmDufWlR7l3RvQLvzt8r8HXH zOaYSkY+wIz8kpiRDzAjvyRm5APMyC+JGfkAM/JLYkY+wIz8kpiRDzAjvyRm5APMyC+JGfkAM/JL YkY+wIz8kpiRDzAjvyRm5APMyC+JGfkAM/JLYkY+wIz8kpiRDzAjvyRm5APMti+J2dbQmaHOrLh2 fkqdz0SdQeoERJ1+6gyIOlepc03U+YA6H4o6l6nTJurUUed9Uec8dS6IOm9Tp0zUKaZOiaiTT51C UcdKnWxR5zR1zoo6J6lzStRJok6KqBNPnURR5xh14kSdw9SJFXX2U+eQqLObOvtEne3U2SnqxFBH ydpHmLWPiteYjzA/HxXz8xHm56Nifj7C/HxUzM9HmJ+Pivn5CPPzUTE/H2F+Pirm5yPMz0fF/HyE +fmomJ+PMD8fFfPzEebeo2LuPcLce1TMvUeYe4+KufcIc+9RMfceYe49KubeI7w2fFS8NnyE+fmo mJ+P8NrwUfHa8BHm8KNiDj/Ca8xHxWvMR5jnj4p5/givVR8Vr1UfIRccFbnA0LlBHZ+oc5U610Sd y9RpE3UaqdMk6tRR531Rp5I674k656lzQdR5mzplok4xdUpEnXzqFIo6udTJE3Ws1MkWdU5T56yo c5I6p0SdJOqkiDrx1EkUdY5RJ07UOUydWFFnP3UOiTq7qbNP1NlOnZ2iTgx1FE6JI6ckiJwSR05J EDkljpySIHJKHDklQeSUOHJKgsgpceSUBJFT4sgpCSKnxJFTEkROiSOnJIicEkdOSRA5JY7X+RPE 6/xx5J0EkXfiyDsJIu/EkXcSRN6JI+8kiLwTR95JEHknjryTIPJOHHknQeSdOPJOgsg7ceSUBJFT 4sgpCSKnxJEvEkS+iOPnDgni5w5x5JQEkVMMnR7q9Io6V6lzTdS5TJ02UcdFHbeoU0md90SdFd0r kD1/r8B5jnlBHPNt6pSJOsXUKRF18qlTKOrkUidP1LFSJ1vUOU2ds6LOSeqcEnWSqJMi6sRTJ1HU OUadOFHnMHViRZ391Dkk6uymzj5RZzt1doo6MdRRmCaJTJMiMk0SmSZFZJokMk2KyDRJZJoUkWmS qJEiMk0SmSZFZJokMk2KyDRJZJoUkWmSyDQpItMkkWlSRKZJItOkiEyTZP7i321VdJ6mzjOizibq PCXqPEqdJ0Sdh6jziKhzP3UeFHUMpkkRmSaJTJMiMk0SP8NJET/DSTJ/8d8lUnQ+pc5nos5H1PlE 1OmnzoCoc5U610SdD6jzoahzmTptok4ddd4XdVzUcYs6ldR5T9Q5T50Los7b1CkTdYqpUyLq5FOn UNTJpU6eqGOlTraoc5o6Z0Wdk9Q5JeokUSdF1ImnTqKoc4w6caLOYerEijr7qXNI1NlNnX2iznbq 7BR1Yqij8I6FvJMm8o6FvJMm8o6FvJMm8o6FvJMm8o6FWSxN5B0LeSdN5B0LeSdN5B0LeSdN5B0L eSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN 5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LeSdN5B0LP1NKEz9TspCb0kRusvAzpTTxMyULP1NK Ez9TspDj0kSOs5Dj0kSOs/AeujTxHjoLOS5N5DgLOS5N5DgLOS5N5DgLOS5N5DgLOS5N5DgLOS5N 5DgLOS5N5DgL+StN5C8L+StN5C8L+StN5C8L+StN5C8L+StN5C8L+StN5C8L+StN5C8L+StN5C8L +StN5C8L+StN5C8L+StN5K8M8pdV5K8M8pdV5K8M8pdV5K8M8pf1a/DXiu4ViJ+/V2DpmCkc0yL2 nkidH4s6P6DOq6JOLHW+L+ocos7Los4+6hwQdXZSZ4+os5U6O0SdTdR5StR5lDpPiDoPUecRUed+ 6jwo6hiMZRUZK4OMZRUZK4OMZRUZK4OMZRUZK4OMZRUZK4OMZRUZK4OMZRUZK4OMZRUZK4OMZRUZ K4OMZRUZK4OMZRUZK4OMZRUZK4OflVnFz8oyyGpWkdUyyGpWkdUyyGpWkdUyyGpWkdUyyGpWkdUy yGpWkdUyyGpWkdUyyGpWkdUy+JmbVfzMLYPMZxWZL4PMZxWZL4PMZxWZL4PMZxWZL4PMZxWZL4PM ZxWZL4PMZxWZL4PMZxWZL4PMZxWZL4PMZxWZL4PMZxWZL4fMZxOZL4fMZxOZL4fMZxOZL4fMZxM/ c8shx9lEjsshx9lEjsshx9lEjsshx9lEjsshx9lEjsshx9lEjsshx9lEjsshx9lEjsvhZ2428TO3 HPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKg TeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHPKgTeTBHHKcTeS4HHKcTeS4HHKcTeS4 HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKc TeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4HHKcTeS4fHJcochx+eS4QpHj 8slxhSLH5ZPjCkWOyyfHFYocl0+OKxQ5Lp8cVyhyXD45rlDkuHxyXKHIcfnkuEKR4/LJcYUix+WT 4wpFjssnxxWKHJdPjisUOS6fHFcoclw+Oa5Q5Lh8clyhyHH55LhCkePyyV+FIn/lk78KV8BfK7pX 4LnP/w2CLxvzU475mdj7R9T5RNQZpE5A1OmnzoCoc4M6PlGnhzq9os5V6lwTdT6gzoeiTiN1mkSd Ouq8L+q4qOMWdSqp856oc546F0Sdt6lTJuoUU6dE1MmnTqGok0udPFHHSp1sUec0dc6KOiepc0rU SaJOiqgTT51EUecYdeJEncPUiRV19lPnkKizmzr7RJ3t1Nkp6sRQR2E1O1mtVGQ1O1mtVGQ1O1mt VGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1 O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1O1mtVGQ1Oz9zKxU/c7OT+UpF5rOT+UrFz9zs 5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhS kePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5LhSkePs5K9Skb/s5K9Skb/s 5K9Skb/s5K9Skb/s5K9Skb/s5K9Skb/s5K9Skb/s5K9Skb/s5K9Skb/s5K9Skb/KyF8Okb/KyF8O kb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/K yF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyF8Okb/KyE0O kZvKyE0OkZvKyE0OkZvKyE0OkZvKyE0OkZvKyE0OkZvKyDsOkXfK+HfOHOLfOSsjNzlEbiojNzlE biojNzlEbiojNzlEbiojNzlEbiojNzlEbiojNzlEbiojNzlEbiojNzlEbirj51+OZT7/WtG9Ag/N 3ytwt2Oe5phnxd5PUueUqJNEnRRRJ546iaLOMerEiTqHqRMr6uynziFRZzd19ok626mzU9SJoY7C ak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6y WpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXIak6yWpXI ak6yWpXIak5+VlYlflbmJPNVicznJPNVicznJPNVicznJPNVicznJKtViazm5GdcVeJnXE4yX5XI fE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4yX5XIfE4y X5XIfE4yX5V4z6OTHFclcpyTHFclcpyTHFclcpyTHFclcpyTHFclcpyTHFclcpyTHFclcpyTHFcl cpyTHFclcpyTHFclcpyLHOcWOc5FjnOLHOcix7lFjnOR49wix7nIcW6R41zkOLfIcS5ynFvkOBc5 zi1ynIsc5xY5zkWOc4sc5yLHuUWOc5Hj3CLHuchxbpHjXOQ4t8hxLnKcW+Q4FznOLXKcixznFjnO RY5zixznIse5RY5zkePcIse5yHFukeNc5Di3yHEucpxb5DgXOc4tcpyLHOcWOc7Fz+7c4md3LvKg W+RBF3nQLfKgizzoFnnQRR50izzoIg+6RR50kQfdIg+6yINukQdd5EG3yIMu8qBb5EEXedAt8qCL POgWedBFHnSLPOgiD7pFHnSRB90iD7rIg26RB13kQbfIgy7yoFvkQRd50C3yoIs86BZ50EUedIs8 6CIPukUedJEH3SIPusiDbpEHXeRBt8iDF8mDjSIPXiQPNoo8eJE82Cjy4EXyYKPIgxfJg4bOiu4V mDTfvlfAIo6ZyDF/LOr8gDqvijqx1Pm+qHOIOi+LOvuoc0DU2UmdPaLOVursEHWeps4zos4m6jwl 6jxKnSdEnYeo84iocz91HhR1DHZsFNnxItmxUWTHi2THRpEdL5IdG0V2vEh2bBTZ8SLZsVFkx4tk x0aRHS+SHRtFdrxIdmwU2fEi2bFRZMeLZMdGkR0vkh0bRXa8SHZsFNnxItmxUWTHi2THRpEdL5Id G0V2vEh2bBTZ8SLZsVFkx4tkvkaR+S6S+RpF5rtI5msUme8ima9RZL6LZL5GkfkukvkaRea7SOZr FJnvIpmvUWS+i2S+RpH5LpL5GkXmu0jmaxSZ7yKZr1FkvotkvkaR+S6S+RpF5msi87WIzNdE5msR ma+JzNciMl8Tma9FZL4mMl+L+BlgEzmuReS4JnJci8hxTeS4FpHjmshxLSLHNZHjWkSOayLHtYgc 10SOaxE5rokc1yJyXBM5rkXkuCZyXIvIcU3kuBaR45rIcS0ixzWR41pEjmsix7WIHNdEjmsROa6J HNciclwTOa5F5LgmclyLyHFN5LgWkeOayHEtIsc1keNaRI5rIse1iBzXRI5rETmuiRzXInJcEzmu ReS4JnJci8hxTeS4FpHjmshxLSLHNZHjWkSOayLHtYgc18TPAFvEzwCbyIMtIg82kQdbRB5sIg+2 iDzYRB5sEXmwiTzYIvJgE3mwReTBJvJgi8iDTeTBFpEHm8iDLSIPNpEHW0QebCIPtog82EQebBF5 sIk82CLyYBN5sEXkwXby4BWRB9vJg1dEHmwnD14RebCdPHhF5MF28uAVkQfbyYNXRB5sJw9eEXmw nTx4ReTBdvLgFZEH28mDV0QebCcPXhF5sJ08eEXkwXby4BWRB9vJg1dEHmwnD14RebCdPHhF5MF2 8uAVkQfbyYNXRB5sJw9eEXhwRfcK+ObvFVgYc4ZjzopreZg6U6LOp9T5TNT5iDqfiDqD1AmIOv3U GRB1blDHJ+r0UKdX1LlKnWuizgfU+VDUuUydNlGnkTpNok4ddd4XdVzUcYs6ldR5T9Q5T50Los7b 1CkTdYqpUyLq5FOnUNTJpU6eqGOlTraoc5o6Z0Wdk9Q5JeokUSdF1ImnTqKoc4w6caLOYerEijr7 qXNI1NlNnX2iznbq7BR1YqijsGMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX 2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEjsmMX2dEj smMX2dEjsmMX2dEjsmMX2dEjfpbYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzY RR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70iDzYRR70 iDzYRR70iDzYRR70iDzYRY7ziBzXRY7ziBzXRY7ziBzXRY7ziBzXRY7ziBzXRY7ziBzXRY7ziBzX RY7ziBzXRY7ziBzXRY7ziBzXRY7ziBzXRY7ziBznJcddFznOS467LnKclxx3XeQ4LznuushxXnLc dZHjvOS46yLHeclx10WO85Ljrosc5yXHXRc5zkuOuy5ynJccd13kOC857rrIcV5y3HWR47zkuOsi x3nJcddFjvOS466LHOclx10XOc5LjrsucpyXHHdd5DgvOe66yHFectx1keO85LjrIsd5yXHXRY7z kuOuixznJcddFznOS467LnKclxx3XeQ4LznuushxXnLcdZHjvOS46yLHeclx10WO85Ljrosc5yXH XRc5zkuOu34XHLeiewUum7/wbxAsHfM8x7wg9v42dcpEnWLqlIg6+dQpFHVyqZMn6lipky3qnKbO WVHnJHVOiTpJ1EkRdeKpkyjqHKNOnKhzmDqxos5+6hwSdXZTZ5+os506O0WdGOoo7OgjO/pFdvSR Hf0iO/rIjn6RHX1kR7/Ijj6yo19kRx/Z0S+yo4/s6BfZ0Ud29Ivs6CM7+kV29JEd/SI7+siOfpEd fWRHv8iOPrKjX2RHH9nRL7Kjj+zoF9nRR3b0i+zoIzv6RXb0kR39Ijv6yI5+kR19ZEe/yI4+sqNf ZEcf2dEvsqOP7OgX2dFHdvSL7OgjO/pFdvSRHf0iO/rIjn6RHX1kR7/Ijj6yo19kRx/Z0S+yo4/s 6BfZ0Ud29Ivs6CM7+kV29JEd/eJngD7yoF/kQR950C/yoI886Bd50Ece9Is86CMP+kUe9JEH/SIP +siDfpEHfeRBv8iDPvKgX+RBH3nQL/KgjzzoF3nQRx70izzoIw/6RR70kQf9Ig/6yIN+kQd95EG/ yIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD 5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGA yIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD 5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5MGAyIOD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiA yHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD5LiAyHGD 5LjAl3Dciu4VeHf+XoGVjhnDMRXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXm C5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5 RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXm C5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC5L5RkXmC/IzwFHxM8Ag 2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFU ZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFUZMcg2XFU/AwwSB4cFXkwRB4MizwYIg+GRR4MkQfDIg+G yINhkQdD5MGwyIMh8mBY5MEQeTAs8mCIPBgWeTBEHgyLPBgiD4ZFHgyRB8MiD4bIg2GRB0PkwbDI gyHyYFjkwRB5MCzyYIg8GBZ5MEQeDIs8GCIPhkUeDJEHwyIPhsiDYZEHQ+TBsMiDIfJgWOTBEHkw LPJgiDwYFnkwRB4MizwYIg+GRR4MkQfDIg+GyINhkQdD5MGwyIMh8mBY5MEQeTAs8mCIPBgWeTBE HgyLPBgiD4ZFHgyRB8MiD4bIg2GRB0PkwbDIgyHyYFjkwRB5MCzyYIg8GBZ5MEQeDIs8GCIPhkUe DJEHwyIPhsiDYZEHQ+TBsMiDIfJgWOTBEHkwLPJgiDwYFnkwRB4MizwYIg+GRR6cJg9GRB6cJg9G RB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6cJg9GRB6c Jg9GRB6cJg9GFvHgiu4VyJ6/V+DrjrmJYz4l9v4odZ4QdR6iziOizv3UeVDUMdgxIrLjNNkxIrLj NNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkxIrLjNNkx IrLjNNkxIrLjNNkxIrLjNNkxIrLjNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLf NJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkvIjLfNJkv IjLfNJkv8hXMZ8yBcRy/i/rGfJy8nS+NQvQ0vcLv/R7feNP8p19vXTS2cZ6+wD+/wIy68Ps9i3Lr fKL6/Nfi/hf+bDz/71F/fc/8mDZUFuoERN663dcL5ldMT6IeQz2M+ibKjPotHougJlG/Rn2MGkL5 UX0oL+pXqF+i6nEuf8ecCJ1E6CRCIxEaidBIhEYiNBKhkQiNRGgkQiMRGonQSIRGIjQSTa2oSyhD q9Z0wnTUfML0PdQLqF2obajvoJ5EPYZ6GPVNlBn1Wzw/gppE/Rr1MWoI5Uf1obyoX6F+iWpFXULV o4xxqlDvms6YjpvPmH6I2oXahvoO6knUY6iHUd9EmVG/xXMjqEnUr1Efo4ZQflQfyov6FeqXqFbU JVQ9qhZVhTLGegdzcMKchTGz4DMLPrPgMwtjZ2HsLIydhbGzMHYWxs7C2FkYOwtjZ2HsLIydhbGz MHYWxs7C2FkYOwtjZ2HsLIydhbGzMHYWxs7C2FkYOwtjZ90e244qQNn+eG4Y50QyqgL1Bs8N4/E8 9JaH3vLQWx56y0NveegtD73lobc89JaH3vLQWx56y0NveegtD73lobc89JaH3vLQWx56y0Nveegt D73lobc89JaH3vLQWx56y0Nveegtz/Q69sc3zMU4RsU4RsXooxh9FKOPYvRRjD6K0Ucx+ihGH8Xo oxh9FKOPYvRRjD6K0Ucx+ihGH8Xooxh9FKOPYvRRjD6K0Ucx+ihGH8Xooxh9FKOPYvRRjD6K0Ucx jkExjkTx7X5eM/3c9FPzz03pqBOo46gfoo6ivod6AbULtQ31HdSTqMdQD6O+iTKjfguNCGoS9WvU x6ghlB/Vh/KifoX6JaoVdQlVj6pFVaHeRb2DsqMKUDbU6yijt2TM3xvmCvRWgd4q0FsFeqtAbxXo rQK9VaC3CvRWgd4q0FsFeqtAbxXorQK9VaC3CvRWgd4Wzg3jnDiCurTo3DAer0DvFei9Ar1XoPcK 9F6B3ivQewV6r8ArKtB7BXqvQO8V6L0CvVeg9wr0XoHeK3BsK9B/BfqvwEpSgxFq0HsNeq9B7zXo uwZ916DvGvRdg75r0HcN+q5B3zXouwZ916DvGvRdg75q0FcN+qpBXzXoqwZ91aCvGvRVg75q0FcN +qpBXzXoqwZ91aCvGvRVg75q0FcN+qpBXzXoqwbHtcaUgDz7hrkO81+H+a/DMa5Dj3XosQ7Htw59 1qHPOvRZhz7r0Gcd+qxDn3Xosw591qHPOvRZhz7r0Gcd+qxDn3Xosw591qHPOvRZhz7r0Gcd+qxD n3Xosw591qHPOvRZhz7r0Gcd+qxDn3Xosw591qHPutt9voI6gle/Yb6EXrHaoY6jfoj6HuoF1C7U NtR3UE+iHkM9jPomyoz6LV4fQU2ifo36GDWE8qP6/nhuGOfE86Z5psrnuWE8fgleLsHLJXi5BC+X bv9Xj6pFVaHeRb2DsqMKULbbq9UleLkEL5fgoQ39t6H/NvTeht7bcKzb0H8b+m9D/23ovw39t6H/ NvTfhv7b0H8b+m9D/23ovw39t6H/NvTfhv7b0H8b+mtDf23orw39taG/NozYhv7a0F8b+mtDf23o rw39taG/NvTXhv7a0F8b+mu73d+LqOdNnXDeiR470WMneuxEj53osRM9dqLHTvTYiR470WMneuxE j53osRM9dqLHTvTYiR470WMneuxEj53osRM9dqLHTvTYiR470WMneuxEj53osRM9dqLHTvTYiR47 0WMneuxEj53osRPnQyfOh0702onzoRO99uB49uB86EG/Pei3B/32oN8e9NuDfnvQbw/67UG/Pei3 B/32oN8e9NuDfnvQbw/67UG/Pei3B/32oN8e9NuDfnvQ78K5YZwTz6GGF50bxuM98NMDPz3w0wM/ PfDTAz898NMDPz3w0wM/PfDTAz898NMDPz3w03M7fZww96H/PvTfh/770H8f+u9D/33ovw/996H/ PvTfh/770H8f+u9D/33ovw/996H/PvTfh/770H/f7f+8qF+hfolqvX0W96G/PvTXh/760F8f+utD f33orw/99d3eeV9HvYZKRiXcTkd9cD6APgfQ5wD6HECfA+hzAH0OoM8B9DmAPgfQ5wD6HECfA+hz AH0OoM8B9DmAPgfQ5wD6HECfA+hzAMoD6HMAfQ6gzwH0OYA+B9DnAPocQJ8D6HMAfQ6gzwH0OYA+ B9DnAPocQJ8D6HMAfQ7g/B3GzAxjTRvGeTGMXofR6zB6HUavw+h1GL0Oo9dh9DqMXofR6zB6HUav w+h1GL0Oo9dh9DqMXofR6zB6HUavw+h1GL0Oo9dh9DqMXhfOjWFm4rlFWdR4fBhehuFlGF6G4WUY XobhZRhehuFlGF6G4WUYPoahMI4ex9HjOHocR4/j6HEcPY6jx3H0OI4ex9HjOHocR4/j6HEcPY6j x3H0OI4ex9HjOHocR4/j6GEcPYyjh3H0MI4extHDOHoYRw/j6GEcPYyjh3Ecz0kcp0n0MIkeJjH+ JMafxPiTGH8S409i/EmMP4nxJ2//92vUx6ghlP/2mTiJ8Scx/iTGn8T4k7dX3XpULaoK9S7qHZQd VXA7/U1i7FkcuVmMPQvvsxh7FmPPYuxZjD2LsWcx9izGnsXYs3jFLMaexdizGHsWY89i7FmMPYux ZzH2LMaexdizGHsWY89i7FmMPYuxZzH2LMaehfdZjD+L8ecw7hzGncO4cxh3DuPOYdw5jDuHcecw 7hzGncO4cxh3DuPOYdw5jDuHcecw7hzGncO4cxjXODcu8Trxa6b5zxK+Y3yWdPs8uYSqR9WiqlDv ot5B2VEFqCzUa6YHzN8xP2B+EvUw6psoM+q3+H4ENYn6Nepj1BDKj+pDeVG/Qv0S1Yq6hKpH1aKq UO+i3kHZUTbTeoyxHvrrob8e+uuhvx7666G/Hvrrob8e+uuhvx7666G/Hvrrob8e+uuhvx7666G/ HtrrobsRPW+E3kbobYTeRuhthN5G6G2E3kbobYTeRuhthN5G6G2E3kbobYTeRuhthN5G9LsRmhuh uQV6W6C1BVpboLUFWlugtQVaW6C1BVpboLUFWlugtQVaW6C1BV634fnb8NxteN42PGcbnrMNz9mD P39dBv4jtdrna+nXW5fhV4PL/4Ysft/nr7hnKe8u8LpxTWQfr4kon2kZOpuo85So8yh1nhB1HqLO I6LO/dR5UNQx3qeGzr2izi3qKJ/nGDoz1JkVdcLUmRJ1PqXOZ6LOR9T5RNQZpE5A1OmnzoCoc4M6 PlGnhzq9os5V6lwTdT6gzoeiTh11lM9PEkzzn4UfN2n3RidwPTwurocJXA+Pi+thAtex4+I6lsB1 7PiidWxF9wrEf/17BRK45h0X17wErnnHxTUvgWvecXHNS+Cad1xc8xK45h0X17wErnnHxTUvgWve cXHNS+Cad1xc8xK45h0X17wErnnHxTUvgWvecXHNM3QuU6dN1GmkTpOoU0cddQ12UUe5/9n43NG4 f8xi0u4fM3QOUedlUWcfdQ6IOjups0fU2UqdHaLO09R5RtTZRJ2nRJ1HqfOEqPMQdR4Rde6nzoOi jrH3WkSGSOF+ahH30xTupxZxP03hfmoR99MU7qcWcT9N4X5qEffTFO6nFnE/TeF+ahH30xTupxZx P03hfmoR99MU7qcWcT9N4X5qEffTFO6nFnE/TeF+ahH30xTupxZxP03hfmqJwn5aSZ33RJ3z1FHu LUs3zd+vnmHS7lc3dH5AnVdFnZ3U2SPqbKXODlHnaeo8I+psos5Tos6j1HlC1HmIOo+IOvdT50FR x9iXM8R9OZ37coa4L6dzX84Q9+V07ssZ4r6czn05Q9yX07kvZ4j7cjr35QxxX07nvpwh7svp3Jcz xH05nftyhrgvp3NfzhD35XTuyxnivpzOfTlD3JfTuS9niPtyOvflDHFfTue+nCHuy+nclzPEfTmd +3JGFPblt6mj3PNt3Bedwvuilb/XZmVOyBZzgpXXE7LF6wlWXk/IFq8nWHk9IVu8nmBlbskWc4uV uSVbzC1W5pZsMbdYmVuyxdxiZW7JFnOLlbklW8wtVuaWbDG3WJlbssXcYmVuyRZzi5W5JVvMLVbm lmwxt1iZW7LF3GJlbskWc4uVuSVbzC1W5pZsMbdYmVuyxdxiZW7JFnOLlbklW8wtVuaWbDG3WJlb ssXcYmVuyRZzi5W5JVvMLVbmlmwxt1iZW7LF3GJlbskWc4uVuSU7CrmlmDolok4+dQpFnVzqKH9X zcbcki/mFhtzS76YW2zMLflibrExt+SLucXG3JIv5hYbc0u+mFtszC35Ym6xMbfki7nFttJ7BZ6b v1fgfo75oDjmHN9H94o6t6jze/E9NEOdWVEnTJ0pUedT6nwm6nxEnU9EnUHqBESdfuoMiDo3qOMT dXqo0yvqXKXONVHnA+p8KOpcpk6bqNNInSZRp44674s6Luq4RZ1K6rwn6pynzgVR523qlIk6xdQp EXXyWYWizmnqKH//vcg0/zOijNutlZ8RVcRrTnbxmlMRrznZxWtORcxudjG7FTG72cXsVsTsZhez WxGzm13MbkXMbnYxuxUxu9nF7FbE7GYXs1sRs5tdzG5FvOZkF/NYEfOYXcxjRcxjdjGPFTGP2cU8 VsQ8ZhfzWBHzmF3MY0XMY3YxjxUxj9nFPFbEPGYX81gR85hdzGNFzGN2MY8VMY/ZxTxWxDxmF/NY EfOYXcxjRcxjdjGPFTGP2cU8VsQ8ZhfzWBHzmF3MY0XMY3YxjxUxj9nFPFbEPGYX81gRM5RdzGNF vOZkF685FfH6ld2k/XykIuZDexTy4UnqKD8fqdQ0/zNEz5m0nyFq6JyizhlRJ4U6FlEnkTo/FnV+ QJ1XRZ1Y6nxf1DlEnZdFnX3UOSDq7KTOHlFnK3V2iDpPU+cZUWcTdZ4SdR6lzhOizkPUeUTUuZ86 D4o6Rn4+J+bnUubnc2J+LmV+Pifm51Lm53Nifi5lfj4n5udS5udzYn4uZX4+J+bnUubnc2J+LmV+ Pifm51Lm53Nifi5lfj4n5udS5udzYn4uZX4+J+bnUubnc2J+LmV+Pifm51Lm53Nifi5lfj4n5udS 5udzYn4uZX4+J+bnUubnc2J+LmV+Pifm51Lm53Nifi5l7j0n5t5S5t5zUci9SdRRfi6og9dpneJ1 Wgdzr1PMvQ7mXqeYex3MvU4x9zqYe51i7nUw9zrF3Otg7nWKudfB3OsUc6+Dudcp5l4Hc69TzL0O 5l6nmHsdzL1OMfc6mHudYu51MPc6v0buXdG9Ag/N3yuwdEwjIzvFjOxgRnaKGdnBjOwUM7KDGdkp ZmQHM7JTzMgOZmSnmJEdzMhOMSM7mJGdYkZ2MCM7xYzsYEZ2ihnZwYzsFDOygxnZKWZkBzOyU8zI DmZkp5iRHczITjEjO5iRnWJGdjAjO8WM7GBGdooZ2cGM7BQzsoMZ2SlmZAczslPMyA5mZKeYkR28 xuwUrzE7mLWdYtZ2MGs7xaxt6ByjjvIz76uZtV1i1q5mRnaJGbmaGdklZuRqZmSXmJGrmW1dYrat ZrZ1idm2mtnWJWbbamZbl5htq5ltXWK2rWa2dYnZtprZ1iVm22pmW5d4TbeaedUl5tVq5lWXmFer mVddYl6tZl51iXm1mnnVJebVauZVl5hXq5lXXWJerWZedYl5tZp51SXm1WrmVZeYV6uZV11iXq1m XnWJebWaedUl5tVq5lWXmFermVddYl6tpoZLzKvVzKsuMa9WM6+6xLxazbzqEvNqNfOqS8yr1cyr LjGvVjOvusS8Wr3k72ApOqepc1bUOUmdU6JOEnVSRJ146ij/ZpSbubdezL1u3utRL97r4ea9HvXi vR5uXvOuF695u5nD68Uc7mYOrxdzuJvXquvFa9Vu5vl6Mc+7mefrxTzvZp6vF/O8m3m+Xszzbub5 ejHPu5nn68U872aerxfzvJt5vl7M827m+Xoxz7uZ5+vFPO9mnq8X87ybeb5ezPNu5vl6Mc+7mefr xTzvZp6vF/O8m3m+Xszzbub5ejHPu5nn68U872aerxfzvJt5vl7M827m+Xoxz7uZ5+vFPO9mnq8X 87ybmaVezPNu5vl6Mc+7mefrxTzvZp6vF/O8m3m+Xszzbub5ejHPu5nn68U872aerxfzvJt5vl7M 827m+Xoxz7uZ5+vFPO9mnq+PQp4/Rp04UecwdZR/A7aRfNEk8kUjuaBJ5IJGckGTyAWNvM7fJF7n byRfNK2AL1Z0r8Ck6Y//BsGXjXmIY74s9r6POgdEnZ3U2SPqbKXODlHn6YV/61zU2USdp0SdR6nz hKjzEHUeEXXup86Dos4cc8m9os4t6vxeXKtmqDMr6oSpMyXqfEqdz0Sdj6jziagzSJ2AqNNPnQFR 5wZ1fKJOD3V6RZ2r1Lkm6nxAnQ9FncvUaRN1GqnTJOrUUeN9UWfh8wm3qFNJnfdEnfPUuSDqvE2d MlGnmDolos7CzyopFHVyqZMn6lipky3qnKbOWVHnJHVOiTrHqKNk9lZm7XYxa7cya7eLWbuVGbld zMitzMjt4jX4Vl6DbxevwbcyP7eL+bmV+bldzM+tzM/tYn5uZX5uF/NzK/Nzu5ifW5mf28X83Mr8 3C7m51bm53YxP7cyP7eL+bmV+bldzM+tzM/tYn5uZX5uF/NzK/Nzu5ifW5mf28X83Mr83C7m51bm 53YxP7cyP7eL+bmV+bldzM+tzM/tYn5uZX5uF/NzK/Nzu5ifW5mf28X83Mo9vl3Mz63Mz+1ifm5l fm4X83Mr83O7mJ9bmZ/bxfzcyvzcLubnVubndjE/tzI/t4v5uZX5uV3Mz63Mz+1ifm5lfm4X83Mr 83O7mJ9bmZ/bo5Cf91PnkKizmzr7BJ0rxnyb59fEArOmk0idH4s6P6DOq6JOLHW+L+ocos7Los4+ 6hwQdXZSZ4+os5U6O0Sdp6nzjKiziTpPiTqPUucJUech6jwi6txPnQdFnTlml3tFnVvU+b24bsxQ Z1bUCVNnStT5lDqfiTofUecTUWeQOgFRp586A6LODer4RJ0e6vSKOldZ10SdD6jzoahzmTptos7i a/SKTh113hd1XNRxizqV1HlP1DlPnQuiztvUKRN1iqlTIurkU6dQ1MmlTp6os/jfcVF0TlNnaQ5f 0b0Cvvl7Be52zJMc85TYexJ1UkSdeOokijrHqBMn6hymTqyos5s6CkN4eC3fK17L9/C+Ga9434yH LOIVWcRDFvGKLOIhi3hFFvGQRbwii3jIIl6RRTxkEa/IIh6yiFdkEQ9ZxCuyiIcs4hVZxEMW8Yos 4iGLeEUW8ZBFvCKLeMgiXpFFPGQRr8giHrKIV2QRD1nEK7KIhyziFVnEQxbxiiziIYt4RRbxkEW8 Iot4yCJekUU8ZBGvyCIe7oFekUU8ZBGvyCIesohXZBEPWcQrsoiHLOIVWcRDFvGKLOIhi3hFFvGQ Rbwii3jIIl6RRTxkEa/IIh6yiFdkEQ9ZxCuyiIcs4hVZxEMW8YqfCXjIF16RLzzkC6/IFx5ygVfk ghvMzz4xP99gfvaJ+fkG87NPzM83mJ99Yn6+wfzsE/PzDeZnn5ifbzA/+8T8fIP52Sfm5xvMzz4x P99gfvaJ+fkG87NPzM83mJ99Yn6+wfzsE/PzDeZnn5ifbzA/+8T8fIP52Sfm5xvMzz4xP99gfvaJ +fkG87NPzM83mJ99Yn6+wfzsE/PzDZZPzM83mJ99Yn5e/LnCNVHnA+p8KOpcpk6bqNNInSZRp446 74s6Cz+byC3qVFLnPVHnPHUuiDpvU6dM1CmmTomos/DvlBaKOrnUyRN1rNTJFnVOU+esqHOSOqdE nSTqpIg68dRJFHWOUUfN4dups1PQ8TPP3xTzvJ95/qaY5/3M8zfFPO9nnr8p5nk/8/xNMc/7medv innezzx/U8zzfub5m2Ke9zPP3xTzvJ95/qaY5/3M8zfFPO9nnr8p5nk/8/xNMc/7medvinnezzx/ U8zzfub5m2Ke9zPP3xTzvJ95/qaY5/3M8zfFPO9nnr8p5nk/8/xNMc/7uWfcFPO8n3n+ppjn/czz N8U872eev7nSnytwef5egQ/FMS9zzDZRp5E6TaJOHXXeF3Vc1HGLOpXUeU/UOU+dC6LO29QpE3WK qVMi6uRTp1DUyaVOnqhjpU62qHOaOmdFnZPUOSXqJFEnRdSJp06iqLOfOsp9+QHeTx8U76c3dM5S 5x9EnVPUOSPqJFLnx6LOD6jzqqgTS53vizqHqPOyqLOPOgdEnZ3U2SPqbKXODlHnaeo8I+psos5T os6j1HlC1HmIOo+IOvdT50FRx2CIoMgQATJEUGSIABkiKDJEgAwRFBkiQIYIigwRIEMERYYIkCGC IkMEyBBBkSECZIigyBABMkRQZIgAGSIoMkSADBEUPxMIkAuCIhcEyAVBkQsC5IKgyAUBckFQ5IIA uSAockGAXBAUuSBALgiKXBAgFwRFLgiQC4IiFwTIBUGRCwLkgqDIBQHm+aCY5wO8dh4Ur52PMUeF xBw1xhwVEnPUGHNUSMxRY8xRITFHjTFHhcQcNcYcFRJz1BhzVEjMUWPMUSExR40xR4XEHDXGHBUS c9QYc1RIzFFjzFEhMUeNMUeFxBw1xhwVEnPUGHNUSMxRY8xRITFHjTFHhcQcNcYcFRJz1BhzVEjM UWPMUSExR40xR4XEHDXGHBUSc9QYc1RIzFFjzFEhMUeNMUeFxBw1xhwVEnPUGHNUSMxRY8xRITFH jTFHhcQcNcYcFRJz1BhzVEjMUWO8vhoSr6+GeT1qSrweFWaOmhJzVJg5akrMUWHmnykx/4SZf6bE /BNm/pkS80+Y+WdKzD9h5p8pMf+EmX+mxPwTZv6ZEvNPmPlnSsw/YdaUmH/CzD9TYv4JM/9Mifkn zPwzJeafMPPPlJh/wsw/U2L+CTP/TIn5Z/HPBrkm6nxAnQ9FncvUaRN1GqnTJOrUUed9UWfh3xd1 izqV1Pm6+WdF9wq8O3+vwMKY5znmBbH3t6lTJuoUU6dE1MmnTqGoY6VOtqhzmjpKxonwvsZb4n2N EWacW2LGifAazy3xGk+EGeeWmHEizDi3xIwTYca5JWacCDPOLTHjRJhxbokZJ8KMc0vMOBFq3BIz ToQZ55aYcSJLcpei8yl1PhN1PqLOJ6LOIHUCok4/dQZEncV/d0rR6aFOr6hzlTrXRJ0PqPOhqHOZ Om2izuJ/N0XRqaPO+6KOizpuUaeSOu+JOuepc0HUeZs6ZaJOMXVKRJ186hSKOrnUyRN1rNTJFnVO U0fJP39g3jCbtbzxB+YNQ+cZUWcTdZ4SdR6lzhOizkPUeUTUuZ86D4o6Rt4wdO4VdW5R5/fi+TND nVlRJ0ydKVHnU+p8Jup8RJ1PRJ1B6gREnX7qDIg6N6jjE3V6qNMr6lylzjVR5wPqfCjqXKZOm6jT SJ0mUaeOOu+LOi7quEWdSuq8J+qcp84FUedt6pSJOsXUKRF18qlTKOpYqZMt6pykjnJPzTfM8/v7 t8T93dDZRJ2nRJ2HqPOIqHM/dR4UdYx9+VvivvwNXpP4lrgvGzoz1JkVdcLUmRJ1PqXOZ6LOR9T5 RNQZpE5A1OmnzoCoc4M6PlGnhzq9os5V6lwTdT6gzoeizmXqtIk6jdRpEnXqqPO+qOOijlvUqaTO e6LOeepcEHXepk6ZqFNMnRJRJ5c6Cr+v4z64QdwH13H/2iDuX+u4f20Q96913L82iPvXOu5fG8T9 ax33rw3i/rWO+9cGcf9ax/1rg7h/reP+tUHcv9Zx/9og7l/ruH9tEPevddy/Noj71zruXxvE/Wsd 968N4v61jvvXBnH/Wsf9a4O4f63j/rVB3L/Wcf/aIO5f67h/bRD3r3XcdzaI+8467hcbxP3icXLK prvglBXdK5D9xX+DYOmY93NMZS1/nGv5JnEtf5xr+SZxLX+ca/kmcS1/3PzFe+4UnU+p85mo8xF1 PhF1BqkTEHX6qTMg6iz+uZmKTg91ekWdq9S5Jup8QJ0PRZ3L1GkTdRqp0yTq1FHnfVHHRR23qFNJ nfdEnfPUuSDqFFNH2RM2c32OEdfnzVxXY8R1dTPX1RhxXd3MdTVGXFc3c12NEdfVzVxXY8R1dTPX 1RhxXd3MdTVGXFc3c12NEdfVzVxXY8R1dTPX1RhxXd3MdTVGXFc3c12NEdfVzVxXY8R1dTOvYcSI 1zC28v21Q3x/beX7Yof4vtjK83mHeD5v5Xm4QzwPF//bn9dEnQ+oo5yHu+hr71f4Mv5u8WHUd43r S/Px/nbeNwooYHqF3/s9vvGm+U+/XtC613TnP3/VGAu/5uzztfTrxVpbF3n9qrEXHxuM+9amRZ62 rmDvu5fHkt+/ZyP/PDc3d89jn3//9s9dSqKusT8nGPNgXuCm6y+b3rr68sJjxvd+aPrirzn++hH+ nD7fz4qO6938+U7HIo/3Dkf7WBi6cfz88MSSY7Hw2Fo7FsZ6XbQKx8LQzTHz30ZcciwWHltrx2Kc cxTtY2HovsS9JH7JsVh4bK0di/Xm+fvio30sDN2/RT2GemHJsVh4bC0ei52rdCy28FjsWuZYbFmD x2Ibr3NF+1gYusY1wO+inlxyLBYeW4vH4v5VOhb/Zpo/Ft9c5lj8m2ntHYuDZNpoHwtD11gjj5jn 751dfCwWHltrx+IV3tO2Glnrn03zWeuXpj/NWv+8gmNx+s98LM7wPq9oHwtD903ea/XOkmOx8Nha OxY/R726CsfC0H2Z9+X9aMmxWHhsrR2LRP6Mgmgfi0Sum3+P37+3ZO1ceOyrjsWZP/OxMK5xnlqF Y2Ho/nvolhpr4JJjsfDYWjsWxvmaZ16d90gmr0vnm//0PZK5Bo9F5yqdF52c+2vLnBeda/S8MK7r ZK/CsTB006B7Hb+/seRYLDy21o6FkXliV4lT/46cenQZTv27NXgsJk2rw2aGrsFf/8X0RTZb0b0C 8fP3ChjHbUFnrR23WdPq8L2h+7f8fGwp3y88ttaOxZlVyiJnmDesy2SRMyvMIq//mY9FxSqtsxVc S6uWWWcrVrjO/rmPRd8qHYs++vUvcyz61vCxOLtKx+InPBY/XeZY/GQNHou5Vcplc8xexj3PS3PZ 3Apz2V/iWMSt0rF4iccifplj8dIaPRaHVulYfJfH4nvLHIu1uI9s4fXraB8LQ/Pfmec/y4xZciwW HluLx2LTKh2LdTwWTy5zLNat0WPx6Codiwd4LB5b5lg8sAaPhXEt/xem1fn84D9wXbho+tPPD/6D ae0diz3kzmgfiz38d1X2m+eZbvGxWHhsLR6Ly6t0LH7BY9G6zLH4xRo8Fgf5961W43Mlw6zxudLD 5j/9XMlkXpvHYm6VPmOb4Wds5mWOxcwaPS8ur9Kx+AWPResynzeuxfdILO+XjfaxiOV9r8a/tTa0 5FgsPLYWj8XVVToWLTwWv1rmWLSswWMRz3sFon0sDN3/D78fx++/W3IsFh5ba8fC+Iwnd5U+ez3D z15ty3z2emYFxyLjL3DP30nT6tzzZ3y+bdzz95rpT+/5e3UNHovhVToWw/Q7usyxGF6jx+IB8+q8 Rwxd431g5Lil75GFx9basdjGz4pXg83+iWz27jJs9k9r8FgcXCVOPUgWPbIMpx5cIadm/AX21E9X aU/9V+6pv15mT/3XNXosrq7SsWjhsfjVMseiZQ0ei0v8+x7RPhaG7h7U/2W8T5Yci4XHvupYnI2i /5Uei/2rdCx28Fi8uMyx2LHGjsWK7hV4bv5egYVclrRKuewY70lNXiaXHVuD59Aw7+1cjVxmQX2M yloml1n+f/beB6jK+9z3fZc1f5o0hAmZeDgxU2LUq2az67+tSTTqcEmoUS9qGAZUqPgHgpy9QgEB 77pWiCZDCHCEG4YjHOWEcYMjZNamQIDJChso0rMuAynQWcMIFaaUAAeYvYYDtKTtcD8Ll6kYmnxN JPH+qfNUIt/39z7P876/5/t5Fwu4D3sxsED3xYD72g/Nc18M3Kf3xcACzdkB9ywdmmfODtync9Zz gXjd083kXvPwuqfI6991L7xNCzMvXOu6ZsJS05fnxa3P3W+92LBAzy4b3M8nm+Z5dtkgPrt8170Y W6Dn+zH3M/z/nOf5fkx8vo/6Ht4rGbRAvQhw9yJ4nl4E3Ie98FygPeLp3gdPzbNHPMU98n30ImmB enHC3YtT8/TixH3YC+8Fen+Ct/s9CD+e5/0J3uL7E76PXry3QJ7q2gc/nmePeN+ne8Tb/bOIFqIX se5enJmnF7H3YS+WuX9Oy0K8h8f1c6xc7+G5Ynz5PTwX7sNerFmgr7GtcX8dbe08X2NbI36N7fvo xcgC9eKGuxdj8/Tixn3ai98sUC9+5e5Fxzy9+NV92ovKBepFibsXH83Ti5L7tBcfLlAv/pu7F9Z5 evHf7tNeFC9QLy64e3Flnl7crz5yYYF6keHuRcE8vci4D3uxwf3z0hbi9YvfuN//2DvP6xe/uU97 4VigXtjdveiapxf2+7QXv1mgXvzK3YuOeXrxq/u0F40L1Isqdy+a5ulF1X3YC/8Fen+Cv/s9CK/N 8/4Ef/H9Cd/Hc2rIAr1+scv9fSwH5nn9QvnZdtHf8Wu//gv0M8z83T+n7NV5foaZv/gzzL7rXkSZ FubnP0a574lo05d//uOtz91vvUheoPfRJ7vfK/+/z/M++uSveR+99F6Bx26+V+D77FvlAvWtxN23 O5/jksXnuO+6F6kL1ItUd71n5+lF6n3ai/QF+tlv6e6f75Yxz89+Sxd/9tv30YvGBepFlbsXTfP0 ouo+7EXOAvF6jpvJ35+H13NEXv+ue7HM/bsVFuI187fdr5m/P89r5m8LvXjje3h2mV6gZ5dR97PL n+d5dhm9T3tRuUC9KHH34qN5elFyH/Zi6wLNi63umeA3z7zYKs6L77oXaxbo+15Xu7+39R/m+b7X NeL3vZ74Ht7PtBDP9+vdz/D/NM/z/Qbx+f677sXWBfr+gy3u7zHYNs/3H2wVv//g++iFfYF6YXP3 omWeXtjuw174L1Av/ld3va/O0wv/+7QXWxfoaylb3V8v2TbP11K2il9L+U/fcS/2LNDXG/e4Z3Lg PF9v3CN+vfH76EXxAvXigrsXV+bpxYX7sBfBC/Q1tmA3w4XM8zW2YPFrbN9HL36zQL34lbsXHfP0 4lf3aS/sC9QLm7sXLfP0wnYXvdhPxJpu/j6kCT53u9626IsPl/3UOM4xifx/As/7R2d/D8Dd/W/S 9bZY46f/crOa/172yC/O/+LBH7T/L89mehv/6vptgsYL//rILx7e0zv7+Rb/6aTknY8are9NJv30 ocXG5vOTSU+6V0rh39M4Ntm02HDFH0g+7R//q7GDtV3/7frfjn/50R3nrzJ9yMV5w5fePHzz7xd+ /sNfuP6eANZ/+i+m2S78aHfv7PrJgb2GM+3zpLSlk0kzs32ZTPofsT/8hUuz+OZLyMZrnk/te9Kd z5UOV8cXG7s5/jBr3nn8sUTXbWLM+Z23tz5+jCVu/71Ut1+Tf/8Or4mrz66PX/N8ZL+rJles/d96 jW9yTVzZ//nZ6e/8mhxc/dQnC31NnN/hNTG5P+5944e/oCe1//qjHtPelocNn//zb9fEeemhea/J n/j3HT++2X9X+PzmR7/46djKr7km/kG3rknIQ3//mizP6Z1d/7/kfv01GV31VNOT7nxuXZNjOTev yZ3Hf9U1uVcffxXrJgvzevFt83pGZN1o93OFZR7WjRbmtev7pX74HfYifIF6Ee6uN3KeXoTfp71w fa3o5wvQC9e6rt8V8q7p5u+7u/PrU0fuw15cdj+b3eteXHZ/T0CZ+/cc396LW5+733pR5/7+invd C9e6b/J3s/t3Nd/ei1ufu9960UVC+QvQiy739830uX8f9+29uPW5b9OLRx66+V6BRzzmf6+A630C rv+53iuwEH0b52QfLEDfxt1fI/ucvy/f0bdbn7vbvn0Xz0sfLtDzkus1lMPzvL4SLL6+8s/Eo99h L8IX6GtZ4e6vV70xz9eywsWvZX3XvYhdoF7EfsV7ZWLvshe3ng8emOf54LxpYZ+jXR9/1XP0fz// 8LzPB7/h3xfyOdq1vvjM9m9PuvO522e2B/7OM9sDd1yT//EdXpNbz9EHVz9S76rJFXc+R6vX5F4/ R6vX5DXPp17/f+M1ec3zkaD/p14T9knDvbgm9+rjr/p9SJeEr7c/cBfz+px7Xdf7s2bfQz7P70Ny fe6HXzGvDxuu3+n4phHPHeW6q940zLN9ffzmfTo7x105Pej++9a/Pfgd9u4fjZs/++Hb9O72vXdr vz14234794OF2G/f7z6b5fbvcZ/dud9c/b7zHv4/iP9o3Py62Qb3db79Hv6P7n97+CvuYdfPyAme vVZvcpVc9/Gp2fv31j3xozvu40fvuP9W3Zafh+seMt38naaeX5zJ9fRdaNwkZddvNfzUuPnVp38n V9rlSfgQa4kdxF7iMBFLpBD/mSgkrMS/EZ8SvTOuF7JdRXD4v3N4L4d/yuH/xuFWDi/k8P/M4Skc Hsvhhzl8L4fv4PC1HO7D4Z6uww0ONzjc4HCDww0ONzjc4HCDww0ON1wb3QVnHG5wuMHhBofPvk/9 393VfOquzuqu1lV1irsfrvvgB9xMrn49Ovva7mxnF102rV3k+twibinXs+gfke2+ix3y/+3O3jJ7 Vwc3LrpJ6/80+1+u90+53g/21OK1J9eeNIxbn3Ptg88XPXDr/p/V+tJ7l/Yvi27OOsP9r4tm//Xp xcWm9bNX7Oaxf+v9TdUPbj/2i39dbCzl7z8vmpq5Yy4u++nx/5R4PCH66N+bhjePf8B9vOfs6re+ HmBatGjZ66fiE47/89+/I24e/+DNryp8sc7//MZ5LMx6D3xl5x+anSfLf7DY2GkkmnYyQf9sckWi 6VbPb3X59mvyt+v58BdX0DWPVvzggS+t4/rvO9e6WeHD7r8fuiND05w6Hprdv08ucq00f06L7qj8 bzm5jvRa5ENOrpV/PLv9Xf81Xz7f/DjTlzJNkM+QMOcM9/q4L2dWKJ+hcM4Z7vVxX86sRT5Dy5wz 3OvjvpzZlHyGqTlnuNfHfTmzZSb1DMtMt5/hXh/35cz2yGfYM+cM9/q4L2eWLJ8hec4Z7vVxX87s snyGy3POcK+P+3Jmn8pn+HTOGe71cV/O7M/yGf485wz3+jiT4Fbzn8FVw9/OoFekHndnZi/Lmb08 5wwvy5mpx92Z2WI5s8VzzrBYzkw97s7MWmRmaJlzv7TI95l63J2Z5ciZ5cw5Q46cmXrcnZmFypmF zjlDqJyZetydmd0NAS7k1HB+cZyS4fzPRPfvvPrbGe6XSfvNeX7herNQNTzwrc/0c9PtZ/q5Se2W vruWz6lluXw93jH5fUnp+swPXbtj9hUH0yLXM67rc/84+zU1w1j/yCmOcgl81y02HvrWT24/NyXM 6U6C3J0EuTsJc7qTIHcn6q6685i7O+6j1v7jCzf789C3fob8ualwTo8K5R4Vyj0qnNOjQrlHOd+o Rzm3erTp63rUIveoZU6PWuQetcg9apnToxa5R3XfqEd1t3q08et6NCX3aGpOj6bkHk3JPZqa06Mp uUfD36hHw7d6tP7reqQ+qf/cdPuT+s9Ny2QWUs+wfM5rAcvl1wLeMf3YdDc9+qG7Rz92fenHWBv+ 0Ld+jeHnpj1zerNH7s0euTd75vRmj9wb/7vqzY/cvfE3uV5j5/b5mrsnWe5Q8pwOJcsdSpY7lDyn Q8lyh978Rh16092hjV/Toctyhy7P6dBluUOX5Q5dntOhy3KHCr5RhwrcHdr0NR36VO7Qp3M69Knc oU/lDn06p0Ofyh1q/kYdanZ36IWv6dCf5Q79eU6H9OfVP8sd+vOcDv1Z7tCfvtGM/tPsjF73xYx+ 4B48Qe0xbn+C2mPc61cC4Y45z2hTYo8sppe+wVOHmaNmnzo2/P076G5ekbu9Oy/L3bmb1+6m5rx2 p3bH7xt1559udWf93+/O3bwqeHt3FsvduZvXD6fmvH6odueH36g7fzFe+ton1rt5ZfL27rj+S33t U38Nc2rOa5hqd35jbDPufvr8iqO+mhDv5rXR23uTI/fmbl5FnZrzKqram7xv1Jt0d2/W/d3e3M2r s7f3JlTuzd28jjs153VctTfh36g3+9y9Wf93e3M3rw/vmfNq3B7jXju6z5ze+Mi9WfGNevMf3L3Z cE968/+/Ev3tXsX92zurkkx/Md1Sub5qcPv7XuZ7L85i1/t3f5Difi/Om8QbRARxgHid2E28Smwn 1hFriCvoC4j3iUziHeIMkUzEETHEMSKMCCYCCdcfP2ILsZHwJVYa+zjvPs67j/Pu47z7OO8+zruP 8+7jvPs47z7Ou4/z7jMeJ97nmEziHeIMkUzEETHEMSKMCCYCiZ2EH7GF2Ej4EiuNA5zzAOc8wDkP cM4DnPMA5zzAOQ+w7gHWPcC6B1j3AOseYN0DrHuAdQ+w7gHWPcC6B1j3AOseYN0DrHuUdY+y7lHW Pcq6R1n3KDUcpYajrHeU9Y6y3lHWO8p6R1nvKOsdZb2jrHeU9Y6y3lHWO8p6MawXw3oxrBfDejHk F2O4/hwjwohgInC22zEcH8PxMRwfw/ExHJ9guF6lfpN4g4ggDhC7iVeJOD4fQ4QRwV/cB66PE1gz gTUTWDOBNRNYM4E1E1jzNGueZr3TrHeatU6jP43+NPrT6E+jP43+NPrT6N9G/zaff5vPvc3n3uZz b/O5t/lcFp/LYp0s1shCk4UmC00Wmiw0WWjy0ORxvjx0eeSfRy/zOFce2jy0eWjz0OahLTRcrzi/ Tmzh442EL7GSuzXRdIU1rlD3Fc51hXNd4fgrfP4Kny/nc+WsXc6/lfNv5fxbLX2r5d9rOW8t69Xy 77X8eyNrNfK5RtZoRN/I5xo5VyOfb+TzLYbrFd03CT8+3kJsJFYanfx7J//eyZqdrNlJDZ1oOvl8 J8d2ounh8z3k0cPaPXyuh+N7+HwPn+/h8wOsMYBmgDUGWGMA7QD1DnAdBriOA9Q2wLoDHDvAcQMc N8BxYxw3xnFjHDfGcWMcN8ZxYxw3Rk/GuBfGOH7MNTfc94Hr4zHWGSOHMdYaY60x1poyXK/Gvkm8 QUQQrxO7iVeJ7UQYmmAikNhJ+BFbiI2EL7HSmGGNGdaYYY0Z1phhjRnWmOH4Gfb6DPnMcI/PsM4M 68ywzgzrzLDODOvMsM4M6zxkSjQ9ZHqTeIOIIA4QrxO7iVeJ7cQLxDpiDfEUkcxxx4gwIpgIJJix Jj9iC7GR8CVWGp6s78n6nqzvyfqerO/J+p6s78n6nqzvyfqepmeIp4j3OSaZOEaEEcFEILGT8CO2 EBsJX2Kl4c363qzvzfrerO/N+t6s7c3a3qztzdre5O5N7t6m54iniDiOiyHCiGAikNhJ+BFbiI2E L7HSWGZyvSr8JvEGEUEcIF4ndhOvEtuJF4h1xBriOcJEFHxxH7g+XkZdy0xniGQijoghjhFhRDAR SOwk/IgtxEbCl1jJ1E5k5TeJN4gI4gDxOrGbeJXYTrxArCNcf54jniEeJz7n+C4ikzhDJBNxRAxx jAgjgonAWVdaw/nXcP41nH8N51/D+Tdw/g2cfwPn38D5N3D+DZx/A+ffwPk3cP4NnH8D59/AGTdw /g2cfwO93kAOG0wPEybic9YaI3qJj4gPiPeJTOIMkUzEETHEMSKMCCYCiZ2EH7GF2Ej4EiuNreS2 ldy2kttWcttKblvJbSu5bSW3reS2ldy2kttWcttKblvJbSu5bSW3reS11TTBOp8RXUQ78QnxEWEl PiAyiXeIM0QyEUMcI8KIYCKQ2En4EVuIjYQvsdLwd723130fuD72J1d/cvUnV39y9SdXf3L1J1d/ cvUnV3/y9CdPf/L0J09/8vSnh/7k6k8P/cnXnz76k7M/vfQnb3+mpT95+5OzPxPan7z96a0/ufuT uz+5+5O7P/31J39/8vcnf3/y9yd/f/L3J39/8vcnf3/y9yf/PSbXK/9vEm8QEcQB4nViN/EqsZ14 gVhHrCGeI54hniIeJx4mTMTnrDVGfEZ0Ee0Ez8WmJuITwkoUEO8TmcQ7xBkimYgjYohjRBgRTAQS Owk/YguxkfAlVhrB5BxMzsHkHEzOweQcTM7B5BxMzsHkHEzOweQcTM7B5BxMzsHkHEzOweQcTM7B 5BxMr4PJO5i8g+l1MHkHk28w90cwOQfT52DyDibvYPIOJu9g8g4m72DyDibvYPIOJu9b94Hr42Bq CKaGYGoIpoZgagimhmBqCKaGcGoIp4ZwaginhnBqCKeGcGoIp4ZwaginhnBqCKeGcGoIp4Zwagin hnBqCKeGcGoIp4ZwaginhnBqCKf/4dQRTv/D6X849YRTTzj3TTj1hFNPOPWEU0849YRTTzj1hFNL OLWEk384+YeTfzj5h5N/OPmHk384+YeTfxT5R5F/FPlHkX8U+UeRfxT5R5F/FPlHkX8U+UeRfxS5 R5F7FLlHkXsUuUeRexS5R5F7FLlHkXsUuUeRexS5R5F7FLlHkXsU1yKK/KPIP4r8o8g/ivyjyD+K /KPIP4rrEUUNUdQQRQ1R1BBFDVHUEEUNUdQQRQ1R1BBFDbHUEEsNsdQQSw2x1BBLDbHUEEsNsdQQ Sw2x1BBLDbFcg1jqiKWOWOqIpY5Y6oiljljqiKWOWOq4dR+4Po6lplhqiqWmWGqKpaZYaoqlpljq iaWeWOqJpZ5Y6omlnljqiaWeWOqJpZ5Y6omlnljqiaWeWOqJpZ5Y6omlnljqiaWeZJPra1RvEm8Q EcQB4nViN/EqsZ14gVhH4BTUk0w9ydSTTD3J1JNMPcnUkkwtyeSfTP7J5J9M/snkn0z+yeSfTP7J 5J9M/snkn0z+yeSfTP7J5J9M/snkn0z+yeSfTP7J5J9M/snkn0z+yeSfTP7J5J9M/qnkn0r+qeSf Sv6p5J9K/qnkn0r+qeSfSv6p5J9K/qnkn0r+qeSfSv6p5J9K/qlcj1RqSKWGVGpIpYZUakilhlRq SKWGVGpIpYZUakilhlRqSKWGVGpIpYZUakilhlRqSCX/VPJPJf9U8k8l/1TyTyX/VPJPJf9U8k8n /3TyTyf/dPK/dR+4Pk6nlnRqSaeWdGpJp5Z0akmnlnRqSaeWdGpJp5Z0akmnlnRqSaeWdGpJp450 6kinjnTqSKeOdOpIp4Z09kY6daRTRzp1pFNHOnWkU0c6daRTRzp1pHMt0qklnVrSqSWdWtKpJZ1a 0qklnVrSqSWdWnKoJYdacqglh/xzyD+H/HPIP4f8c8g/h/xzyD+H/HPIP4f8c8g/h/xzyD+H/HPI P4f8c7gWOdSQQw051JBDDTnkn8N1yKGGHGrIoYYcasihhhxqyKGGHGrIoYYcasihhhxqyKGGHGrI oYYcasihhhxqyKGGHGrIoYYcaiighgJqKKCGAmoooIYCaiighgJqKKCGAmoooIYCaiighgJqKKCG AmoooIYCaiighgJqKKCGAmoooIYCaiighgKuQwF1FFBHAXUUUEcBdRRQRwF1FFDHrfvA9XEBNRVQ UwE1FVBTATUVUFMBNRVQUwE1FVBTATUVUFMBNRVQUwE1FVDTZZPrq6xvEm8QEcQB4nViN/EqsZ14 gVhHrCGeI54hniIeJx4mTMTnrDVBjBGfEb1EF9FOtBBNxCfER4SVuEJ8QBQQ7xOZxBkimYgjYohj RBgRTAQSOwk/YguxkfAlVhpl1FBGDWXUUEYNZdRQRg1l1FBGDWXUUEYNZdRQRg1l1FBGDWXUUEYN ZdRQRg1l1FBGDWXUUEYNZdRQRg1l1FBGDWXUUEYNZdRQRg1l1FBGDWXUUEYNZdRQxrUoo44y6iij jjLqKKOOMuooo44y6iijjjLqKKOOMuooo44y6qiijirqqKKOKuqooo4q6qiijirqqKKOKuqooo4q 6rh1H7g+rqKmKmqqoqYqaqqipipqqqKmKmqqoqYqaqqipipqqqKmKmqqoqYqaqqipipqqqKmKmqq oqYqaqqipipqqqKmKmqqoqYqaqqipipqqqKmKmqqoqYqaqqipipqqqKmKmqqo6Y6aqqjpjpqqqOm Omqqo6Y6aqqjpjpqqqOmOuqoo4466qijjjrqqKOOOuqoo4466qijjjrqqKOOOuqoo4466qijjjrq qKOOOuqoo4466qijjjrqqKOOOuqoo446aqijhjpqqKOGOmqoo4Y6aqijhjpqqKOGOmqoo4Zmamim hmZqaKaGZmpopoZmamimhmZqaKaGZmpopoZmamimhmZqaKaGZmpopoZmamimhmZqaKaGZmpopoZm amimhmZqaKaGZmpopoZmamimhmZqaKaGZmpopoZmamjmWjRTRzN13LoPXB83U1MzNTVTUzM1NVNT MzU1U1MzNTVTUzM1fWpyvXfgTeINIoI4QLxO7CZeJbYTLxDriDXEc8QzxFPE48TDhIn4nLUmiDHi M6KX6CLaiRaiifiE+IiwEleID4gC4n0ik3iHOEMkE3FEDHGMCCOCiUBiJ+FHbCE2Er7ESqOLOrqo o4s6uqijizq6qKOLOrqoo4s6uqijizq6qKOLOrqoo4s6uqijizq6qKOLOrqoo4s6uqijizq6qKOL Orqoo4s6uqijizq6qKOLOrqoo4s6uqihixq6qKGLGrqooYsauqihixq6qKGLGrqooYsauqihixq6 qKGLGvqooY8a+qihjxr6qKGPGvqooY8a+qihjxr6qKGPGvqooY8a+qihjxr6qKGPGm7dB66P+6in j3r6qKePevqop496+qinj3r6qKePevqop496+qinj3r6qKeP69JHTX3U1EdNfdTUR0191NRHTX3U 1EdNfdTUR0191NRHTX3U1EdNfdQ0TE3D1DRMTcPUNExNw9Q0TE3D1DRMTcPUNExNw9Q0TE3D1DRM TcPUNExNw9QxTB3D1DFMHcPUMUwdw9QxTB3D1DFMHcPUMUwdw9QxTB3D1DFMHcPUMUwdw9QxTB3D 1DFMHcPUMUwdw9QxTB3D1DFMHcPUMUwdw9QxTB3D1DFOHePUMU4d49QxTh3j1DFOHePUMU4d49Qx Th3j1DFOHePUMU4d49QxTh3j1DFOHePUMU4d49QxTh3j1DFOHePUMU4d49QxTh3j1DFOHePUMU4N 49QwTg3j1DBODePUME4N49QwTg3j1DBODePUME4Nt77X1/XxOPWMU8849YxTz1d/JePW94jf+f0P 831v8sOz//ofFp8y7TYSTAFf87W1W+o41P8sq2NQR8rqY6h/JqvDUIfI6mDU+2R1IOrXZPVO1P6y 2g/1y7J6I+qfyGpf1Ktkdamx2yg2VPUl1BdkdR7qHFl9HvV7svpd1Odk9VuoT8tqC+pEWX0Sdays NqOOltVRqI/I6sOoD8rqUNRBsno/6j2yejfqAFn9Cuodsnob6hdl9WbU62X1WtTPy+rVqJfL6lOm IGZsoDxjg5ixgfKMDWLGBsozNogZGyjP2CBmbKA8Y4OYsYHyjA1ixgbKMzZo9nVrdcYGMWMD5Rkb xIwNlGdsEDNWVT+B+jFZnceuzDEC5akZxNQMlKdmEFMzUJ6aQbNfo1anZhBTM1CemkFMzUB5agYx NQPlqRnE1AyUp2YQUzNQnppB/AmUp2YQUzNQnpqueRwoT80gpmagPDWDmJqB8tQMYmoGylMziKkZ KE/NIKZmoDw1w5iaIfLUDGNqhshTM4ypGSJPzTCmZog8NcOYmiHy1AxjaobIUzOMqRki7/kw9nyI vOfD2PMh8p4PY8+HyHs+jD0fIu/5MPZ8iLznw9jzIfKeD2PPh8h7Pox7NkTe82Hs+RB5z4ex50Pk PR/Gng+R93wYez5E3vNh7PkQec+HsedD5D0fxp4Pkfd8JHs+Qt7zkez5CHnPR7LnI+Q9H8mej5D3 fCR7PkKmmUhoJkKmmUhoJkLel5Hsywh5X0ayLyPkfRnJvoyQ92Uk+zJC3peR7MsIeV9Gsi8j5H0Z yb6MkPdlJPsyQt6XkezLCHlfRrIvI+R9Gcm+jJD3ZST7MkLel5Hsywh5X5rZl9HyvjSzL6Plfcl9 JauPof6ZrA5E/ZqsNhtm7u9o+f42c39Hy/e3mfs7Wr6/b3Kven+bub+j5fvbzP0dLd/fZu7vaPn+ NnN/R8v3t5n7O1q+v83c39Hy/W3m/o6W7+8k7u84+f5O4v6Ok+/vJO7vOPn+TuL+jpN9JwnfiZN3 QxK7IU52qaTZ94+pTpKEk8TJOy2JaxMn750k9k6cvHeS2Dtx8t5JYu/EyXsnib0TJ++dJPZOnLx3 ktg7cfLeSWLvxMl7J4m9EyfvnST2Tpy8d1LYOxZ5N6SwGyzybkhhN1jk+zuF+9si3ycp3CcW+T5J 4T6xyPdJCveJRb5PUrhPLPJ9ksJ9YpHvkxTuE4t8n6Rwn1jk+ySF+8Qi3ydp3Cdn5X6n0e+zcgfT 6OBZuYNpdPCs3ME0OnhW7mAaHTwrdzCNDp6VO5hNBzPkvZPN3smQd0M2uyFDvjrZXJ0M+epkc3Uy 5KuTzdXJkK9ONlcnQ7462VydDPnqZHN1MuSrk8/VyZXnYD5zMFe+lvlcy1yZCvKhglz5+TKf58tc ebLlM9ly5Sufz5XPla98Plc+V77y+Vz5XPnK53Plc+Urn8+Vz5WvfBFX/qL8il+RsU9WbzOK6MlF uSdF9OSi3JMienJR7kkRPbko96SUnhTLu6GU3VAsk2kpZFosT7ZSJluxPNlKmWzF8v1dyv1dLPe7 lH4Xy/0upd96ByvpoFWeEJVMCKtcZSVVWuUqK6nSKldZSZVW+cnLxpNXtdwTGz2plmesjRlbLe9L G/uyWu6JjZ5Uyz2x0ZNqeaddY6fVyx28Rgfr5d1wjd1QL98n17hP6uUOXqOD9fJku8Zkq5f7fY1+ 18v9vka/6+V+t9Fvu9zvNvptlzvYRgftcgfb6KBd7mAbHbTLPWmjJ3a5Jw560i73xEFP2uVd7GAX t8u72MEubpdpxgHNtMtXx8HVaZf77aDf7fId6+CObZevjoOro/e7l35fl72hF2+4Lk+IXibEdbmD vXTwunx/93J/X5f73Uu/r8v97qXf1+V+99Lv6/JuGGI39MtXZ4ir0y/vhiF2Q7+8G4bYDf3ylR/i yvfLHDsEx/bLr80MGa/J6sPGkHHQ6JcJbwjC65f3/BB7vl++Y4e4Y/vle3CIe7BfvgeHuAf75Xtw iHuwX74HndyDI/I96OQeHJHvQSf34Ih8Dzq5B0fke9DJPTgi34NO7sER+R50cg+OyE8ZTp4yRuTX v51GrDEi399O7u8R+Y51cseOyHeskzt2RJ6xTmbsiHx/O7m/R+T728n9PSLf307u7xH5/p7m/p6Q 7+9p7u8J+f6e5v6ekO/vae7vCfmOneaOnZDv2Gnu2An5jp2e/WkB6us+08bLsvqwMc0dOyFP5Gkm 8oTMENMwxIS8G6bZDRPybphmN0zIu2Ga3TAh74ZpdsOEvBum2Q0T8m6YZjdMyLthkSnB9Fd5Nywy /bOsjkEdKauPof6ZrA5GvU9WB6J+TVb7oX5ZVm9E/RNZfdJYZIo1/iq/U2CR6YisDkUdJKv3o94j q3ejDpDVr6DeIau3oX5RVm9GvV5Wr0X9vKxejXq5rD5leoS984BJ3TuPsHdUdQzqSFl9DPXPZHUY 6hBZHYx6n6wORP2arN6J2l9W+6F+WVZvQb1JVm9E/RNZ7Yt6laz2Rv2krLYYj5gSDVUdhfqIrD6M +qCsDkUdJKv3o94jq3ejDpDVr6DeIau3oX5RVm9GvV5Wr0X9vKxejXq5rD5l8mKeeMjzxIt54iHP Ey/miYc8T7yYJx7yPPFinnjI88SLeeIhzxMv5omHPE+8mCce8jzxYp54yPPEi3miqn1QPy2rvVE/ KavzDC9TjuEhzxMv5omHPE+8mCce8jzxYp54yPPEi3niIc8TL+aJhzxPvJgnHvI88WKeeMjzxIt5 4iHPEy/miYc8T7yYJx7yPPFinnjI82Qp82SJPE+WMk+WyPNkKfNkiTxPljJPlsjzZCnzZIk8IZYy IZbIE2IpE2KJPCGWMiGWyBNiKRNiiUwcSyGOJTJxLIU4VPVK1M/Kam/UT8rqk8ZSnmBUtRl1tKw+ jPqgrA5FHSSr96PeI6t3ow6Q1a+g3iGrt6F+UVZvRr1eVq9F/bysXo16uaw+ZVrBPPGR58kK5omP PE9WME985HmygnniI8+TFcwTH5lPVsAnPvL0WcH08ZGnzwqmj488fVYwfXzk6bOC6eMjT58VTB8f efqsYPr4yNNnBdNHVS9GPWOo6kvGCtMFWZ2HOkdWv4X6tKy2oE6U1SdRx8pqM+poWR2F+oisPoz6 oKwORR0kq/ej3iOrd6MOkNWvoN4hq7ehflFWb0a9XlavRf28rF6NermsPmXyZcaukmesLzN2lTxj fZmxq+QZ68uMXSXPWF9m7Cp5xvoyY1fJM9aXGbtKnrG+zNhV8oz1ZcaukmesLzN2lTxjfZmxq+QZ 64t2lTxjfZmxq+SnUV+eRlX1E6gfk9V/5eM/Gaq6m48dsvo8H78nq9/i49Oy2sLHibL6JB/Hymoz H0fL6ig+PiKrD/PxQVkdatz8uSPq/PZlfq+S57cv83uVPL99md+r5Pnty/xeJc9vX+b3Knl++zK/ V8nz25f5vUqe35uY3+vk+b2J+b1Ont+bmN/r5Pm9ifm9Tp7fm5jf6+T5vYn5vU6e35uY3+vk+b2J +b1Ont+bmN/r5Pm9iT/r5Pm9ifm9Tp7fLmdYJ8/vTczvdfL83sT8Xic/z2/ieX6dPO03Me1V9aOo H5TVi1G7fsuI6iSbcBJV7UQ9Iqt/j/p3sroWdaWsvoy6UFbnoc6R1edRvyer30J9WlZbUCfK6pOo Y2W1GXW0rI5CfURWH0Z9UFaHog6S1ftR75HVu1EHyOpXUO+Q1dtQvyirN6NeL6vXon5eVq9GvVxW nzJtxwFfkh1wOw74kuyA23HAl2QH3I4DviQ74HYc8CXZAbfjgC/JDrgdB3xJdkDX73d5SXbA7Tjg S7IDbmcmvyQ74HYc8CXZAbfjgC/JDrgdB3xJdsDtOOBLsgNuxwFfkh1wOw74kuxprt+cpar/aGw3 jcvqYdR/kNXdqB2y+reoP5XV9ag/ltW1qCtl9S9RfyirL6MulNXnUb8nq99FfU5Wv4X6tKy2oE6U 1WbU0bI6CvURWX0Y9UFZHYo6SFbvR71HVu9GHSCrX0G9Q1ZvQ/2irN6Mer2sXov6eVm9GvVyWX3K FIBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf+sl+GYBf +skOGIAD+skOGIAD+skOGIAD+skOGIAD+skOGIAD+snPgAE8A/rJfhmAX/rJz4ABPAP6ye4agLv6 yU+MATwx+sleHIAX+8nPlwE8X/rJzh2Ac6vqNtR2WV2P+mNZ/UvUH8rqUtTFsvoy6kJZnYc6R1af R/2erH4X9TlZ/Rbq07LagjpRVp9EHSurzaijZXUU6iOy+jDqg7I6FHWQrN6Peo+s3o06QFa/gnqH rN6G+kVZvRn1elm9FvXzsno16uWy+pRpL1SwS6aCvVDBLpkK9kIFu2Qq2AsV7JKpYC9UsEumgr1Q wS6ZCvZCBbtkKtgLFeySqWAvVLBLpoK9UMEu+Sl6L0/Ru2SG2AtD7JIZYi8MsUtmiL0wxC6ZIfbC ELtkhtgLQ+ySGWIvDLFLZoi9MMQumSH2whC7ZCrYCxXskqlgL1SwS/b5vfj8LvkJfS9P6LtkKtgL FajqX6NulNX1qD+W1b9E/aGsvoT6gqzOQ50jq8+jfk9Wv4v6nKx+C/VpWW1BnSirT6KOldVm1NGy Ogr1EVl9GPVBWR2KOkhW70e9R1bvRh0gq19BvUNWb0P9oqzejHq9rF6L+nlZvRr1cll9yhQKFQTJ VBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAK FQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJVBAKFQTJryyEmsZl tRP1iKweRv0HWf171L+T1b9F/amsrkf9sayuRV0pq3+J+kNZfRl1oay+hPqCrM5DnSOrz6N+T1a/ i/qcrH4L9WlZbUGdKKtPoo6V1WbU0bI6CvURWX0Y9UFZHYo6SFbvR71HVu9GHSCrX0G9Q1ZvQ/2i rN6Mer2sXov6eVm9GvVyWX3KFAFDHJIZIgKGOCQzRAQMcUhmiAgY4pDMEBEwxCGZISJgiEMyQ0TA EIdkhoiAIQ7JDBEBQxySGSIChjgkM0QEDHFIZogIGOKQzBARMMQhmSEiYIhDMkNEwBCHZIaIgCEO yQwRAUMckhkiAoY4JDNEBAxxSGaICBjikMwQETDEIZkhImCIQzJDRMAQh+RXLSJMDln9W9Sfyuo2 1HZZ/WvUjbK6HvXHsroWdaWsLkVdLKsvoy6U1ZdQX5DVeahzZPV51O/J6ndRn5PVb6E+LastqBNl tRl1tKyOQn1EVh9GfVBWh6IOktX7Ue+R1btRB8jqV1DvkNXbUL8oqzejXi+r16J+XlavRr1cVp8y nYBPjst8cgI+OS7zyQn45LjMJyfgk+Myn5yAT47LfHICPjku88kJ+OS4zCcn4JPjMp+cgE+Oy3xy Aj45LvPJCfjkuMwnJ+CT4zJxnIA4jsvEcQLiOC4TxwmI47hMHCcgjuMycZyAOI7LxHEC4jguE8cJ iOO4TBwnII7jMnGcgDiOy8RxAuI4LhPHCYjjuEwcJyCO4zJxnIA4jsvEcQLiOC4TxwmI47hMHCcg juPy6ycnTB/K6lLUxbL6MupCWX0J9QVZnYc6R1afR/2erH4X9TlZ/Rbq07LagjpRVp9EHSurzaij ZXUU6iOy+jDqg7I6FHWQrN6Peo+s3o06QFa/gnqHrN6G+kVZvRn1elm9FvXzsno16uWy+pQpHpox yzQTD82YZZqJh2bMMs3EQzNmmWbioRmzTDPx0IxZppl4aMYs00w8NGOWaSYemjHLNBMPzZhlmomH ZswyzcRDM2b51ZZ407Oy2gf107LaG/WTsvoJ1I/J6kdRPyirF6OeMcwy+8TDPmaZfeJhH7PMPvGw j1lmn3jYxyyzTzzsY5bZJx72McvsEw/7mGX2iYd9zDL7xMM+Zpl94mEfs8w+8bCPWaaZeGjGLNNM PDRjlmkmHpoxyzQTD82YZZqJh2bMMs3EQzNmmWbioRmzTDPx0IxZppl4aMYs00w8NGOWaSYemjHL NBMPzZhlmomHZswyzcRDM2aZZuKhGbNMM/HQjFmmmXhoxizTTDw0Y5ZpJh6aMcs0Ew/NmGWasUAz iTLNWKCZRJlmLNBMokwzFmgmUaYZCzSTKNOMBZpJlGnGAs0kyjRjgWYSZZqxQDOJMs1YoJlEmWYs 0EyiTDMWaCZRphkLNJMo04wFmkmUacYCzSTKNGOBZhJlmrFAM4kyzVigmUSZTyzwSaLMJxb4JFHm Ewt8kijziQU+SZT5xAKfJMp8YoFPEmU+scAniTKfWOCTRJlPLPBJoswnFvgkUeYTC3ySKPOJBT5J lPnEAp8kynxigU8SZT6xwCeJMp9Y4JNEmU8s8EmizCcW+CRR5hMLfJIo84kFPkmU+cQCnyTKfGKB TxJlPrHAJ4kyn1jgk0SZTyzwSaLMJxb4JFHmEwt8kijziQU+SZT5xAKfJMp8YoFPEmU+OQefnJH5 5Bx8ckbmk3PwyRmZT87BJ2dkPjkHn5yR+eQcfHJG5pNz8MkZmU/OwSdnZD45B5+ckfnkHHxyRuaT c/DJGZlPzsEnZ2Q+OQefnJH55Bx8ckbmk3PwyRmZT87BJ2dkPjkHn5yR+eQcfHJGfrXlnOlPsvqP qMdltRP1iKweRv0HWf171L+T1d2oHbL6t6g/ldVtqO2y+teoG2V1PeqPZXUt6kpZXYq6WFZfRl0o qy+hviCr81DnyOrzqN+T1e+iPier30J9WlZbUCfK6pOoY2V1FOojsvow6oOyOhR1kKzej3qPrN6N OkBWv4J6h6zehvpFWb0Z9XpZvRb187J6NerlsvqUKRM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+ SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5 JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZP5JBM+SZOJIxPiSJOJIxPiSJOJIxPiSJOJIxPi SJOJIxPiSJOJIxPiSJMZIhOGSJPfrZJp+lBWl6IultWXURfK6kuoL8jqPNQ5svo86vdk9buoz8nq t1CfltUW1Imy+iTqWFltRh0tq6NQH5HVh1EflNWhqINk9X7Ue2T1btQBsvoV1Dtk9TbUL8rqzajX y+q1qJ+X1atRL5fVp0y58Ml5mU9y4ZPzMp/kwifnZT7JhU/Oy3ySC5+cl/kkFz45L/NJLnxyXuaT XPjkvMwnufDJeZlPcuGT8zKf5MIn52U+yYVPzst8kgufnJf5JBc+OS/zSS58cl7mk1z45LzMJ7nw yXmZT3Lhk/Myn+TCJ+dlPsmFT87Lr5/kmkZk9TDqP8jq36P+nazuRu2Q1b9F/ams/jXqRlldj/pj WV2LulJW/xL1h7K6FHWxrL6MulBWX0J9QVbnoc6R1edRvyer30V9Tla/hfq0rLagTpTVJ1HHymoz 6mhZHYX6iKw+jPqgrA5FHSSr96PeI6t3ow6Q1a+g3iGrt6F+UVZvRr1eVq9F/bysXo16uaw+ZboE zVyQaeYSNHNBpplL0MwFmWYuQTMXZJq5BM1ckGnmEjRzQaaZS9DMBZlmLkEzF2SauQTNXJBp5hI0 c0GmmUvQzAWZZi5BMxdkmrkEzVyQaeYSNHNBpplL0MwFmWYuQTMXZJq5BM1ckGnmEjRzQaaZS9DM BZlmLkEzF2SauQTNXJBp5hI0c0GmmUvQzAWZZi5BMxdkmrkEzVyQX5u5ZLLL6l+jbpTV9ag/ltW1 qCtl9S9RfyirS1EXy+rLqAtl9SXUF2R1HuocWX0e9Xuy+l3U52T1W6hPy2oL6kRZfRJ1rKw2o46W 1VGoj8jqw6gPyupQ1EGyej/qPbJ6N+oAWf0K6h2yehvqF2X1ZtTrZfVa1M/L6tWol8vqU6YS2KdI Zp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S 2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdI Zp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S 2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIZp8S2KdIppkSaKZIppkSaKZIppkSaKZIppkSaKZI ppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkSaKZIppkS aKZIphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNV phkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkr NHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNV phkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVphkrNHNVfiXHajonq99CfVpWW1An yuqTqGNltRl1tKyOQn1EVh9GfVBWh6IOktX7Ue+R1btRB8jqV1DvkNXbUL8oqzejXi+r16J+Xlav Rr1cVp8y1cA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+ FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL7 1MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+ FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL7 1MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+FTL71MA+ FTL71MA+FTL71MA+FTL71MA+FTL7NMA+Npl9GmAfm8w+DbCPTWafBtjHJrNPA+xjk9mnAfaxyezT APvYZPZpgH1sMvs0wD42mX0aYB+bzD4NsI9NZp8G2Mcms08D7GOT2acB9rHJ7NMA+9hk9mmAfWwy +zTAPjaZfRpgH5vMPg2wj01mnwbYxyazTwPsY5PZpwH2scns0wD72GT2aYB9bDL7NMA+Npl9GmAf m8w+DbCPTWafBtjHJrNPA+xjk9mnAfaxyezTAPvYZPZpgH1sMvs0wD42mX0aYB+bzD4NsI9NZp8G 2Mcms08D7GOTaaYBmrHJNNMAzdhkmmmAZmwyzTRAMzaZZhqgGZtMMw3QjE2mmQZoxibTTAM0Y5Np pgGasck00wDN2GSaaYBmbDLNNEAzNplm7NBMk0wzdmimSaYZOzTTJNOMHZppkmnGDs00yTRjh2aa ZJqxQzNNMs3YoZkmmWbs0EyTTDN2aKZJphk7NNMk04wdmmmSacYOzTTJNGOHZppkmrFDM00yzdih mSaZZuzQTJNMM3ZopkmmGTs00yTTjB2aaZJpxg7NNMk0Y4dmmmSasUMzTTLN2KGZJplm7NBMk0wz dmimSaYZOzTTJNOMHZppkmnGDs00yTRjh2aaZJqxQzNNMs3YoZkmmWbs0EyTTDN2aKZJphk7NNMk 04wdmmmSacYOzTTJr+TYTYmy+iTqWFltRh0tq6NQH5HVh1EflNWhqINk9X7Ue2T1btQBsvoV1Dtk 9TbUL8rqzajXy+q1qJ+X1atRL5fVp0wdsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+r zD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4d sE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+r zD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4d sE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+r zD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD4dsE+rzD7dsI9DZp9u2Mchs0837OOQ2acb 9nHI7NMN+zhk9umGfRwy+3TDPg6ZfbphH4fMPt2wj0Nmn27YxyGzTzfs45DZpxv2ccjs0w37OGT2 6YZ9HDL7dMM+Dpl9umEfh8w+3bCPQ2afbtjHIbNPN+zjkNmnG/ZxyOzTDfs4ZPbphn0cMvt0wz4O mX26YR+HzD7dsI9DZp9u2Mchs0837OOQ2acb9nHI7NMN+zhk9umGfRwy+3TDPg6ZfbphH4fMPt2w j0OmmW5oxiHTTDc045Bpphuaccg00w3NOGSa6YZmHDLNdEMzDplmuqEZh0wz3dCMQ6aZbmjGIdNM NzTjkGmmG5pxyDTTDc04ZJrphmYcMs10QzMOmWa6oRmHTDPd0IxDppl+aOaGTDP90MwNmWb6oZkb Ms30QzM3ZJrph2ZuyDTTD83ckGmmH5q5IdNMPzRzQ6aZfmjmhkwz/dDMDZlm+qGZGzLN9EMzN2Sa 6Ydmbsg00w/N3JBpph+auSHTTD80c0OmmX5o5oZMM/3QzA2ZZvqhmRsyzfRDMzdkmumHZm7INNMP zdyQaaYfmrkh00w/NHNDppl+aOaGTDP90MwNmWb6oZkbMs30QzM3ZJrph2ZuyDTTD83ckGmmH5q5 IdNMPzRzQ6aZfmjmhvxKTr8pR1afR/2erH4X9TlZ/Rbq07LagjpRVp9EHSurzaijZXUU6iOy+jDq g7I6FHWQrN6Peo+s3o06QFa/gnqHrN6G+kVZvRn1elm9FvXzsno16uWy+pRpFPYZlNlnFPYZlNln FPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZ lNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNln FPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZ lNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNln FPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZlNlnFPYZ lNlnEvZxyuwzCfs4ZfaZhH2cMvtMwj5OmX0mYR+nzD6TsI9TZp9J2Mcps88k7OOU2WcS9nHK7DMJ +zhl9pmEfZwy+0zCPk6ZfSZhH6fMPpOwj1Nmn0nYxymzzyTs45TZZxL2ccrsMwn7OGX2mYR9nDL7 TMI+Tpl9JmEfp8w+k7CPU2afSdjHKbPPJOzjlNlnEvZxyuwzCfs4ZfaZhH2cMvtMwj5OmX0mYR+n zD6TsI9TZp9J2Mcps88k7OOUaWYSmnHKNDMJzThlmpmEZpwyzUxCM06ZZiahGadMM5PQjFOmmUlo xinTzCQ045RpZhKacco0MwnNOGWamYRmnDLNTEIzTplmJqEZp0wzk9CMU6aZSWjGKdPMJDTjlGlm EpqZX/0PJmNW9dTiJNNfTD7Gztl/dXLsw/z9+aJFxs3/zXyx5s2/FxmLOXLFD1I44gXTTuMZ4ini ceJhwkR8zucmiDHiM6KX6CLaiRaiifiE+Ij4wNhnPGfaxzr7WGcfa+xjjX2ssY819rHGPtbYxxr7 WGMfa+xjjX2ssY819rHGPsNKXCFcaxUYB4xXTQeM7cQLxDpiDfEc8QzxFPE48TBhIj5HP0GMEZ8R vUQX0U60EE3EJ8RHhJW4QnxAuM7zPpFpHDVeNx01dhPriDXEc8QzxFPE48TDhIn4HO0EMUZ8RvQS XUQ70UI0EZ8QHxFW4grxAVFAvE+4zvWOEWMcMMVwzhjqjKHOGOqM4dwxnDuGc8dw7hjOHcO5Yzh3 DOeO4dwxnDuGc8dw7hjOHcO5Yzh3DOeO4dwxnDuGc8dw7hjOHcO5Yzh3DOeO4dwxs+c+QyQTcV/c B66PE8gjgTwSyCOBPBLII4E8EsgjgTwSyCOBPBLII4E8EsgjgTwSyCOBPBLII4E8EsgjgTwSyCOB PBLII4E8EsgjgTwSyCOBPBLII4E8EoxjxmnjTWbn68SrxHbiBWIdsYZ4jniGeIp4nHiYMBGfc9wE MUZ8RvQSXUQ70UI0EZ8QHxFW4grxAVFAvE9kEu8QZ4hkIo6IIVz5hBlvG2+Y3jYiiAPE68Ru4lVi O/ECsY5YQzxHPEM8RTxOPEyYiM9ZY4IYIz4jeokuop1oIZqIT4iPCCtxhfiAKCDeJzKJd4gzRDIR RxwjXLkFG1n0LIvcssgti9yyyC2L3LLILYvcssgti9yyyC2L3LLILYvcssgti9yyyC2L3LLI7dZ9 4Po4izyzyDOLPLPIM4s8s8gzizyzyDOLPLPIM4s8s8gzizyzyDOLPLPIM4s8s+hjFrlmkWsWEyKP XPPIM48888gzjxzzyDGPHPPIMY8c88gxjxzzyDGPHPPIMY8c88gxj7zyyCuPvPLIK4+88sgrj7zy yCuPvPLIK4+88sgrj7zyyCuPvPLIK4+88sgrj7zyyCuPvPLoYZ4RaBSSWyHXupBrXUg/C8mxkBwL 6WUheRaSZyF5FpJnIXkWkmcheRaSZyF5FpJnIXkWkmcheRaSZyF5FpJnIXkWkmcheRaSZyF5FpJn IXkWkmcheRaSZyF5FpJnIXkWkmcheRaSZyF5FpJn4WyeOwk/jn7TdIVcmWLE68RuYjvxArGOWEM8 RzxDPEU8TjxMmIjPOX6CGCM+I3qJLqKdaPniPnB9fIW8r5D3FfK+Qt5XZv98QBQQ7xOZxDvEGSKZ iJudOFfI+wp5XyHfcnItJ9dy8iwnz3L6Wk6u5eRaTq7l5FpOruXkWk6u5eRaTq7l5FpOruXkWk6u 5eRaTq7l5FpOruXkV05+5eRXTn7l5FfOGcvJr5z8ysmvnPzKya+c/MrJr5z8ysmvnPzKya98Nr8t xEaj1kg01ZJjLTnWkmMtOdaSYy051pJjLTnWkmMtOdaSYy051pJjLTnWkmMtOdaSYy051pJjLTnW kmMtOdaSYy051pJjLTnWkmMtOdaSYy051pJjLTnWkmMtOdaSYy051pJjLde+lmtfS661XPtacm2k n41c+0bybSTfRvJtJN9G8m0k30bybSTfRvJtJN9G8m0k30bybSTfRvJtJN9G8m0k30bybSTfRvJt JN9G8r11H7g+biT3RnJvJPdGcm8k90ZybyT3RnJvJPdGcm8k90ZybyT3RnJvJPfGWVo4YGoh1xZy bSHXFnJtIdcWcm0h1xZybSHXFnJtIdcWcm0h1xZybSHXFnJtIdcWcm0h1xZybZn900R8QnxEWGfv 2BbyayG/FvJrIb8W8mshvxbyayG/llmnPEaEEcFE4CzNtBi+Rid5dpJnJ3l2kmcneXaSZyd5dpJn J3l2kmcneXaSZyd5dpJnJ3l2kmcneXaSZyd5dpJnJ3l2snIneXaSZyd5dpJnJ3l2kmcneXaSZyd5 dpJnJ3l2kmcneXaSZyd5dpJnJ3l2kmcn92oP92kPs6qHe6CHXHvItYdce8i1h1x7yLWHXHvItYdc e8i1h1x7yLWHXHvItYdce8i1h1x7yLWHXHvItYdce8i1h1x7yLWHXG/dB66Pe8i7h7x7yLuHvHvI u4e8e8i7h7x7yLuHvHvIuYeeDpDPAPkMkM8A+QyQzwD5DJDPAPkMkM8A+QyQzwD5DJDPAPkMkM8A +QyQzwD5DJDPAPkMkMMAOQyQwwA5DJDDADkMkMMAOQyQwwA5DJDDAL0boydj5DBGDmOcf4zzj3H+ Mc4/xvnHOP8Y5x/j/GOzfz4jeokuon32rhvj/GOcf4zzj3H+sdnJ+QFRQLxPZBLvEGeI5FkyG+Pc U9xLU5x7itqnOPcU557i3FOce4pzT3HuKc49xbmnOGKKc09x7inOPcW5pzj3FOee4txTnHuKc09x 7inOPcW5pzj3FOee4txTnHuKc09R+xTnn+L8M5x3hvPOcN4ZzjvDeWc47wznneG8M5x3hvPOcN4Z zjvDeWc47wznneG8M5x3hvPOcN4Zzvt/Ga774OjsxzPkMEMOM+QwQw4z5DBDDjPkMMP5Z7j+D5me Mz1keoZ4nHiYMBGf8+8TxBjxGdFLdBHtRAvRRHxCfERYiSvEB0QB8T6RSbxDnCHiDE/O4cn6nqzv yfqerO/J+p6s78n6nqzvyfqerO/J+p6s78n6nqzvyfqerO/J+p6s78nanqzrTc7erOfNet6s5816 3qznzXrerOfNet6s58163qznzXrerOfNet6s58163qznTb7erOnNmstYbxlrLWOtZay1jLWWsdYy 1lrGWstYaxlrLWOtZay1jLWWsdYyal2Dfg3aNejWoFmDZg2aDXz81U+Ea0+uPel69rv1RHjr6dL1 2b/MPhjOLDK+9GrrbmMTz9Tqa5y7jadltTfqJ2X1E6gfk9WPon5QVi9GPWOo6r8au40/yeo/oh6X 1U7UI7J6GPUfZPXvUf9OVnejdsjq36L+VFa3obbL6l+jbpTV9ag/ltXQnVEpqy+jLpTVK01BxrOm QHnvBLF3AuW9E8TeCZR3QxC7IVDeDUHshkB5NwSxGwLl3RDEbgiUd0MQuyFQ3g1B7IZAeTcEsRsC 5d0QxG4IlHdDELshUN4NQeyGQHk3BLEbAuXdEMRuCJR3QxC7QVX/EvWHsroUdbGsvoy6UFZfQn1B Vu80hRn+phBD/epkmPGyrN6CepOs3oj6J7LaF/UqWb0S9bOy2gf107LaG/WTsvoJ1I/J6kdRPyir F6OeMULkWRXGrAqRZ1UYsypEnlVhzKoQeVaFMatC5FkVxqwKkWdVGLMqRJ5VYcyqEHlWhTGrQuRZ 5XrmD5FnVRizKkSeVWHMqhB5VoUxq0LkWRXGrAqRZ5XrFbkQeVaFMatUtetVyRxZfR71e7I62BRp 7DNFGOo7QCKN12T1RtQ/kdW+qFfJ6pWon5XVPqifltXeqJ+U1U+gfkxWP4r6QVm9GPWMESFPtkgm W4Q82SKZbBHyZItkskXIky2SyRYhT7ZIJluEPNkimWwR8mSLZLJFyJMtkskWIU+2SNerrvJki2Sy RciTLZLJFiFPtkgmW4Q82SKZbBHyZItkskXIky2SyRYhT7ZIJluEPNkimWyq+l3U52R1mMlshJii 5alpZmpGy6xphjWjZdY0w5rRMmuaYc1oeSKbmcjR8kQ2M5Gj5YlsZiJHyxPZzESOlieymYkcLU9k MxM5Wp7IZiZytDyRzUzkaHkim5nI0fJENjORo+WJbGYiR8sT2cxEjpYnspmJHC1PZDMTOVqeyGYm crQ8kc1M5Gh5IpuZyNHyRDYzkaPliWxmIkfLE9nMRI6WJ7KZiRwtT2QzEzlanshmJnK0PJHNTORo eSKbmcjR8kQ2M5FV9VuoT8tqC+pEWX0SdayhT+QkJnKcPJGTmMhx8kROYiLHyRM5iYkcJ0/kJCZy nDyRk5jIcfJETmIix8kTOYmJHCdP5CQmcpw8kZOYyHHyRE5iIsfJEzmJiRwnT+QkJnKcPJGTmMhx 8kROYiLHyRM5iYkcJ0/kJCZynDyRk5jIcfJETpp9B4A6kZOYyHHyRE5iIsfJEzmJiRwnT+QkJnKc PJGTZt+fo07kJCZynDyRk5jIcfJETmIix8kTOYmJHCdP5KTZd8OpEzmJiayqo1AfkdVxphTjn00W mdZToHWLTOsp0LpF9oYUvMEie0MK3mCRvSEFb7DI3pCCN1hkb0jBGyyyN6TgDRbZG1LwBovsDSl4 g0X2hhS8wSJ7QwreYJG9IQVvsMjekII3WGRvSMEbLLI3pOANFtkbUvAGi+wNKXiDRfaGFLzBIntD Ct5gkb3B9U4yi+wNKXiDRfaGFLzBIntDCt5gkb0hBW+wyN6QgjdYZG9IwRsssjek4A0W2RtS8AaL 7A0peINF9oYUvMEie0MK3mCRaT0FWlfVZtTRsjoK9RFZfRj1QVkdY0ozIk1nDfX7PNOMn8nqMNQh sjoY9T5ZHYj6NVm9E7W/rPZD/bKs3oJ6k6zeiPonstoX9SpZvRL1s7LaB/XTstob9ZOy+gnUj8nq R1E/KKsXo54xzsp+mYZfnpX9Mg2/PCv7ZRp+eVb2yzT88qzsl2n45VnZL9Pwy7OyX6bhl2dlv0zD L8/KfpmGX56V/TINvzwr+2UafnlW9ss0/PKs7Jdp+OVZ2S/T8Muzsl+m4ZdnZb9Mwy/Pyn6Zhl+e lf0yDb88K/tlGn55VvbLNPzyrOyXafjlWdkB03DAs7IDpuGAqjoUdZCsjjNl81SXITtgNg6YITtg Ng6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6Y ITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtg Ng6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6Y ITtgNg6YITtgNg6YITtgNg6YITtgNg6YITtgNg6YIT8xZvPEmCH7ZTZ+mSH7ZTZ+qap3ow6Q1XGm fPwyV3bAfBwwV3bAfBwwV3bAfBwwV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwt V/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0 fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwt V/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV/a0fDwtV34GzOcZUFXv R71HVseZinDAi/LrsUVGpKw+hvpnsjoMdYisDkT9mqzeidpfVvuhfllWb0G9SVZvRP0TWe2LepWs Xon6WVntg/ppWe2N+klZ/QTqx2T1o6gflNWLUc8YF2UvLsKLL8peXIQXX5S9uAgvvih7cRFefFH2 4iK8+KLsxUV48UXZi4vw4ouyFxfhxRdlLy7Ciy/KXlyEF1+UvbgIL74oe3ERXnxR9uIivPii7MVF ePFF2YuL8OKLshcX4cUXZS8uwosvyl5chBdflL24CC++KHtxEV58UfbiIrz4ouzFRXjxRdmLi/Di i7IXF+HFF2UvLsKLL8peXIQXX5SfXYt4dlXVr6DeIavjTKX4fLHs3KU4d7Hs3KU4d7H8FF3KU3Sx 7POl+Hyx7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7Nyl OHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex7NylOHex 7NylOHex7NylOHex7NylOHex7NylOHex7NylaItl5y7FuYtl5y7FuYtl5y7FuYtl5y7FuYtl5y7F uYtl5y7FuYtl5y7FuYtl5y7FuYtl5y7FuYtl5y7FuYtl5y7FuYtldy3FXYtlv6zEL62yX1bil1bZ AStxQKvsgJU4oFV+0q3kSdcq+2UlfmmV/bISv7TKflmJX1plv6zEL62yX1bil1bZLyvxS6vsl5X4 pVX2y0r80ir7ZSV+aZX9snL254SpflmJX1plv6zEL62yX1bil1bZLyvxS6vsl5X4pVX2y0r80ir7 ZSV+aZX9shK/tMp+WTn70xtVv6zEL62yX1byxyr7ZSV+aZX90uXEVtkvK2d/tqrql5X4pVX2y0r8 0ir7ZSV+aZX9shK/tMp+WTn7k4xVv6zEL62yX1bil1bZL13fIWmV/bISv7TKflmJX1plv3T9RC9V vQ31i7J6M+r1svqUyWYkmKpld7XhrtWyu9pw12rZXW24a7XsrjbctVp2VxvuWi27qw13rZbd1Ya7 VsvuasNdq2V3teGu1bK72nDXatldbbhrteyuNty1WnZXG+5aLburDXetlt3VhrtWy+5qw12rZXe1 4a7VsrvacNdq2V1tuGu17K423LVadlcb7lotu6sNd62W3dWGu1bL7mpj/lTL7mrDXatld7XhrtWy u9pw12rZXW24a7XsrjbctVp2VxvuWi27qw13rZbd1Ya7VsvuasNdq2V3teGu1bK72nDXatldbbhr teyuNty1Wn4d2WYEyer9qPfI6t2oA2T1K6h3yOrNqNfL6mOmazwX18uvI18zQmR1MOp9sjoQ9Wuy eidqf1nth/plWb0F9SZZvRH1T2S1L+pVsnol6mdltQ/qp2W1N+onZfUTqB+T1Y+iflBWL0Y9Y9TL Pn8Nn6+Xff4aPl8v+/w1fL5e9vlr+Hy97PPX8Pl62eev4fP1ss9fw+frZZ+/hs/Xyz5/DZ+vl33+ Gj5fL/v8NXy+Xvb5a/h8vezz1/D5etnnr+Hz9bLPX8Pn62Wfv4bP18s+fw2fr5d9/ho+Xy/7/DV8 vl72+Wv4fL3s89fw+XrZ56/h8/Wyz1/D5+tln7+Gz9fLPn8Nn6+Xnfsazq37ZRt+aZf9sg2/tMt+ 2YZf2mW/bMMv7bJftuGXdtkv2/BLu+yXbfilXfbLNvzSLvtlG35pl/2yDb+0y37Zhl/aZb9swy/t sl+24Zd22S/b8Eu77Jdt+KVd9ss2/NIu+2UbfmmX/bINv7TLftmGX9plv2zDL+2yX7bhl3bZL9vQ 2mW/bMMv7bJftuGXdtkvb74CrvplG35pl/2yDb+0y37Zhl/aZb9swy/tsl+24Zd22S/b8Eu77Jdt +KVd9ss2/NIu+2UbfmmX/bINv7TLftmGX9plv2zDL+2yX7bhl3bZL2/+HH/1ubiN52K77K5tuKuq Xov6eVkdZnLgxe2yFzvw4nbZix14cbvsxQ68uF32Ygde3C57sQMvbpe92IEXt8te7MCL22UvduDF 7bIXO/DidtmLHXhxu+zFDry4XfZiB17cLnuxY/a3R6le7MCL22UvduDF7bIXO/DidtmLHXhxu+zF Dry4XfZiB3/aZS924MXtshe7XL5d9mLH7O/3U73YgRe3y17swIvbZS924MXtshe73nHWLnuxY/Y3 bqpe7MCL22UvduDF7bIXO/DidtmLHXhxu+zFjtnfb6t6sQMvbpe92IEXt8te7PpJwO2yFzvw4nbZ ix14cbvsxQ68uF32Ygde3C5/vdhhvCirT5l6jQTTdfm7lHqNSFl9DPXPZHUw6n2yOhD1a7J6J2p/ We2H+mVZvQX1Jlm9EfVPZLUv6lWyeiXqZ2W1D+qnZbXrNwk+KaufQP2YrH4U9YOyerHJ9bsfr8vO 3YtzX5eduxfnvi47dy/OfV127l6c+7rs3L0493XZuXvZ9ddl5+7Fua/Lzt2Lc1+Xndv123mvy87d i3Nfl527F+e+Ljt3L859XXbuXpz7uuzcrt+XfV127l6c+7rs3L0493XZuXtx7uuyc/fi3Ndl53b9 dvrrsnP34tzXZefuxbmvy87di3Nfl724Fy++Lj+79vLsqk/7IaZ9vzzth5j2/fK0H2La98vTfohp 3y9P+yGmfb887YeY9v3ytB9i2vfL036Iad8vT/shpn2/PO2HmPb98rQfYtr3y9N+iGnfL0/7IaZ9 vzzth5j2/fK0H2La98vTfohp3y9P+yGmfb887YeY9v3ytB9i2vfL036Iad8vT/shpn2/PO2HmPb9 8rQfYtr3y9N+iGnfL0/7IaZ9vzzth5j2/fK0H2La98vTfohp3y9P+yGmfb887YeY9v3ytB9i2vfL z2lDPKf1y2zvhO1H5GnvZNqPyNPeybQfkee3k/k9Is9vJ/N7RJ7fTub3iDy/nczvEXl+O5nfI/L8 djK/R+T57WR+j8jz28n8HpHntxPtiDy/nczvEXl+O5nfI/L8vvmanzq/nczvEXl+O5nfI/L8djK/ R+T57WR+j8jz++b3tqjz28n8HpHnt5P5PSLPbyfze0Se307m94g8v53M7xF5fjuZ3yPy/HYyv0fk +e1kfo/I89vJ/B6RJ7KTiTwiT2QnE3lE/vrOtBFimpAn8jQTeUIm6mmIekKeyNNM5Al5Ik8zkSfk iTzNRJ6QJ/I0E3lCnsjTTOQJeSJPM5En5Ik8zUSekCfyNBN5Qp7I09wlE/JEnmYiT8gTeZqJPCFP 5Gkm8oQ8kaeZyBPyRJ5mIk/IE3maiTwhT+RpJvKEPJGnmcgT8kSeZiJPyBN5mok8IU/kaSbyhDyR p5nIE/JEnmYiT8gTeZqJPCFP5Gkm8oQ8kaeZyBPyRJ5mIk/IRD0NUU/I83ua+T0hz+9p5rc+YxeZ Vpn+Ks/YRaZnZbUP6qdltTfqJ2X1E6gfk9WPon5QVi9GPWP8VZ6xi0x/ktV/RD0uq52oR2T1MOo/ yOrfo/6drO5G7ZDVv0X9qaxuQ22X1b9G3Sir61F/LKtrUVfK6l+i/lBWl6IultWXURfK6kuoL8jq PNQ5svo86vdk9buoz8nqt1CfltUW1Imy2ow6WlYfRn3Q0OfgI8zBB0zqHHyEOaiqn0D9mKx+FPWD snox6hlDVf/VeITJpqr/iHpcVjtRj8jqYdR/kNW/R/07Wd2N2iGrf4v6U1ndhtouq3+NulFW16P+ WFbXoq6U1b9E/aGsLkVdLKsvoy6U1ZdQX5DVeahzZPV51O/J6ndRn5PVb6E+LatPoo6V1StNXkwf D3meeDFPPOR54sU88ZDniRfzxEOeJ17MEw95nngxTzzkeeLFPPGQ54kX88RDnidezBMPeZ54MU88 5HnixTzxkOeJF/PEQ54nXswTD3meeDFPPOR54sU88ZDniRfzxEOeJ17MEw95nngxTzzkeeLFPPGQ J4QXE8JD3vNe7HkPmQqWQgVL5H25lH25RN6XS9mXS+R9uZR9uUTel0vZl0vkfbmUfblE3pdL2ZdL 5H25lH25RN6XS9mXS+R9uZR9uUTel0vZl0vkfbmUfblE3pdL2ZdL5H25lH25RN6XS9mXS+R9uZR9 uUTel0vZl0vkfbmUfblE9vml+PwSeRcvZRcvkXfxUnaxvi9XsC995J22gp3mI++0Few0H3mnrWCn +cg7bQU7zUfeaSvYaT7yTlvBTvORd9oKdpqPvNNWsNN85J22gp3mI++0Few0H3mnrWCn+cg7bQU7 zUfeaSvYaT4yma6ATPX7xJf7ZJV85X258qvka+nLtVwlXx1frs4q+er4cnVWyVfn5k9OVDPZRCbr 5lX/g8mYVT21OMn0F5OPsdP92Qfm/ftv6rUn157kgzmfdXLG+Y++ldFDxgP8/eSiVaadf+e1iAdm FT9cxP8tcn1kWjQzM7PI9bl/5Nw/5O/1jzzN0fyHERH+0Fdm+/f/vjOfEFPCt8hnF0ff23wspvJv kY+Zo+9tPrtMA98in20cfW/z2WTyNH3zfFZz9L3N5yffKp9l9zyfp01rvkU+Hhx9b/N58FvlM23c 63zGja3fIp8Bjr63+VQa32Yelhg35+GRe5bPOePot8gniaPvbT7/dzN3ENJ0FMcB/PcU1KVmkB5W HWKnkAjhD0Yww2BgIEqFDMKY1Zpzpm3RKDs4EPLiRGKe1iEWhHYxCPQgSQc9iZfd7KCd9FJKoiFI h/W2/9NSUtq+XyFB/rv8/n7+v/f/v/d+7/1nqwwCnms6mutpUDcBT52Oznr8NE+HntMX7rmlo7me qELaK6gG6fmZhvIzTfb06vlY4Z67Oprr8UDj+2UzvvuJ4+m6IOPpOtlTr7YFmW9skz0Nyg/1P3b/ /IB4P49A9/MI3bMAeRbInk7Ic5vu6VAZqD/MkD3NkOcq3dMAeeroHpdyAfPVGh3N9ZyFPCfpnmrI U0r3pASpd0ZNvcPzfBML8HzR0VzPBORJ0T2VCqkHRbnJngxUn26Km95ebqi92J7s/zoq3DOvo7me j5DnPd2zIy2AZ01Hcz0RqH73m/o9QPO0C7K+2iof6J4lyLNE9kSkVCHtVaq4njg0nsbMeBogju9u aHx3kz0r0PO+aJ73ALE/bIH6Q7anUd4Bz5elo7OeTl8xyXMF8lyke9qg/rDJ9Ic8TwjqD++Y/pCZ n1UoP6tkTyPkseieiJyCxgt7/4t5/zgVcv84yZ44VO/ETL3D87TLOjTfsNdXg7T+uQ3yNNE9ceh+ jpn7med5Bnm66Z5q5YTWW5yK3V5OqL3Ynn7IE6F7xgRZH0uKi+zZgebza2Y+HySuj9VC62NsTxry zNE9k5BnnO6ZgDwpumcM8iTpniTkGaZ7lqH5T9rMf4LE9UMLWj+06M+7BT3vbM8s5Jmie1LQ+uqo WV8NEt8XRcb3ZjP/CdHm8/XKA70P6SF7LqigQt7vDZI9GemD9lP6yJ5JyDN+DJ4ByDNA9mzKEPR+ 5hDZMwt5puietCSg8SJB9iSg+mLQ1BddxPrCguoLi+yZhDzjdE8a2r+YM/sXXcT991po/92er3bT 7ucU1F6jpr14nhWovRZNe/E885Bn5hg8HsjjIXuSUH6GTX56aJ4JaD8uZfbjeJ4xyJOke5bFC9Wn XrInDXnm6J55yDPzh+ecPv4s+pGR3M9MkX0U1/VAz9NANOS/J3//2f1Go30s2fu+oX2+7+TzbYDn y/d4MN99h/YfZSbfmSPyHTL9x0Ofg+bxQR4f2fP40HrlXzz3Tb3C8wzLW8DTr6O5nph8AjxhHc31 vJLPgCeuo7meN7IJeBI6er8HH5+90Phs96+9vnJaveKD6hUf3ROBPJE9z/7+/aXKr393HBgvHOZ8 X/+z8+V7PJjv14d+38lxZL5rcvke0tHZzN8IPwlFQ+FH51WVVEq5jizRv9lPJSTnllzK25nN9u8s vyjOP8snZPd6T+eud0vOSJn+5A30hP2h6PMqqdB/vcJccfm+a9jQ0cW5c/0CDTu3JfjPBQD= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0057_image023.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhZQH7AHcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAABk AfoAggAAAAAAAAAA//8AAP///wECAwECAwECAwP/SLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bz+i0es1uu9/wuBwZ qAfm+Pzpru/7OXV/goMVfAR2iImKi4yNjo+QkZKTlJWWl5iZmpuIUHyGhiChHqOkIqUdqICnI6oa rq9RgQuwGbUXt7isuyG5ha28ole+hMUkxGTILsotzCzOENAO0tNz1CjXXtlC28beZd094d/kD+Pl 6Frnf+vp3NbYsxkCPu0f4d348NgQ9BL++7DYyxMKoACADxC6OzLQRsNWhhAqFLbQYcVC8iRCfHhR /wLHOB8PdawRssnAggwmplw5gWNJXcFMxXRzkqUClQs0thzzUkZPNihz/rMZ4edIBUafKTGI8ygd pxaYXmgKleSwL1SrzkjK0OPUCTr7EbUqDlgvsyA9yhvKlsDBtm5TqZtb7BMFhXhzUg17E6bWY1eb IdWQl09WuLT+CrYS0u7hsXb6Co2QNTLStYpnylo267HXuGMbHAY1ODNgOcTyTl590zBftwYpkBZ5 moe+s5qBdniNN+K/2KFn3+H6maLAjrxFR2z6VnLOUaEiEy9KUJa86RUUDoctV5Vd02hNlqbNIjnl BpGbO0+IHjMty+Bxbx6/wrwFV3vRf8bujb/89/8MgLLWdYEMR6CAANWRIEIBCEBaRoYd4logB1nW yWUGDnidfgEe2OF7AoYI4ogYAkgiefRJKKKKK6p4InwazpZYIvPtsVtKr4FGD3N8PbZdPblJ4R8p 8OkgEW/+HMQjUepNM8uQAVbHGAqqbVAlaFZ65oCW8SU25QlXXrmlaEwyZ46VsZgGZXgkhJljVKKZ 2WGErLGHi3tVrQlknXGWGVqSWDapXpPkvRnaTinmcJtxjL5hqFh6/XnoTT0GtZ6Ystn435dxcElm pKBSGuqWlTI5aXEosknCAFPouakHvEnT426WYmnrrdENmASrQgZmgna0CQqWTYLKCel6dg4W3RP/ vPLqhKuNwgpZZ48Oi+N5lElVJ4O6OsGqs8/O8VpBR446UVjFjjoma1eW4oqepQwArl/3qNpGjnja KuynyF7L7qzbtjdnCPNGKUHB0TjwbY1weOqnuUkS6u+l2l6KwXcfIJxoAxpX0KwnvhLG760Sf7Va yVFVrO/I1aTKwccZwJxMyPMgNiimpv6LbQhUYezBwhoADUTH9oLMgZsTE+owycYmpDJewA1cYtBE H1w1DldHe4aYS4a6L0s7Qqxu0jlGzaFdygiNgdoJBwmBzBuvwgbXJovK56Nfc8fuxM5pt6HLHitQ cC1wn0kD2yVACwMsS1PM983ZPmwxLY3Tdo7i/xfIuwLmct1J595813xy5YPqGOZXRaqBeG1UIOOY zSrfOvLTTa/bL8ttcT5D4ay3evFgSHuNM2G12953n8j6rftxZ7w+Vtd2690vuuVGn+zx/ObdMqqL 9ZJ1pm5r0xnytx8rvex491i99Eqba5mDeObqSz6rliXu0Y9Xr2TkewMacYBa+o7PlGWwX/iAdzRY XprwRyzyiWx0N/Kf9VZGHcsBT2pxO8LqEkgz0V2PfcObnOPM94GelcYxMvJSEzS3AwXaAgPr09f6 2gcClMmOZWYDYUpgRKFcFZAJG9wKGuxTvvNB7m59eZqObsi+QNluX3dIkIta2AoE0qteDPMg0/9O ZbywTS+J50GX6SIWO5W974zboF8WXKiWFDROI7QTCk68eDwxUm43BsmX4faIml6p8IPGc2LoYJgz NNkRdDtjn4fahobvbe8JPhRQa8jzI/jlyh92eIuC4pJJAioLkw6iVEQa9CDQaAiTDTLRZRK0odiA gpWLFIkPpzZFCX1oRi/S0Isoqcs/8jAAvPpRLTGkxyQsK5EylFwTq4VMOuIuaYIUHjT5NcAfmsGK FTQamgJ2KjqWLoeT8+YSawdOJWbTgoxMwaKwyIAgnpMMSOLi7KJZSDKJsWLfJFY8kWWhR3JIawd0 pLPYSIM3riQ2COXOHCkYFwmeL4pZymE//zn/RYJmjADfGyjNmFklqEmrb6Qb5wTpmaz9Re9C0EGn RQkmuMBhdKXgQyQJn5dQPmXPVAsVQVbyc8IUTTSFbFgYTBGlM2kWNXrOrNxBZ4rDgP1vbANDKUUx NtQVsLCqOs3hucSCT5KCUaRfdSY0wQlWhvZUqgKjZRqLtsCXYRRwUAhhE0PnGTja1IsLvSelyCjH T4lTL1KUZU/BV8wrRqFZWCWhUg3a16821LH2bOAhRfLGwP4opVHiIR4S+8EYBlKri5VpzSaLvUQO ykKxTJ0KL7QGaBWWe54N5LZqatOTmYuJPC2ibr3qlssOtprj2Q5n2aqo1342f41dWVflqljc//4m uXlx6GNx9aSfDnZ7wtUNEVhL1OxoVbaOxStvdSJO6R43eWUN4Al9C9dSKgMz0FDjq2zTrZZk5LnJ Y27/7jqppELXdP+VbnT5mdr4uZdF5pAOZrPYQlTkIpW+YeADw+vEHYk1uYIUb5MgSrwlekmSBkyY MHE5YrguYbjooYdroFrIAUs4doScroXBi623OJS1xs1UiQVbYhSj2GCfOx8g0WteJpZWuc11I1uo utYdZ3fB2rTfH2VzX7FROMCQNfJXcopMDze2vLkjUHep42QeUxSS9WBEjFPGzf5W+MqRTW9ZyQco ksm5tmcuyt/YCdC28tk2sAph6ZaK53mSVf+87CmbQhc9XRnT08brUW3b0NZnBktBv4iBE5bD6lw7 ktdgS0P0BgSYY3DYbxEXm+FO/RrnbZpVdJ+mcUnBasOWAZcfTPjxlGP7TDsXObfmVMnpCq3pvNpK 0rUstUxOHASg3mV2M3z0mxu9287ebi/AqfOXJSPqOGMygxzMNRCg/OypfLfXYG60sMHG7TnjTtuc 9nKaXnKL+YXvdz8gN1gE/V1mgg7e/241eh1dZ4cW3Nv7oQ+54xtTJbg2z9YidsTX7OF0d5vaFWe3 l0It7wT7bOHqFM+4uXuXsCnpf09lNKRvZnJV++vke7WyEclo4SPO2HrNgfnNTWq3j49ZDbr/TpFd u0zkdhMb3us+r5LD2J4U1iLo8zUmfE5X079S+OASr5s8t2pnR3dc4Amesj9D3hU0F4qQAOf6ugxu 9Hlm3OuyTXu8ufTLy4gd4iG+zzs7F/XiIvu6/uVq0bNOQYDPndVObfnb2Z5MagtwvQPUt97xbYSH WxPIAPwEtzAoTPjlskMLgigpOekgC2/S85MMPelT2dvmoDGzJ6pQQ6tL+xZJx+4ZUmWyKclLENs9 l6X8vfB3H3xiKntxeEcnZb2m5cKDHdbyZEusew0ptj+el3hvzFPynXwcG3riTm27c71csqTLVKkY jlPdhY9s7Zc93yQv934P73yvG6p418bv/9sng+gi++v6Z+MHUMclF2d+Gtd/+UdoB2d4X9R1CHgf B+Jg4DZqxEV5ldYEYmVsvkZ9OZNbWYZ0iGdyRsd4qnFhmGdrPiU3zNNgaqZFzZdp2aFx/NdQvxZO wsMg1UYqWeYk3vEkyWcGUMdhBDdxchdagRJSSFV+aCdgwGJi3AN0G0VoROdiIGgnVfhMmPZR/AMi zgB1F+gQBeKCQ2d/NkiGteUwBghIcqduX+dAKmWBrTVu1xWDWzZwWNhuWMdlZnhxSOSA0wYIPogq DHdvk9d3NwCAdHh/HZeBdCh+TpN+VwhYMadtGoZhOfeCyncm2eCFE0gWt0aENzJhi8htl/9YgjLI cgXkKUcUDXvmcQGRZn/XNiQIF0X4Uf62M2PIgWOCTzjxiU4Yh5xShkxlimsYgoeShujXiP2wfk/H iYS4ONkwbLNYf4zHX0x1Q9hmW8QIXSb0hr8oJeJADQQ4iuCVh/tHZ4vXhoOGcQL2agGYiUrxjA33 Z/TFQIpYi8fodqsGiSK0be24XNmDg974cyq4d+83cqGYg9fYVDN4gO5WhWbjTBG2ZoTCjJJWEkO1 PMpGJw9oWn7IcUeYkKg4ZNliTk93iCLXbMwwCtM3U6bYZg45Ug14flmlFrAQi93jcCopYWUoaoGH dXYYfWoIfh/IfGSFb9ORkfCnhTFYjFz/tICOOH9DOH5gB5B8h2AGCYd89IQFSY+HGI2ihYeIl4CI lI1F6Y4mlZZ/mE9GtUsQeHxj1zpbcIsLCZMRSYo1uIFJKJCYSDYnaEsG5IxfyGycxoA4xWoGRZez ln6EF14C6Xt354uFGIy+04FE+ZJh+YdUCZNLN2m6d5GdOI9yaQXSKJaQVXVuR33FY5aaGZK3ZZOq dGs4mXdVMIAxxogkWZiMiVTTZnXMd5YzJ5TeEZqpIF9e2ZWXlpl5iSnDpmW5SI7+eJpxVE9/B1z5 wImCqZiyZpfhN5UNaYkYVivKyC16BJryqJMM4xvVaD64aYJNyWLjOUiNaURbmQ6cVV8i/+SBA9ec ksViatePyUhss1kXfhSUqElbvZmaQhl3RMmOMqdlkimYKlBVt/ccYjY18BM24wMjoaRKpdcaHVp6 pYRKO9JJlqRJrAeYKlYoKXoITLEcLQKYOFZgsOdJ7Cc/uGSjFnmhatV7n7khDxJ/4laWVuiHktJi +oebaKktSTWdOHd57/iDWjmaUXZTPJlkRyZjq3hkgWc94rlvRHGSgCOhhpiSeuObXfSR+KOdCmpD BNgUYkqmuFabqamfRqqc4DmUC1pCIkZlcqopVdCclfiacNePeLaaScqPOKcRqLVjC2GbJqOkHDio GCdz3lROUaOk6YJ5kol858mVyBkFpP9lmt85k/kJO1Y6PUzaeCP0YfBIgWVqdlpwlMqIpSRFqTSE ZEeFIp5Cad8omm8QhH65WwjqbnUok5rWqtuZEI2KlRUhrEhqM2NpqHXkn0rnjh/FXsQ5CJBKpOb1 NQWYj65GrP3An9R1fQNKCN2aiF23nZe6m8q6immoLkSEgp06oRVoWFc5BR21dndqqgy1jxN0l3RW Rs8TOgoGl79Km3TKBbTaoAyZpe86sFLJqqmiJRu2rVnZBtAKQeQKnLU2FGxaT7U2jq7qqWIQhO/T QNLKskFZrYcasZ7aE38qpUzwbQHrmk8KZ+0as82UqPR6sLCJsmEwgANqsLmpbsWqTPL/qoPK+moB agLvkq+TeZxmVwpql6uKRrK02K93w4u6uAxTqrHoWZuqhYQ/K1ry6rUgy5cKuTljywUd67S65Vkw Zm7wKX8glC89k65HMbdpmra2VU+t6qYy24TSULPNdgUeuKV5s47UylhHl4q9qrCPyrgUKWhJymvx ak5hihQhG6VU62fLZrWkK6pdhrQ927PVYreceTMUgnp+ug7HNJhm2gVoa6v7pEMuu6ijcq93x2eK G6pV+jgRp7WkM7LcGbx6RruCALipiQr2Ya4WS05he7KRSbYMu1mYm1ZHkkn8CbloeaUCq344yoPO yw7di71Alqvu6qRd5HKztauvSpix/1q1pWsSVfZHiKu1MmtWjcu2+dWWc+gTn2pMRjO1y/eXdcm6 NEXAt+qBmEUcwwurNeIuQaZieKK6obuFfQlI4zO0XQKsDBGGXGgilsRJHqLBpsRhLTp6ffFKMCpK IqLBokchLmIggaV7gDlltldgPwyZulJ8PVzEQAx8P1p8IwZfQlq24Ca/53duAPuUujiRQjbCLwRJ fhuQyceam3o9X+yOA4Qy9zm6JGzBWMHFqfi/AMOe4Yu+srXFUnvA2wcGkqu7D7yzUZxOKbiwc9pH lPm/HqyzHYastTpMnonFBFm8MIil1BtDjcsynQq8XQK9tNi1y9tNoGiTWMuDirzIdv8MGURHwIYr tNTayAVqu2d8umEwLq/bpi1Lk1pHRfd7u3JQvgBcsKZssaCMv7XcPB2EyiyCvHi0p8vaO39hyaOs HwK8ulAMxtdLx+WgzOzpyW+8y7qovG/Lx8l8L6J7qOojuABKcVBan4AmzXFJvGWgPFG0vw78lCEs y8l6Ir28r3BQwZ0jvQras86zx1VLyYAKyA07x4SrRbKSz58MqrKaOKVBRHTDvnFZOQANHtRcQu5s rpXxLpYrwgmdznoQyVRcJ8AVtdqlyqvsBhiNzfs80vqKA/icxR8dy87MRSztyy4NjlRaAsPmDMwk aSTd0R4dLmAyOyv7yiJ9vtEM1Kz/vNBtMjLfUa8JV79K/cf90dAq/bkMPJ96YJymi9JRDdVYjYiN 6SovHbeO4iStEc8zfdQYtKdkjdOV+SsBaITbnNRT3dJMTXZyfaV3bdIj14qmO9FdLURS5tcbu9RO /AIbPScNkrvTjM5BTQeYMz8d3MDljJLcq79Mxgmc3dme/dmgHdqiPdqbkMBCXdZ9HQ9avNipvdWA GIEXUyRyTHIrKduW28Sz29a5fdnNi9RvGSLUoLJyIxxIybyIInnHDdEwMYjEvdw2i4Lae12CzcNx ncWz5Ny0hN3DN3nuMtze/dzsC5ebPW+3ZM+arb0TJdWezHt+nN5FbF+BaNzJDavIyF1B163dE01V 0S3ftszdyh3V6p3I2R2YvE1Y343eCnd8+u3H0r3fDj4E8dVPCi7htR2IFS5ccnzCo3Z7uO1xYo0R PijeFcraqN3aJn7iKJ7iKr7iLN7iLv7iMB7jmaEIaPzgMg4V9X3SN57QOV7PO/7JUJY6MFJR9CxL Jf7jMbBgMtLczY0i8Y3kTkHjqyV2C+50/A3lj81ISF3lvnTlWP4NCjymDc7ewfXlUZ5NSz7mVm7j Zp7ZrAgdfxPfQu7lbV7ndn7nVZEAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0058.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Car-Hill (Performance)
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0058_image024.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC+2dDXgVxbnHZ3ZPEpJgwYpIFTVYUzUWTCICvVTECi1aaMGqVVGEi1HIBySQkETy RTQhH2iI4SNggCCCRsEopUQFlfrkUqymqLfUSvlQoQ9WuDW2otKq3N/s2U1ONufkbM454BfxeZnZ 3TOzM/+Z93fe3Z2zvtmyfblw/4Ut0acZmQ8mSBEphL7iWiF6ir691T717w/Ce4kLYn497qfXSONz UlxUo1Jd/RMjxPrh2ES1X4gFwuMvTIglusd2PvXeQJHR436qPq3KJHK8B4d6YanY1WY54RJiJBW2 Ctvfuy5R1KCLcXzyBjGWf39CE34qpou7RZaYJZLMj50uVB/cddMnoToz2bRo27b9+HfNsqo/12Pp 0p3/+Pjx455NWSjbsheM4cypYg7/ZtKWqWKK6O7f0Wz1b48fhRuK5f3fPiPfj/SVAxn3qH3TyL+N qb9XRh3LnnZnpGipPJo9JsIlhi48mt3HrCmd/SPP7ylypMuwYec1SqOp1Ke2hZnveP4muQHdpw0U 4u0odzosI7JApYcvlbRFVeESB4/sM+rv/+E+0Vr27+yy/kezjxu6HM1+Pz2yQH3G5Z4W4pmefcP6 mO2Zdpc0BvVC2j+ZOu3l75xjDJWI9dDeyp8mRNHFPsbkXydxTJTOKv9Mz6hw1SdlPT4IbEyUFv8Z cOykj8l1l/T9yTdxTK67JOqar+uY4CcRoRiTU/mvRz4BG0xGM75Er3U476uwVVgj9iK2E9uvvrWp k2nVG4vB4rGR2HhsMpaOFWFV2CqsEXsR24ntxz7AmJa9j39A8f0U30nxFyneSPFVFK+ieBHF0yk+ meLjKT6S4vEUj6F4b1VcUFxQXFBcUFxQXFBcUFxQXFBcUFxQXFBcUFxQXFBctZ5euHuz0+xdo9lb 1WuKi3hTP6XZYPN7/efYSPNYmOlL0rTeHsodL3SbfdtbnbHtcUPRRaT9rDhHlTt+XPP0PbX/UvMz 1p+kIUv19s9EUI+VFx556eNcLh/nUvtvwn6IrcNGY6XsTNDdx9aRW8feUpeyBN3X+e/gn2GmXsPM cw3zbH/7+TvMPquOMFv/f23qb29TmJc2qW1f7brYo/6LPfrsqZFHG4rOMcdYnTehC72t/rr89Nf+ F+ulDaq/Q8zzWv1UcWsfj201nfyNfaB98dcmawysNlnbTtokQ6C1k/ZdpHds30VBarZGXTN00aZ+ 7Xnte2Ze+Vbf9v1cR7hN1TWYuhpIr9GtubHjalG05WrrmNo31jZXjpt/401fiHCgZR8ZnJaqvKeW 1nYwWkbL0Gg52axL6dVfdtbSOqZ3oeUNhpaDxBWmni6HuhYHqWuxTdfiEOhaKEM3RwtN7Uq86God 8zdHExxq2Rqklq02LVtDoOWhEM7RQ6Zen3rR8pDDOZoQwBydpAWnqyrvqau1HYyuE7XQzVFVl9Ju itZZV+uYvzma6FDLliC1bLFp2RICLZu10M3RZlOvXV60tI75m6OJAczR4Xpwuqrynrpa28HoOkwP 3RxVdSntRuiddbWO+ZujlzvUsiFILRtsWjaEQMt6PXRztN7U62kvWlrH/M3RywOYo2e5gtNVlffU 1doORtc+rtDNUVWX0u5sV2ddrWP+5uhgh1qWBqllqU3L0hBome8K3RzNN/V6wIuW1jF/c3RwF3PU yfXTNArdEoL+3GnWNZa6ikgn2vpjHXP56U+OiCF+ie/QnwiPe7nqWvzYSbyX6zLzt43dL67eH11w +FI5TuX7j9vfdi/3PR/3cls73csVbfdvg72XG0MbVP1i2X4n99d/2cdsj3UvdzLl1b1ce3n7vdww H/fXw2xj8smXc399vOqTslG/CHRMQnd/vTtjUhvb941QjEmYAx+fx0nHh4j/qq6fqWs90uttPm4d 88f/IbZ2y24yK9h7Po2i430yazuY74USVX8INE4167qEulabMaGnxtaxsC40vpn8QONORbxxzeKy fT9EONR4rxacxqq8p8Z7teA0fkg9ate6p7H3tQLuv7a1Aue41wr4mvPqvOpaR43HQa3jeFjHQnUP TuleEaTuFTbdK4LUvVw989BCEyOouopNLattWlrH/MUI1swOZE4PDVLboTZth4ZA2x+HUNvLTG1H eNH2MsfaDg5I24MyOG1VeU9tD8rgtf27DJ22e6Rb28Oys7Z7pFNthwSkbXWQ2lbbtK0OgbZLQqht ualtrRdtyx1rOywgbUcEqe0Im7YjgtR2oXrWI0NzjanqGmJqO9qmrXXMyTOPhG7eB1G6HhbB6XpY dNT1sAh+zraK0M3ZA8Kt6z9F5zl7QDibswkBfo/VBqltrU3b2hBouyKE2i40tV3lRduFjrUN7Hts dJDajrZpOzoE2l4XQm2vNLUd60XbKx1r2/l7zN/aF0vb+S73Glv7tucai67WSGgd1kgcd7RGQq0J afZow6+l+3okzti3V3scexL7DdaEbcW2Yc3Yy1gL9gb2Z2w3th87gL2HHcFasaPiWe2YWK59IVZp mizUImSOFi3naL3kLK2PHKP1kz/SzpVDtQEyXouVA7Q4+Vc5SDbJy2W5HCqvkMOlS46U74pRcpW4 VlaKcTJXTJA3ipvkGNo7uttzUvW5P30sk+4+b1bPv8y1S2XUVyl/IaqwRXI837vjRB351fKXYq28 Xjwux4onyW8k3yRvFlvkRLFN3iaa5e3iZfItpK/LyeLP2G55h9iPHSB/CDtCvhU7KqeIY3KS+Jy8 pk0S4dodIpq0F+kZWD9tMu2bLAZgsVgcNojjidoU4uo7xHBsJPtGkY5h/zhsAvmb2HcrNon8VOxu LJVj6ezLJJ9LPl+7nWuJSaIUq8Sq2F/DvlqsDluNrcUatNvEk6QbsSY+t4XtF8k3k99B2oK9Tn4X +3eT34cdwA6x7whpK+lHpMewz7WJQtNvF+H6RBGF9SJ/BtYP6892DBaLXaLfJgZhieSHYMP1SeIq PjOKfWNIx5JOYN+N2K1sT+IzU9l3N/kULB3LZDuX4/nYPLZLsQryVVgN+aWkdVg9thZrYN8Gymwk 3cy2mhtvm3NionlvfJL5HF4d38LnXuTzzeR3kH+V/Ovkd2G72d5Hm94lf4j9h9luxT5i+1O2PyeV LnRwTRJRrtvEd8ifQXoW1p98DBaLXcL2QNJEbAj54aRXkV5DOoZ0LOkE7Ebyt1DXJNL/xu4mn0I6 E8vkeA5pPjaPfCnpfCwQX7nP477H6eY6g7NDyofP4MOZ+hfidF2TA/QI+QM9Wl6k95KX6KfLRD1Q /1Yx3DzyU81nekVGXJeoD5UD9eHybH2kPE0fJf+oXSu3aePkSm2CnK7dJMdpt0q8UP5TTpV/kNNk g0yVN8l0CJQlz5W58gtRIHVZDJNK5b9hxVuw4k+wYiesaIEVO2DFdlixBVbsgBXPw4rnYEUjrFgP K9bBinWwYi2sqIcVK2BFHax4CFYshxXLYMVSWLEUVtTCiiWwYgmsWAwrFsOKxbBiEayohhVVsGIh fa2CFVWw4gFYcT+sqIIV98OKBbBiAayohBULYMUCWLEAViyAFZWwohJWVMKKSlhRDisqYEU5rKiA FRWwohxWlMGKMlhRBivKYUUFrKiAFeWwogxWzIcVpbCiFFaUwIpSWFEKK0pgRQmsKIEVJbCiBFaU wIoSWFECK0pgxX2w4j5YUQor5sOKMlhRBivmw4oyWFEKK0phRQmsKIUVpbDiPlhRAitKYMV8WFEC K+6DFffCihJYcS+suBdW3AsrimFFMayYByuKmQfzjLnxQ+meE9mmf6s5MsqYJ/NgSREsKYIlhbBk HiwpgCUFsKQAlhTAknxYkgdL8mFJPizJgyX5sCQPluTDknxYkgcr8mBFHqzIhxV5sCIfVuTDirmw Ig9WzIUVc2HFXFhxD6zIhRX3wIpcWJELK3JhxT2wIhdW5MKKXFiRCytyYEUOrMiBFTmwIgdW5MKK HFiRAyuyYUUurMiGFdmwIgdW5MCKHFiRDSuyYUUOrMiGFdmwIhtWZMOKHFiRDSuy0SQbVmQLf+s/ /cUjQpNGvNOdNZuW7yrepGFvkX9Hc/tu6HizG968STzyF+KR14hHdhKP/JF4pIV45CXikeeJR7YS jzxLPLKJeGSRNkgWa5fDh6EyQRsue2kj5SdylNwsr5Xr5Di5QE6Qd0CJ6+St8vvkesKOT8U0+TeR KreJdNkksuQTIleuhh0Pi2L8vhS/r5SLYEc17KiCHffDjkrYUQE7SmFHBezAp7AmeS/suBd2zIMd 82BHEewohB2FsKMAduTDjnzYkQc78mDHXNiRBzvyYMc9sOMe2JELO+6BHbmwIwd2ZMMO5hLpACwW i9PmwI5s2JENO7JgxxzYMQd2ZMGOLNiRBTvmwI4s2JEFO7JgRxbsyIQdmbAjE3ZkwY5M2DEbdsyG HbNhx2zYkQk7ZsOOTNgxG3bMhh2zYEcG7JgFOzJgxyzYkQE7MmDHLNiRATsyYEcG7Mgw5sZ6fPd1 5kOS6btzzPhdHd+N7cMOYIewI1gr9XyEHcM+xzR9FmyZBVsyYMss2JIBWzJgSwZsSYct6bAlA7Zk wJZ02DITtmTAlgzYMhO2pMOWmbAlHbakw5aZsGUm7JgBO2bAjhmwYwbsmAE7ZsCOGbAjDXakwY40 2JEGO9JgRxrsSIMdabAjDXakwY5U2JEKO9JgRyrsSIUdabAjDXakwo5U2JEGO1JhRwrsSIMdKbAj BXakwo5U2JECO1JgRwrsSIEdybAjBXakwI5k2JEMO5JhRwrsSIYdKbAjBXYkw45k2JECO5Jhx3TY kQI7UmBHMuxIhh3JsCMZdiTDjmTYkQw7kmFHMuyYDjuSGZNk2DE9aHYEei1jsaPGfP6wN+Ts2As7 9sKOvbBjL+zYAzvehB2vmOz4Lex4EnY8AjsWwY4i2JEKO26DHWNgx2jY8V/aKDmQuON84o4o4o5P Ycde2NEMO9bBjgrijpnEHbeYcccPiTu+JwtkNHHHf2DHEdjxDuz4C+x4A3bsgB3bYcc22PEC7Hje jDtegB3Pw46tsONZ2PEc7NgKO7bAjudhx1bY8RzseA52bIEdz8KOZ2HHM7CjCXb8FnZsgh2/gR1P w45G2NEIOxphxwa0Xg87noAdDbDjCdjRADsaYMc62PEo7FgHOx6BHWthxyOwYw3seBh2rIEda2BH PexYDTtWwY5VsKMedqyAHSthxwrYASO1OtixHHbUwY4VsGM57FgGO5bDjmWwow521MKOZbBjKexY AjsWw45FsGMRrFjMPKhpY4eaE3NNdtS0saMGdjwIO2pgRw3seBB21MCOKtjxIOyohh0LYUc17HgQ dlTBjoWw437YQfxFGqMvgB33w44FsKMCdlTCDuIv0quwUdgYvRx2lMOOCthRBjvmw45y2FEBO8ph RxnsIP4izSXWyef4PNJS0gpimyryNeSXkq/D6ol71rKvgfwG9m8kv5n8FuxF4p9m0h3EPa+Sf510 F7Yb20dc8y4xziHyhznWSvoR+z5l3+ek0jUPdhTBjnmwoxB2FMGOYthRCDuKYEcB7CiAHYWwoxB2 FMCOIthRCDvyYUch7CiAHfmwowB2FMKOAthRCDsKYEc+7CiAHXNhx1zYQXxFWurKgx15AV/PK79/ 2fT7eP1E+/1+/P4Afn8Ev/8Iv/8cv5f6ABmux8poPU6eoQ+S5+qXy1iuNwZxvTGQ643LuN6I169l exzbE7jGuYnP3Cr76HfISH2q/EKbJj/UUuVBLV2+oWXJF7Vc2aAVyAe0YpmulcobtUp5pVYlv68t ktFaLdcodXBjteRqQm6Vj8tG+aRcJTfKctkk8+UWOUtuk2kGVV6Wv5ItRCuvw5M/yx/L3XKo3C8T 5QE5UB6SF8sjMla2yhh5lGucY/Ic+Tm80bQzZbh2BmfqRcRzGldFUbKf1kP218LlAM0lYzUh47BB qJKofYbfH8PvP8Xvj+L3H+H3H+L3H+L3/8Dv/4Hfv4/fH8Lv38PvD+H3f8Pv38Xv38Hv38bv9+P3 e/D7Pfj9bvz+L/j9W/j9m/j9Lvx+F36/C7//X/z+Dfz+Nfz+Nfy+Bb9vwe9b8PsW/P4V/P4V5sHL Hfy+xvT7l9v8/mX8fjt+vx2//z1+34zfN+P3L+H3v8PvX8Lvt+H3L+D3L+D3W/D75/H7Lfj9c/j9 M/h9E37fhN9vxu+b8PtN+P1m/H4Tfr8Jv9+I32/C73+D3z+N32/E7zfi90/h90/h9434/Qb8fj1+ vwG/fwK/X4/fr8fv1+P3j+P3Dfj9Y/j9Y/j9o/j9o/j9I/j9Gvz+Efx+HX6/Dr9/GL9fi9+vw+/X 4Pdr8PuH8ft6/L4ev6/H71fj9/X4fT1+vwq/X4nf1+H3K/D7h/D7lfh9HX5fh9/X4fd14shp7rUC fzzH+1oBtU5A/am1Av2pIwaLdS2DEQ/BiGUwohZGLIcRy2BELYxYCiOWwYglMGIxjFgMI5bAiCUw YimMWAwjFsGIGhixGEYsghE1MGIR41cDI2pCem1yPIBrk8dP2LWJnTN74cxePHAPnPkrnHkLzrxJ fLGL+GIn8cUfiC+aiS+2EV9sIb54lvhiE/FFI/HFo8QXDxFfLNRukvnarcQhd8ibtalyFJyJhzPn w5lIOPMx8cU7xBevEl9slqUQpVJW8V+BXCRnyFp5O5y5nr2j4MyVcGYInBkEZy6EM+fCmbPgzBlw 5jQ4EwFnNDhznPjiM+KLY8QXHxNffER80Up88QHxxRHii/eJL94jvjhEfHGQ+OJd4ov9xBf7iC/2 El/sRuu3iC/eJL7YRXyxi/jiT3DmDTizE87shDMtcOZVOPMKnPkDnNkBZ3bAmd/Dmf+BM81wphnO vARntsGZbXDmBTjzApzZCmeehzNb4MyzcOY5OPMcnHkGzjTBmSY4sxnONMGZTXDmt3BmI5x5Gs48 DWeehjONcOYpOPJkh2uTe03ObGzjzJNwZj2c2QBnNsCZJ+DME3CmAc48DmcehzOPwZnH4MxjcGYt nFkHZx6BM2vhzBo48zCceRjOrIYz9XCmHs6shjP1cGYVnFkJZ1bCmZVwZiWcWQFnVsCZOjizAs7U wZnlcOYhOLMcziyDM8vgzDI4swzOLIUztXBmKZxZAmeWwJnFcGYxnFkEZxbDmSVwZjGcWQRnFsEZ 4iy9Bs7UwJkH4Uw1nKmGMwvhTDWcqYYz1XBmIZypgjMPwJkqOPMAnKmCM1Vw5n448wCcqYIdxFFY rGsB7LgfdiyAHQtgB3EU21e5KmFHBexYADsqYEc57KiAHRWwowJ2VMIO4ihXGewogx1lsKMMdpTB jvmMSRnsKPPJDmvdZrh6HnoS122q50cqn7mtp6hfHCFcv9gvcrAeHus2eyR5X7fZOynSnXkqqsB4 l8rpA+Z/TzwVknWbPcfuN+rP+aWjtbS/6mO2x1q3OdZct2kvb1+3Ge5jLW24x1palX//JIxJ61Wn CWst7Zjze9KvqBtUn5TFd2NMvK2lfY+6rTF5zzyPtzHRe3Qek9aruj8mtRF9t1tjMj7tmzEmKl8b EfX7r+uY4Cc3hmJMusoH+64Fz7q8/b7f17l9PadP0rr+LYVnXU6f2d9n1jtad//u7nbbemDrWGQX z+wnG3VNF2nM0Bgxm1maJNLZ9x1Tt+96tCvaTHua1t0xCVS7XOw8LTjtPP3Y8l2Xh+/ep58I3/1y fda4NvkSfdbuu6ou+xx+TbjvY99GlQXmOHeYw+a+vl3M4RzyE8RARmoK45VlzOSZ4i7+vYtxm2XM avfcnmLM79PNNSnRHr+T6Okxh6zUsjCPz4d78YVwL/7gjRnBvENlO/Mz3Pz9VJz5W6ntepy+XQ/X i1zK4vy+QyUiwHeoRPh4h4q9TWFe2qS2/V2nRnTzHSrWeU/2O1RUP8M9fhtibQezfkv1JS6I36tY 2nu2KS7INll1BNOmr5pO1pyx2mRtf9PG7quqU/g3bD6d8rsTP59C9Q47X307QjvOdPjuM6cx+R6z 3h5YFqKcZYvJrWP9uohn8ox4JoZYJomIZZaYQRQzQ9zNnmxymWIauWuMSMcdr6sIJ8ocA91jza21 /ta+L9Lc5/Kyz17W6Zp+1a/YEGtZZtbbT81B8/f5di3VsSg/1zdKy7E2LaNs65N96eXvOtKaz+r3 rZfqnbdP9DqhC9XzPjZ6Yu9hfXT3uoEWvbd+EAvU79+int5djOeJfsZg9au/2a8L2/oVQ79igupX jMN+ec5Zf/PgMzN4tu4f2LdPdHl/9zPU9Xw/6X5fdfvnv93v6Oxn3gPQZQ9D02jj6su4GtSW6CLM ZUxfKfqT/kf7+LjtDsEFY5JS5yRlTp/q676AKj+QT7vL9zZ4IsRW4yxS0y74Ve7szKQ033cP3OWl kba21fOvgNtxYuo7lfpK1e+/P2F6/dzh3aJvtzdadx+UeoM1zfgmuMLYUmuVle/0dcVnxWdZT9yv MG4u/VsLa7t34dZdNz77mWbdT3HvdRl7z3HV6u4C7rL2OR3mWbZtb7htZL2fIcL4rrhQd3V6j6xV t1Wb57nb292jraWK1bF6mNf30drrcvtyDzONsLVQduhdhMG2Ppqqqas2tZdob5MqeQYlS12q5vMN NKotb+1R5/d/Ru9nGE2v28+gtryXE0Gf6SLd80wX+SxnP8PV+ppOn1RHIpV6xuSSmpoX6tgg6Y6s E6MGU0qNRvykCJ8t7yOdtryP9Gy5r3KdWx4tu9PynmbLo6Xyt/hBVwjVdt/tL3bc/uIO7S923P5C GYjyhcb/TSKhC+VbHbe8tUPLWx23/FBAyh8ylE/wq/wkzWn7J2me7fdVrnP7J2qBKD/RIFBiF8q3 OG55S4eWtzhuebMWiPLNmlI+0a/yw3Wn7R+ue7bfV7nO7R+mB6L8MF0pf3kXyjc4bnlDh5Y3OG55 vR6I8vW6Uv5yv8qf5XLa/rNcnu0/y+W0/X1cgSjfx6WUH9yF8qWOW17aoeWljlue7wpE+XyXUn6w TXlf37S36NO6dZbTjLOMNUvlxCTEu88SYV6HHAvwOiTSdl0Tadb3SZD1Rfro93h9XgDz4meUUg0Z MinMx/VC4JFMgt4oPKOyRoeRzBC9pFuRTC+jJ+pXh+rYwPhB8YnmPInw6ad7Nae92Kt59kJtOevF u9pDAcRjuyjVdTxW4bjlFR1aXuG45eq3pN33oGKzlJK/a+2HOu7B0A49GOq4Bz8OqAeXtfdgcNc9 OCid9uCg9OyB2nLWg7/LQHqwR7b1YEjXPah23IPqDj2odtyDJQH1oLy9B8O67sEIxz0Y0aEHIxz3 4Bq5UHT/+2oIpdzXJQldRgqHhdP2Hxae7VdbztrfKgIZgQPCGoEEP35c67gHtR16UOu4BysC6sHC 9h748ePRwvn1f2OH63+nPbguoB5c2d6DDn7cfscpUZ/vslqv7rO475NobU807L242bh3E/e1eW+M v1FRQUusXnTqnTDf8nfC3GXM64RvxPtf/M35YUZfzz5p73Zx6oOn3tvy7X5vywJjXo469Y4W2zta fD0TsZ5phPl4pmH51al3qny736nyquFXd556f8pX7P0pXfu17xg89H596n0nX8f3nbT79al3m3yV 3m3iNN499d6Sb/d7S9r999Q7Spy8o+RkvXfESbx9/KTE26feE/J1fE9Iu1+feifIV+mdIHa/dj9H bw3wOXqU7bl8lFnf+1+x+rynndeceh5tbaOavXTnVRJJPlYqRXX5fOJM4/nEaEqr59U3TE9Lipmd mZQuvkOp74qeIlxEYz35z0lv/LfzPK7fu9vOgW1r49UYqF/Qd3cM3Cqq/l5s9Pc8Lcn4pfeEgXOm pGYlxcy8K+au6bNmZ9L3KZlJpzMi0fQ4zOh3uGHqvzBjf3ibJuE2VdpHq+sVvD3aVvDaf8XsbAVv dKcVvN5+De392zPa1lL7+IRTNrA1u+rM7U/U1JazJ2pxXs7o/QxxHc4Q5+AMzuo98S1XqnqeYbv+ 5Wtz4vvwVVH/mz9vuvubEXuLztSP+Ky5q2+PSw2a9qC0+q3lhJjUpCmzZkyfcXdM9vTMaTHXpGbx XTJrtqKjbj7zjmjLRQrdXE0R6XHU9/qK2ADb2NdoYz9KRxltHGu2Mcp8Dt+xVZ7fyJfq81yW5k7u 8/cwaKy1/SbS2Wzorb+lB3ql035G968VnZ0xposzes4Te2zy2XHne83/s6X4f6+ZBRdQkwAA ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0058_image025.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh7QFMAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAEAAQDs AUoBggAAAAAAAAAA//8AAP///wECAwECAwECAwP/SLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bz+i0es1uu9/wuHxO r9vv+Lx+z+/7/4CBgoOEhYaHiImKi4yNjo+QkZKTlJWWl5iZmpsBDp2boKEuAZ8EnaWiqaokpwqn pKuxshykn6+utVsCs7xjqFq7vcJev0XBGsfDylfFRMkcAs/L00zNQtLIDbvB2NTePdZhyd3f5Tfh ZuTm6zHo7O+U7j7qJ9H019ER9gv7Cv0E//4BPBYwnz+DAwmOQziQH0KGEB8QbOiQwUOIExVqdLjx /6BCixfvyOtxb8UzjB0T6hsHkKPLgxIEliiZDULJkyArWtTGM6YDmh6Azhm5Q2gKhi2hIWWZwSg8 Q0SfSlUUdWoiWAxqad3KtavXr2Cx1qlqFUzYs2jTql3LViwdsjecDvvl1lTbu3i34spFDa4NuWj0 espLuHDesir80gDcxLDjx2rt8hWM+IdiP5AjZ+VaGdDlt4U7S8TIL6lKjy9P2/lshjA8lKkLkpMG WObYI4w9uNaUUnbspfWY5vRn2rYe1i9ybzgLqrdyDEhBOsep03RpqEaeV+iax7hS5yOiD1d9Urwy 5FBgURZ3cXrT39pHx+Y5EcmA+/jz3xdZxuuTbf/jCWReCAWJACBM8713A34M6OfggxAOwF8YnNk1 T3upWUdPfAQulJJ1xPkQ4YgklrifIOixEN96NfjWBXBKmCjjgwvkh0g4KZoUAl8tcjjEgRSBSMSJ BMxI5Cpi3WKKMSDUNYOPT0rX2089GBkhPh+m0cqSWOU4hJMmwAZFREVYOaICDP4FXx6TNaCVFmCC IBOULrgohJkO/gAeJF4CESc0uK1pBJ5H5mDngPE4s9yb4V2zk314/ojoLH3WowGLH9DZAWnXDWEk EHauU2mYl476F4xDmqjDQ9poGoupNvyJjKsTeFemqnGx2mpPoi1ZhazQJUfrCrj2agWsNAC7Kqr/ npa4bJbG0vJjBsouNmwKzhY1abQ7TntBtVFCeqW10nHbjrcWgFtrh7eeCQOGj5qb7BTq+iQnEDO+ 26m8ftKrmHAZBbhtDDLK8BG/RiDbjl/2hBSvkDy42yPC1aA7Qb1WQIjDwBQTofC928ElXBI0Tvxw x0t8nGm6ZJHZro3khogyFSq3UC+zIo775LUz41AzCzenqnHPjfy8AsbNwhyuakRvYXRNIVfAMwhp nto0GU8rSEG1QO5QssFTX/2lo1GvG3YFOquoq9hnZH0C1zrkWefZbKccBdIz6Gdw3UPd7TYGeifH 92pkb+0O3Q5UPffgbBRjS+ESKIt4kUqbNDnj/0UkucCWoJZt7wuBo1Af5m5s+bhkf+94OAyKH0U6 aI93mV7qNYauwuWv0/wE3iXYLnrugdC+ruEU0Fr5UbgD/2sQJQHLsQbHz6R8IcJ/HoHzEFMt4e3J T49F9bqtfkKh3vMCfge8e9B6eN2XD6cTf9IZvYHuX9U58SuNsD7I9VN1/8XN4AYJyJep9vWvC+ej Ev56wiECHpBSjbGGAEPgwAfKIoGlakEFm/I8C/4Bg9T62AY9iKT/XQ9Z+yMhBPW0wJ0oJ4VaU6Ej QLivB8SpaxsYodQMKEMKVUOC2bOADnuYChqyDFvbI+JcfoiCIdqEh0oUgxHpkb4JOFEiUUzUPP8A aJMgUuCKWdTEFLkoHw6AMYyZMGILxwO9JAYLjZNQIxkH6MYLQBGOUkxZpc7oRTyiyIYsjFyf+OhH RrjldCQR5BPNWMdCYoJziOSBOm4oF0I6kiqm24scB9PFNnLwknFUQhUT10jrgZJPosyRJU8ZiTGe UASrZCUqt/hKLAKulLIMBQ0nKUGjgPGOufzCJm0IxFu+MZizPMIoa3dMZM4wCcuknDNfRUsIRDOW 01zELhVpSyHiMpuPrCYxO4m2b3YTnCLhXLqQQMkL/BKdeGgFWYbpJvS8E57phAs9N0NBc+KzD6WY pzjrqb1/QgVTFxsoP6lEE2wa9A2X2aY1Awj/lCsC86FBQGjk2ElRb9oRo7dR53Y4+gEnXhSk/RLp 1pSJHIeidA22EKgka9kBl75USxq9XiInisWS2PSmgUHfTsfJ0C/682RAHcQ+cdFTo1rgpEnlQSTX mbDPmDSqcciFTDNn1aNi9S0W8sTmXEHWmfK0pl79Kh+6pCQQdgNME5TAT9Wax7JmhUuSseuqzspA K3oVqnQ9B6a02lYe+SmAMoPAEAEb2FgJlavq+2tjlVrVyE72OHql6l6JysjLHkc91DIrJxVIj7l6 lhhh/ZZoCUpav0rttGqIKGRbG4GrwjYwjFKtxy5j29uudbeW9S3sDDvSonA2h2kVbmtku9nR/9L2 AaZV7vdyq9mN8TVe5OitdOu6HOA+F7ppZex2M/ol3oZ3vL5gbkbNi1455NSaxnWuJ9sLB/VaV76P ym5y6StFpCUQG3CtYYOOKl7+oiCgoRVCgL2owwIb2AQIxsAw0zdCBz8Ywv4tL1ov/EdDXdeYHA5e c1mL3druN8RZHfFCS6xYf1oYxefQMK/0OzwYcwG0ElawYhps4xvjNcfr7WyPcWuNCE+1RccF8ZDr G1NfOTkuSS6OiZe8XOLi9RU4vi+JO9UNHlN5umAibFipuwMnja4BXv5y43SM3OKp+VjvBWSQ5/vm CWl5xe48cZ2lEGex3hnPGgLvnvOo0oT+ef9zvxgZM2s86Lv9WLeH5ZWQFttoejEVyJaBC6Ur3bY5 K5nTNybzSg+d2jyDmrvVRTJ+FS3NUyNQ1IaG8pYTK41Nu3p33c20AhOLJgLf2tGXivSn9fHr9OTa BxiLbrEt/WekKXvZZc5yqpeGaElPWs/QTli1Ma1qQFNnwG7ONjTvyu1oW7vWvhY3O8kNaalqOt3q ZmmGkf3ueMP00MnGtr2/h29U0AbN+t73sfp9MnSHW+Dvo7eAn2FrhPNb4aY+uMOZoeuIT9xpY2U3 tQEt14BfnKVMlV23V4yNZDz742W+ssg3vm2ImVyyKNejkhAdTRFgpRsvf23MlweOqGh358D/jgEv LQ50s+yNm+UsOqrrBMDZgFvpRu+503v9VKjzGdnqOMbPrV6xnntU51yHn9Dn6NSwI3DsSC+72TEu uIttaNFrT3iZ3051icd93JIU39bvrm13f93ufO8708ne8cBnwWj3ANfJDR/jtutDVsFYPOMdWxTI 1x3wk1dw5fUO88wLPgfV2rvn+zX4tE959LorPU0Lj3pmq6h4zpN8689lKOxZdPZju91TbZ/cF49e YUBR1j1xT/rffUsesic+KzJ+5GSNJPnKV92lH/0XyUE/+h+Q54/BJwDr9x77zr8FW2tu7cfLQwC3 Bz84dH9EtWNe/ZS31PHdD3+8Czbp9bdb/1yE7/H8x38xpdZizeR//0cCA7MLwHJGvod6sBJ87qCA BAh6pBJCrBeB9tcoEvaA/WeBM1ApQIGA6HAfC8iBT4Z2siKCMUSCMpA6wXCCA2AUI/h7PhMOfBSD MmiAGSRoKniBBEKBOriD64aD4QYmVWODFpg13WdllONGQmGEnoeESkg+TQiENfA0f+JQTph5fTJ0 AkiFjSGEq/d0Xth19HM9NHZOY1h8KVAL3Qde3zSFaUh7B6YV2ABDfVR+cXg0a5iAGyhgeaiHGDh9 bviHopRZQYE6cTI/hMhmJRgUFdIScgN2ixhjnUAolmiHaEhsk7iHvXaJn7IBH7iJNtMCU/+ThTfY g6LIdmWYig+3ioHIio1HiqBYdbC4hvqSguRUi29zixOoi7v4ir4YdK4YjLgGhsNIjC/gNhxDd8hY Aiz4fkjVjE3iBKaIex6oXJmRjWsRT714W1GxcnKmU3Qxb4QjPQMYjblEFLFDVliWVyVIGU3GjlxS F0ulA9yXi7tWSDHFHWUlT4TVjuOHOsynHlhGZvUogcYniX64kGi0j/zYJm/CVoLYjjQnZv/oJtwo f64DSshCfvBlZ8Z4WgcZjrdxat/YfBg5iiDZjRJXjWWhjvLYjxEpbRCWkQkpiVN3Sp+gM+uYVwSJ iIlhkxpJi0Hkkt+wk+MyGeKHYz9piyv/qSiypDAe6Wfl+HqNCIgwYIUrKIejsFFiN5JYeTo/g4Bc 2ZVZuZVnmZYTpY1sCRZXiTXxSH06cn44025mmYxoiZdqeZd62TagpX1v6ZQxsRCPtZcqaZhAU5aJ qZj98ZPj15aQGZmSOZmUWZmWeZmYmRZVuZiBeXV9yZlG8zMRJndMNgooeWB6mTWi2Zm/qJKi+ZqG WDqsaXO+MpU6dZe2eZuHmZsfuZsqw5ucBZx+Y5ozx5mJqYRzaJruaDM1I2b8JpzSGJ3SOZ3UCVJ6 4RdgySbZ6UEBRV2OU5A0Nw2oAJ4P1p3b92RJkkmzWURjVWjoZZ6mQ48V2Y/LMJ4ByF/wPVmbYhWX +lmdgveXLZdajIKc/nl2BdofB5qgCrqgDNqgDvqgEBqhEjqhFFqhFnqhGJqhGrqhHNqhHkoNCQAA Ow== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0059.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Multi-Valued Maze
Multiple starting points
Multiple rewards (negative/positive)
50 x 50 states
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0060.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Performance (multi-valued maze)
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0060_image026.wmz Content-Transfer-Encoding: base64 Content-Type: image/x-wmz H4sIAAAAAAACC+3dCXwUVdbw4XtvJYRNQUTEbYwIgiCSIKDjfDIggzoKrozLi/MJKgq8dMjGoiYk IEsSEjobBMKasCWB7AEhgIAgggYRFdxBEVQYRWQEVBzz/qu7Ouk0naTS3aiDhN+hequqW6fufXKq qC7e27ktU9h//Gdpw20Pvn1AiiZCaI3/R4jmok1L/TX97+satRDtAh8ZeEc/afucFEkT9Kmf/leg EHcNJCz660IkCqcffyFmaU7Po1nuIGbpP/AO/dP6PON4vzFvtSBGEX2N+fSF92GBx4TLzwE/EZur iYF8cpAYwN+304Q7xAjxrBgjwsUw42MXCX0b7Mtmm4S+MUOMaOby3PX9Vsa8+vY8SIRK++OTlZWV zk1JllUP293FmkeJsfwdSVueEkNFQ39OjNP/vv9Qc9uzqG/22R63ZfrG52HP668N5/GnhP7zxt9+ HLcpv4nYOf3EuLsC/MTNySfGtTaWtJ3X/6Z/7suw5yd/mSvX9Fol9abexfLGS9tesz2u+TPpknzy PryrEI80sU9vCWsyQZ+2vF7SFn0RfuLg1/tsy7/yu33iWNxP4+KuPDGu0paXE+OOhDaZoH/Gz94t xLJWbSa3NtpT1Eradmp72jWEZbrO//RY264SHZxy73h8gRCxnWrZJ//+FfeJnudO25qxXU2n6Nuk R+Nvze8TPZfCKZenr/nR/D4J8M0+Od2+zSfn4j453b7pvv/WfcI4meqLfXL+8X/H42DiFh4ofUdq dzeg71uJhUQhsZHYRezXf3OzXLpWSyKQCCL6EPcTQ4hQIpawEguJQmIjsYvYT3xL0DVbVn7L7PuZ fRezb2T2QmZfyOxWZo9l9lBmH8Ls9zN7H2YPYvZAZm+pzy6YXTC7YHbB7ILZBbMLZhfMLphdMLtg dsHsgtkFswtm11vPVti3ZpexdYXG1upbzewiyMihnrcexu/2e4g+xnv+xniSRrR0ylxljD1cn7tb Zofq2iG2I9O2+m5yzFdZqZzHn/56F+Mzjh9JQzK06s/ARKzjsXB6LGtZl18t69Jff5i4QTeD6E9M 5cVgzf7eMh4t49WpfnoEa7Wt/wm9/xn5usVY1y3O7a9ef43e51iGv8v2P2Lk37VN/m7apD+vrV2d nJbfyWmbnXPk1IbYK4x9rK83uI58O7bXr57tdf3p4KYN+vb2Mtbr2E69dm3t9FzvTvXte0+3pb42 OfaBo02O52baJH2QazPt66jVbF9HL3O2WD9uqKNNbasfq8uMx/rYalP9OscS9tCX1YNl5TLtpzn6 xva+InZdX8d7+msDXPpKpfFzvzEWAkzkspn0Lpf6/M65dDz3JpdK+iaXQ4xl6fm6WJ6ZS8d7Wh25 HGTL5Y2ip5FPP5N5DfUyr6EueQ31QV4t0nd91GLkLsJNXh3v1ddHg03mco+Xudzjkss9PshlhQ/7 aIWRr0/c5LLCZB8N9qCP3qq8y6s+v3NeHc+9yevNynd9VF+Wnrvb1Jl5dbxXXx/tbjKXi7zM5SKX XC7yQS4zlO/6aIaRrxw3uXS8V18f7e5BH/XXvMurPr9zXh3Pvfr9pPmuj+rL0nPXWHPz+0kz10dv MpnLkV7mcqRLLkf6IJdDNd/10aFGvsLd5NLxXn199CYP+uguL/O6yyWvu3yQ1wof9tEKI3dvu8lr hck+2sNkLnv6eZdLfX7nXDqee5PLG/x810f1Zdl+J/mdmUvHe/X10R519FEzx0/DmekxH2zP08ay BrCsWKaDXbbH8Z5fPdszXgRSvwTV2J4Ap/O5+rH4j7/i+Vy9Dfr53McH7Bd99zeb0PJ6+ZD++MqB +6vO597yQlO353P78HrR803OWIMvzucG0gZ9+WLOfjPn2D9r7WiPcT53CPPr53Nd53c9n+tfyzl2 f5d9cuq3Ocd+QN8mPf52r/l9crbOsTdkn0xq36a9L/aJv4kxPpGV3u8j//Vl3cmypjB90GWMO96r z/9eLu2WDTTL23M+haLmeTLHc29+L8zVl++jHOvLup5lZennBV1y7HjPV+d89FweVt7lUp/fOZeH lfe5PKq8z6X7awWEqHGtQFf7tQKOvH+m7Hk/ps7M+2fKd8eIet6zvcx7tkves32Q9+XKd30408hl rptcZirf1Yl6Lgd7mcvBLrkc7INcDvFhLh8ycvmkm1w+pMyZazaXrb3MZWuXXLb2QS4v82Eumxu5 vMJNLpubyOUtDcilfs7Om1zq8zvnskJ6l8t0Yrf0PpePGsvaKu25fEfWzKXjPVXved+gqhrfTD6n eJnPKS75nOKDfCb4MJ/RRj4T3eQz2lQ+uzcon729zGdvl3z29kE++/kwn72MfPZ3k89epvLZo0H5 PCW8y+cpUTOfp4T3+fxZ+C6f+jURej5/EWfm81thJp+9GpTPQi/zWeiSz0If5LPMh/lcbuRztZt8 LjeVz1salM/hXuZzuEs+h/sgnxYf5nOIkc/RbvI5xEQ+uzv9PqrvWh7HtQzTSFh37cznzteM1HXN h6pxzUelqWs+9GtcrlT2dYYI+7/F6MewscY1LnlEAVFCrCbKiZdFN+0V8bPaJj5Rr4uP1JvibfW2 yFJ7xR3qQ3FI7hMp8oB4UH4hWsojYo84KgrEcbFAnBQzxWkxg5YlC03OEAFyjmgmZ4oWMlVcLJPF pbx2hUwUV8sEca2MFx2JLnKa6CanipvkFHGzfFH8RU4SfyX6EXfKiWKAjBH3y1gxiOmjMlo8TgyR L4inieHEKPm8CGUaSYwnoohY+ZyYTMTJ8fx+HCeSiXRithwr5hNZcoxYSuQS+UQxsYpYKyPFBmIz j18ldvB4J7Gb2EN8ICPEJ0w/Iw7x+DDxDY+PMz1J/ET8QigVIRqpcNGUuJBoRbQhLldh4k9EO+I6 ojNxI9Gd6KVCxa1Eb6IfcSdxD3Ef8RDxCDGYeEKNFk8RzxL/S4wmIohxxAtEDPEiMY2YTliJNGK2 ChHziEXEEiKH0PvG99LeJ0YY52z0PjLW1k9CxEqiiCgj1hDriU3EVmK7sogK4i3iXeJ94mPiUzVK HCQOE98Q3/HaCaY/Ev8hpGYR/too0YS4gLiI55cwvYzpVUyvITpo/yuuJ7ryOJjoyeNbid7E7Ty/ g7ibx/cSDxIPE/9D/H/iSeIZ3h/JNIQIJ8YSzxMTiEnEVCKBsBJpRAYxl1hILCaWM/8KpoVEKfES sY7YSGwhXiPeIHYR7xDvER8RnxIHia+0keJr4hjxPfED8TMh/EYKP6Ix0ZxoSbT2GyHaElcSgTxv T3Ti8Q1Mg5j2JP7M49uY9iX6E38nBhIPEP8gHiP+SQwlhhEjCAsRRowhniOiiYlEQ01/xPDHcS5t gzENqMOPDajmyXp0pw4wLSY+dXLqAMs7ILpoX4hD6oiYqI6KYHVcHJAnGcWnGe2V4gmpyf4yQF4t m8mWsoVsIi+WQl4qf8Gb03jzA96cxJvjeHMMb/6FN1/ized4sx9vPsGbj/DmA7x5H2/24s1evNmD N+/gzdt48xbevIk3O/GmAm9ex5sdeLOdFmzHm9fwZhvevIo3r+LNVrzZijdb8GYzLd2MN5vwZhPe bMKbTXizCW824s3LeLMBb9bjzXq8WYc35XizDm/K8aYcb8rxphxvyvGmHG/W4s1avFmLN2vwZg3e rMGbl/BmNd6sxpvVeLMab1bjzWq8WYU3q/BmFd6swpsyvFmFN2V4U4Y3ZXhThjdleFOGN2V4U4Y3 ZXhThjdleFOKN6V4U4o3pXhTijeleFOKN6V4U4o3pXhTgjcleFOCN6V4UoonpXhSgicleFKCJyV4 UoInJXhSgicleFKMJyV4UownxfSDYlvfWC/tfWK54VSxUVPo7x8kDhPfEN8RJ4gfVRHeFOFNMd4U 400x3hTjTRHeFONNMd4U4U0R3hThTRHeFNHnCvGmEG8K8aYQbwrxphBvCvGmEG8K8aYAb/LxpgBv CvCmAG8K8KYAbwrwpgBvCvAmH28K8KYAbwrwpgBv8vEmH2/y8SYfb/LxZgXerMCbPLzJw5s8vMnD mzy8ycObPLzJw5s8vMnDmzy8ycObPLzJw5s8vMnDmzy8ycObPLzJZZDl4k0u3uTiTS7e5OJNLt7k 4E0O3uTgTS7e5OBNDt7k4E0O3uTgTQ7e5OBNDt7k4E0O3uTgTQ7e5OBNDt7k4E0O3uTgTQ7e5OBN Dt7k4E0O+2m5qO+a2fpqHv1iPH/H49g+Da559O91rTRV8wRR83TRtonrtNeJN0U77W3RTNtLDfSh WK72iTB1QPxVfcGoOyJWyKNiNJr0QZWrECYAg05R8xzm0cfUPO9Q8+yk5tlOzbMNg7Zg0GYM2oRB GzBoHQatwaDVGFSKQcUYVIRBhRiUj0ErMWgFBuVhUC4G5WDQcgxahkFLMWgJBi3GoMUYlI1B2RiU hUGLMGghBi3AoAUYtACD5mPQPAyai0FzMSgTgzIxaC4GZWJQJgZlYlAmBs3BoNkYNBuDZmNQBgZl YFAGBmVg0CwMmoVBGRg0C4NmkY1ZGDQLg2Zi0EwMmolBMzEoHYPSMSgdg9IxKB2D0jAoDYPSMCgN g9IwKBWDUjEoFYNSMSgVg1IxKBWDUjEoFYNSMCgVg1IxKBWDUjAoBYNSMCgFg1IwKAWDUjAoBYNS MCgFg5IxKBmDkjEoBYNSqmqeJfSJycL+71Mzq2qeFPpKMkYlY1QyRiVjVDJGJWNUMkZZMcqKUVaM smKUFaOsGGTFICsGWTHIikFWDJqBQTMwaAYGWTFoBgZZMWgGBs3AICsGWTFoBgbNwKAZGDQDg2bQ G2dgUBIGJWFQEgYlYVASBiVhUBIGJWFQIgYlYlASBiVhUBIGJWFQIgYlYlAiBiVhUCIGJWJQEgYl YlAiBiViUCIGJWJQIgYlYtB0DJqOQQkYlIBB0zFoOgZNx6DpGJSAQQkYlIBBCRiUgEHTMWg6Bk3H oAQMSsCgBAxKwKAEDIpncMZjUDwGJWBQPAbFY1A8BsVjUDwGxWNQPAbFY1A8BsVjUDwGxWNQPAbF YVAcBsVhUBwGxWFQPAbFY1A8BsVhUBwGxWFQHAbFYVAcBsVjUBwGxXltkLfHXWGmj7uCMSgYg4Ix qBsGXY9BbTHIX/tQHMWg3RiUh0GTqIPupA66lDroGwzahkGZGBROHXQvdVBX6qAW/KnEoGMYdAiD 9mHQXgzahUGvY9CrGLQRg9Zi0CoMKsOgYgxagUE5GLQEg7IwKAuDFmDQIgxaiEELMGgeBmVi0GwM ysCgmRiUhkEpGJSMQTMwaDoGxWNQHAZNw6CpGDQFg17EoEkYNBGDYjEoBoNiMGgCBkVjUBQGvYBB z2PQ8xg0HoPGYdA4DBqLQeMwaBxbPQ6DxmPQeAwah0HjMGgcBo3FoDEYNAaDxmBQJAZFYlAkBkVg UAQGRWJQBAZFYFAEBkViUCQGRZLVCAwKx6BwDArHoHAMCsegMAwKw6AwDArHoHAMisCgCAyKxKBI DIrAoAgMisSgSAyKxKBIDIrEoEjMiaQfRNY47hpl1DORVQZFYlAkBkVgUAQGhWNQOAaFY1A4BoVj UBgGhWNQGAaFYVAYBoVhUDgGhWFQGAaFYVAYBoVhUDgGhdGLQjEoFINCMSgUg0IxKBSDQjEoFING 0+NCMWg0PXA0BnFMyrQ3cTtxB3E3cS/xIPGwFoJBIRgUgkEhGBSCQSEYFIJBFgyyYJAFgywYZMEg CwaFYJAFg0IwKASDLBgUgkEhGGTBIAsGWTDIgkEWDLJgkAWDLBhkwSALBlkwyIJBFgzimJM4SHxF fE0cY1nfM/2B+JkQfhYMsmCQBYMsGGTBIAsGWTDIgkGjMGgUBo3CoFEYxDEn8WfiNt7vy7Q/8Xdi IPEA8Q/iMeKfxFBiGDGCsPD5MKZjmD5HRBMTCU+Oh/7mdNy1w5g2NenHDi+OwY4z3UJ852TWcZZ3 nBrpJD3ntPiXqqQnajJbBcjnVTP5iGohu6mLZWN1qfxaXiF3cSRWJK+V82VHaZVd5GTZTb4gb5KR 8mY5Sv5FPi3/Kp+Q/eRj8k75sBwgH5T3y4FykLxLPipvl4/LXnKIDOJTN8jhsjNzdJChsh1zXy3H s/Qo2VbGcnQ3WbaWcbKVTOSYLxnt0lFvNsd+82VjmYWCS6W/zCXypZLFWLhK/gdrfsaan7DmR6z5 AWt+wJpTWHMKa05izfdY82+sOY4132HNMaw5ijVHseYbc9cKXGC/VuAn+TUu/QuXjuDSV7j0JS59 gUuHcOkQLh3EpQO49BkufYpL+3FpPy7tx6VPcOlDXPoAl97HpfdwaS8u7cWlvbj0Hi7txaW9uPQO Lu3Gpbdw6U1c2olLb+DSG7j0Oi7twKXXcWkHLu3ApR24tB2XtuPSdlzajjuv4c5ruPMa7mzDnW24 sw13tuHOq+ztrbizFXe24s5W3NmCO1twZwvubMGdLbizhT6zxdaPVkp7/1lp+LbFuPZPf/8E8aN6 BZc249JmXNqMS5twaTO9azMubcKljbi0EZc24tLLuPQyLr1sO7vQk7iV6M3z25neQdxN3Kutx6UN uLQel9bh0npcKselclwqx6VyXFqLS2txqRyXynGpHJfKcWktLq3FpbW4tAaX1uDSGlxai0trcOkl XHoJl17CpTW4tAaX1uDSS7j0Ei6txqVVuLQal1bhUhkuleJSKS6V4FIJLpXiUikuleJSCR2mBJdK cKkEl0pwqQSXSnCpGJeKcakEl4pxqRiXinCpCJeKcKkYl4pxqQiXinCpCJeKcKkIl4pwqRCXCnGp AJcKcakAlwpwKR+X8nFpJS6tZL+s9O3xmcnvITrXRnNMH58510ZB1EZB1EZdqY06Uxtdo+2j1xwQ ftoX9MIj4k1qowJqoxnqJKPiNCOoUlyFVz+gwjsoUYoWGfJiJLlUjkCUh5GlH171wquueNUer67C qzZ4dRFeNcerxnxC4VUltdFpaqNT1EYnqI2OURv9i9roK2qjQ9RGn1Mb7ac2+oTa6CNqo/epjfZS G71LbfQ2tdFb1EY7qY0qqI0YhxwjzpevURttozbaSm20hdroFeMc0ct4tQGv1uPVOrwqx6u1eLUW r9bg1Ut4tRqvVuFVGV6V4VUZXpXiVSlelWJQCQYVY1ARBhViUAEG5WNQPgblY9BKDFqBQXkYlItB uRiUi0G5GJSDQcsxaBkZXIZBSzFoCQYtwaAlGLQEg5Zg0BIMysagLLKdhUGLMGgRBi3EoIUYtBCD FmDQQgxagEELMIhjU2IesYhYQuSo+TWOz+KM47OsqtpoPn1lPkbNx6j5GDUPo+Zh1DyMmotRczFq LkbNxai5GJWJUZkYlYlRmRiVSe/IxKBMDMrEoDkYNAeD5mDQHAyajUGzMWg2vWk2Bs3GoNn0rgwM ysCgDHpcBj1wFgbNwqBZGDQLg2Zh0CwMmoVBMzFoFgbNxKCZGDQTg9IxKB2D0jEoHYPSMCgNg9Iw KB2D0jAoDYPSMSgNg9IwKA2D0jAoDYPSMCgVg1IxKBWDUjEoFYPSMCgVg1IxKBWDUjEoBYNSMCgF g1IwKBmDkjEoGYOSMSgZg5IxKBmDkhmcyRiUjEFWDErGoGQMSsYgKwZZMciKQVYMmoFBMzCIY1Pi z7x2G9GXx/2JvxMDiQeIfxCPEf/0S8KgJAyagUFJGJSEQUkYlIRBSRiUiEGJv/nx2USPjs+CMSgY g4IxKBiDgjAoCIOCMChIO0IcJew1UxA1U5BWSWiym8bxmdZM3qC1IC6WnbRLZQftCtleu1peo10r r9Y6yqu0LvJyrZu8TLtJttJuls20v8gA7a/SX+snNe1OqbQBUmr3y1/UIPmzelSeVo/LH9QQeVw9 Lb9Ww+UXapQ8qELlPhUpP1bj5XsqSr6tYuVONVluV3Fyq0qUG1SyLFXpMk/NlsvUfLlIZcn5aqmc rXJlisqXiapYTlar5AS1lhpvg4xQm2WIelU+q3bIoWqnHKx2y3+oPfIB9YG8R30i+6nP5P9Th2RP dVjeqL6R16nj8nJ1UrZQP0l/9Yv8SSr1nWzEu0359IXM3UptlG3USj41V/5Jxct2aixzDZOd1UMs ob/srm6SvdS18lbVWvZWjVjDCQw6hEHvYVAFBm3FoPUYVIJBuRi0GIMWYtAcDErFoCQMSsSg6RgU j0HTMGgaBk3FoGkYxLEp03k8X6QmY86L9IOJNY7Pooz6JbnKoIkY9CIGTcGgqRgUj0EzjHNEVuMc URIGJWJQIgYlYtB0DJqOQdMxKAGDEjAoAYMSMCgeg+IxKA6D4jBoGgZNw6CpGDQVg6Zg0BQMmoJB kzFoMga9SI97EYMmYdAkDJqEQZMwaBIGTcKgiRgUi0ExGBSDQTEYFINBsRgUi0GxGDQRg2IxKAaD YjAoFoNiMCgGg2IwKAaDYjAoBoNiMCgGgyZgUAwGxWBQDAbFYFAMBsVgUAwGTcCgGAyKwaAYDIrB oBgMisGgGAyKwaAYDIrBoBgMmoBBMQzOCRg0AYMmYNAEDIrGoGgMisagCRgUjUHRGBSNQdEYFIVB 0RgUjUHRGBSNQdEYFI1B0RgUjUFRGBSFQVEYFIVBURgUhUFRGBSFQVEYFIVBUezfKI+Oz6Kcjs+U 8X3jth77EYwfwfgRjB/B+BGMH8H4EYwfwfgRjB/BmpKeH9eV632ax+u0auvKWd4GGaRtxqRXsWiH vFbbKQO13fizR16qfYA/n8jm2meysXYIfw7LSsb2T4ztE4ztbxnbRxjbB5RSH6tGaq9qqnapC9V2 1Uq9otqodepyVar+pPJVO7VYXacyVWdlVTeqF1V3NU71UiPUrepx1VsNVP14dKfqou5RV6n71EXq IZb4iPoeZT6XT6i35VNqK/KUSOoLOVqlodFUOU69IF9AuRjMeFENltNwY7q6W1pVX5mm/oxkQXKe 6oxs7eUSdbXMUZfLlaqVLFLNZRmerKEqXK++Z+x+w9g9zNg9xNg9wNj9lLG7j7H7MWP3I8buB4zd 9xi7exi77zJ232HsvsXYfZOxu5Ox+wZjdwdjdztj9zXG7jbG7hbG7mbG7mbG7ibG7ibG7kanY5iX jWOYdYzdNYzd1Yzd1Yzd1Yzd1YzdVYzdMsZuKWO3lLFb7PRvTPmM3XzG7krG7krG7grG7grG7grG 7krG7krGbh5jN5exm8fYzWXs5jJ2cxm7uYzdHMZuDv0gx9Y3Dhp9IsuwLse4Pl1/fyOxhXiNeENb zthexthezthextheytheytheytheytheytheythewthewtheQmdczNhezNhewthezNhezNhezNhe zNhezNjOZmxnM7azGdvZjO1sxnY2YzubsZ3N2M5mbGcztrMZ29mM7WzGdjZjO5uxncXYzmJsZzG2 sxjbWYztLMZ2NmM7i7GdxfZk+fQYp9KDY5wdHh3jeFpfBFFfBFFfBFFfBFFfdKO+6EZ90Y364kbq ixupL7pSX3SlvriB+qIL9UVn6ovrqS86UV90or7oSH3RkfqigzaIeJTa5HHZThtCbfI0tclweYU2 itokVLbRImVrbTxGRMkWWixOTJZNtThqlUTqlGTqk3TqktnyJPXFcUbuMeqLr6kvvqS++Jz6Yh/1 xYfUF3upL3ZTX7xJfbGd+uIV6osNVAhrqC/KqC8KqBiWU19kU1/Mp4LIwKBkDIrDoBgMGo9BIYjx DAYNwaCHMWggBvXFoJswqCMGXYFBzTFIYNB31BcHqS/2UF9so75YQ32RR30xW95JdXSPel7ep0Lk Q+op+QjV1WBMeUL1xqAeGNQJg67CoFYY1BSD/DDoP9QXp6gvvqO++Ib64ivqi4PUF59SX3xEffE+ 9cW71Be7qR921jjGmW4c4+yuqi920ldep77YbpyH2YpRmzHqZYxaj1HlGLUGo1Zj1CqMKsOoUowq Mf4dvAijCjGqAKPyMWolRq3AqDyMysGoHIxajlHLMGopRi3GqMUYlY1R2RiVjVFZ9LhFGLUIoxZi 1AKMmo9R8zBqHkbNxai5GJWJUXMwag5GzcGo2RiVgVEZGJWBUbMwahZGzcSomRg1E6NmYtRMjJqJ UekYlY5R6RiVjlHpGJWOUekYlYZRacYxThoGpWNQGgalYVAaBqVhUCoGpWJQKgalYlAqBqViUCoG pWBQCgalMDhTMCgFg1JcjnGSMSgZg5IxKBmDkjHIikFWDEp2OsaxYpAVg6wYZMUgKwZZMciKQVYM smKQFYOsGGTFICsGWTHIikFWDLLWapDju5GN9O8S/QrfjbxrafOqaxn170amd7tQLJoZIPzu3S/G E42dvhs5pI/770YO79PU/qCoqe1+qo91uGbEZaJI9Fla/d3IPktr/27ko26+G3nXUvt3I5sP2G9b /vj7TH1f9YvWRnty3rZ/N3KA8d1I1/ldvxvZqJbvqzZy+r6q/vjIr7hPnL6v+qW+TXoENWCfOL6v 6sil/n3VX3uf3NakTeq5uE9ua9I07b91nzBOvvLFPqnrsbf3M3Relrt76NW27tquFx+m6r5fgfOy zF47PtlYbn+Wu5XpP12+c+t4r0kd144PsS1rhLDQQwNFBL10mAjltQuNvLVyalczY9rciIbuE09z p593vU95lzvncewYu35OY3eydjbG7m87ZvWf33LMuo5dfVmufbixtF9r0JlpkrGfnftwpPHa1XX0 Yf2aur62/RTOPnqW6dO0+QHRldeGsgfH2Pr2aPEMfz/Dngy39XN7bx9q6/H274Y0q+5fVX28uVPf ush4r5nTPQyc33dMHeHv9PlGbsZQIzfjyJ013tzfVL+3zUllv7dJZ+M+JhVaZ61CO6li/fToXO/9 TQM8vL9pQC33N3Vtk7+bNunP6ztODmjg/U0d6/2172+qb+dJp+8WO557e2+kzg28l0R5I/u1AiMv cH+tgH6dgM0ry5nt76zVbH9nL+874ViGp9/p+r3k1LlNjv7laJPj+W/ZprOx736veTqpzq3+dH7c nf3+5Kv70de2bcc1+/0azbTDbN2/yVjuD2zjkyQlwKXud7x3cR01k15X9bPVRvaaP5y6KITaKVA8 KZ7j74erKif78YPmdE8zP6epu9cdj5sYj50/YyZfl/g4X28Zy9XvTzmGhl3qJl/6e23qyNf4evI1 iPoylBhFPMsrnuYswCU0l+dm8tfBx/mLM5bblggy7kXomr+2xrWgdR1nPkCeBpC7oWQvxMhfUzfb 7edm2+s7nneMef3fi7poZz4/29/vaW/c8/I06zyq3/9Qs18H8o5WqY5olR7buJ/lVNbxe+1sXxfn 2K7mmn27WldtV0vtCOHNdrXUxG/2b2GO7brS2K72VdsVyHYFerVdgSa3y3ks1te/fzYOshznp1yf n+356ztfpp8vaivt/+9Y9ef/2P/PSlvjHJMmG9ty2sxWk9jOGqhZmvD3s3VfKa7Uz6+ok5UuZ6Da 3TVs1NhhkSOequ28kz5/Vz5tn7+lzUkh1tvWIpVq99BzEZHDLLWfnbLPL23TY1XL+bfH7Tg7yzs/ rW06336fHW1AA85I/rFHpONMlZ7BHkrZfhv0tD27QdrrjDZ+QWOCxgjjN0VP2wnMn5R/1Xkue+41 22d/Vo5zb/ZX/WyvXuE3W7PPYJ/XtV/7O89b9Wojl73rfg0Btt8X7TW/M/4/IMeyHUtzXnd1uxtX tVT3uoPm7/b/FXJdln08NzamAS4tlDW2LsDmW2ulL6muNlXPUd0mfc6LmXOqn77kq2086s/ctUdf f/1rdL+G/mx19Rr0Z+7nE16vqaPmvKaOtc7nuoa+2uIzPqm/00TPnq1zSaX3C/29G6X9SKt70x7M pe+NoMKAWlveTJpteTPp3PLa5juz5Uo2pOXNjZYrqY+3oBt7Cr3ttbc/1HT7Q2u0P9R0+y3Sk8xb bP8zaHAdmd9juuV7arR8j+mWV3iU+Qpb5oPrzfytymz7b1XO7a9tvjPbf7PyJPM32wTqXkfmF5lu +aIaLV9kuuUZypPMZ+j/OYboXm/m/TWz7ffXnNtf23xuxqzmSeaVpmf+pjoyP9J0y0fWaPlI0y0f qnmS+aH64bm4qd7M7zLd/l012r/LdPsrPMp8hS3zPerIfE8/sy3v6efc8p5+Zlt+g58nmb/BT898 D5fM1/ab9jFteIPWcoFtLQOMucYHBgfZ1xJgHIv86OGxSBOXY5smxvJOebm8JrVs9/3aRL+G94s7 mUtvSK9C/1qOGTyvZPT7EDlXZYUmK5le2lwPKpnrmavuSuawMtvyw8q55fozcy0/qjxp+WfK3vLa fx9lm255do2WZ5tu+XKPWp5ptLx2VQabbvngGi0fbLrlQzxq+UNGy3vV2vLWplveukbLW5tu+WUe tby50fJbam15hTTb8grp3HL9mbmW75bpDWp5U1vLtzKX/gFd2NraPsV026fUaPsU021P8Kjt0Y62 d6+97b1Nt713jbb3Nt32fh61vZej7T1qb/spYbbtp4Rz2/Vn5tr+s/Ck7d8Ko+29am97oem2F9Zo e6Hptpd51PbljrbfUnvbh5tu+/AabR9uuu0Wj9o+xGh796qxWn22q7s2zc9xBkQ/x2M/R6Oq/kXF tf16gdFBiz1/r94/+L16j9nOH449f1/ec/S+vPU51sq2/wPqvOdufctwWHL+frp/7PvpvmvrS8PP 3zv3d37v3Nr+Xcjx7zr+xjUzDa8Zzt/r9o9wr9vqmuH8fW3Pxfva1u2DN8cU5+9D+0e4D221D+fv OXsu3nO2vuOBtrb939T0/WTNHl+cv1es/V6xf9T7v1bY+tXT5+/1+ju716up4wk314k1rF44f2/W c/XerNX1wvn7sJ6L92H9dY4nzt839Vy9b2q1D+fvkXou3iO1vvq/h23/tz3r9z81exxy/t6mf+x7 m9r+D2Ot+zlzH1Mz9Xul1/X7+fuOnqv3Ha3+/Xz+HqPn4j1GXX2wXzN9zMNrppu6XIPd1Fjekd/Z 8txPz/x+ofO7x6p0dJ37zCvih9XyrZSmdV4JdontSrD+zK1feTpohGVYYETksFBxIXO1Es1FI9GM aM4fM1tTfzvvU3Gioe3sWvVdaH0f6Hfka+g+sGdR395etu29T0Xa7hzXd+yw8KHPDns68IGuY4eO GjMscPQzgc+MCI+IJAlDI4eJFmx9cxFg2/7mtmxcxDP9NX/juR76H3/b642q8tXIJWPVe7Lub3I2 rvomp+udz8x9k7PZGd/kdHcHNfe/oZu5tNR1351UFZpn393U11x9XaP+zNx1jZ3dfA/G/Ro6a85r 0J/VtwZzyz37Ldez6ryGCu23z83Z34bfS/bP/X7T0PsHuLbIXzte65Lr+s3S3ibtD+q4pt9vqt+o MfxeCR8R8mzgk88FPqxjK5oITQRUfd+r+rH+dxP+drzirlWXeNiqTrZWNWbuNme0atDo0NGjRj/7 XF3tCjD+aMbUfes6eNi6NrbWtWVu/QrtBwIHDBsaHkLrmlat2a9GXqqrhy7aRD9HHzBz7VVj228H VXUfInO9s1Lt1zz915nqNdrvEGRujS01c2usrHON9nv3mFtjYB1rdB4prpXbz5XmX7XfRUKI/wMO KfxQJrQAAL== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0060_image027.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlh+gFVAXcAMSH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACH5BAEAAAAALAAAAQD6 AVMBggAAAAAAAAAA/wCAQP8AAP///wECAwECAwP/WLrc/jDKSau9OOvNu/9gKI5kaZ5oqq5s675w LM90bd94ru987//AoHBILBqPyKRyyWw6n9CodEqtWq/YrHbL7Xq/4LB4TC6bz+i0es1uu9/wuHxO r9vv+Lx+z+/7/4CBgoOEhYaHiImKi4yNjo+QkZKTlJWWl5iZmpucnZ6foKGio6SlpqMBp6qrIams r7AXrrG0tQ2ztrmxuLq9qry+wajCxKTAxciax8nMlcvN0Iqps8/R1oYB1Nfb0tMK1dzhe9nZC+Tg 4ul66OqGAu8L7wLx8Ary9PP29QX3+vn8+/rx48OunQSBCAMq/JeQ4UJ8EP1FBOiw4sSGFx9K3Egx /6NFjv/yFLT0sSNIjSZTYhSxUiXKliEt7DMYYSSclzg/CnQAM2CDmBFK9ixJU4pNLkPjeUA4YaeG mTmLzjn6pOVToSil5qLqAyvQpUq1quO6Y+YKs2JPJD3pNew6JE5TxBXVk8Fal23ZTjxF1sVctWjJ yAtMw2qFr2kz9JXbjuicAZAjO4hMecACyAR5EAaMGGnWFgmDVB5NurTp06hTqx49zsZfFJ09f9js 1rACywxW697Nu7dv0ucCVA5OvHjxzIB70N4S+zJm55UFUH4Xmfrv69h5ezDOPQCB7+DDix9Pvrx5 AovLpIfQeTma7PB5S98NnXKByirImaPWnZx4///nBSjggAQS2NpZbr3QnAm43Rbfg6vV10N/FJ5T 4IUYCsjKehQsCJuHGkBoWgkg2lDhid1lqCED4I0FGhEQ3ifZDSVugOKNFE5wYWIfcGiXawvCB9cE OBaJYgc78jiCjwmeFVJnqllhZJExDKhkC0zaE8NXqHVxJBAG7sFUjW9k+eIDXRr1ZRIBMmdXXnjt xRMgZqow02lPTHkOmTeUVxWcUPFljhKlRVHhYUX4CURUWm5w16OM0qHfNyPwiQFpTvQHVhCKAtme XqDG+cqkBTBp6QSFMpGjFeaBttBgr8JaBU52kHoOCKdCUN1zSGhqxTzzmBcppF+5FwSgLyVSJwj/ 0xWxZmHDRtqpa2VFa8qyHsyYa36HIhoqsS2MJ8NnCjpWDLYbVOeDrx0qRMK2HIhXA5ntRctQOqZm ixm8JBj3rbEswSCvDlC5K+eVEuSbAX4D5RBcSma0eCZF8DjFL8IQKHwBw64VdzEVEuMaK8UfY9wB uhLMCO0/xL0Rskf4vNaoyTygrCuv5TLQMhMlRyCxzDRTYfNkDf5I4r37Pcyzo3mF92HPQfcbMAU4 00NiTeVEASLSFbwcNRg+Blm0CcuR6gXALH5HNtpf1xA21WO/m/CtXWzrddtgu1B1CAsqLRuSBFyN 9xFDOzgnB+6ZLUU/QF+gdqVQD75C4Xs3fBja/4ovPvMIj88muapToxm35f3SrbnRJHTeAduf1xw6 0YgrlnkSBqugeutivH2zo7JrXeXtGLCO+7orVN5h7zxHbsHdFyg/fAw2793z7Mdy3bwIwD8fBsqV l0z9oriGkL32eYcv+uHsYZC1EI0HL37g5G8/qPmwXy9L4UwLwbxM8Suxfql82x3/7gc+/Y3PW/3r 1aS4N7oe/Y9GwqvBARGYQCPY6nsdkN5PZIe/oDiPghqYYAW5gC4NNokCD3SV9WaQKxGOcAsc4tLY nqS+FIqhhfB7IRliWL9W2JAxVXBhU3TYhPXIUAQ//BCrcrg6IoLuBN3TQBLtIMQhOnEJy9Igmf+m yJIPym0DVbyilKAYN9ZxMYBGsFQYxVgFI/bwhERClxdhEy8msvELbmyA8QiYHC+s8QFzvCMK8ugc Bwbij+gTZK8YVEY+fjEMiNygIpNQJxOi0EwRVA4Y7ThJEjIykXN7gxo52UktEFJGgKzAGe3nuxCS spRZOOUeL/k6MkRyc7B0ludy00BaojFirwRhLolgJktijRC3JN0wiVkCY2asg7Rz5TK9UMwZ9i09 gSyMNKfpyU0VUpLHXF02tZmBZHLzBuVABzb12MtwCiKS4zxnxgCozl1+05HvDGZQ5Gki09WkmTOM IzL1uU9+zmBnCQMoOB+wyoJmgU/mNKgMqmH/K4WijqF9iee8yklQiU7In99YH0hZabiL3mIxGmUh Rz26SG/c4qXeRCU+AQFPlh6UniiMqUxVCc0uIDKlNqVUqeo5gqp1pqFU7Cgog4olv/2zqAF9Zuzq hoGIMtUFNsmSUd3pPqo6TqlXpaROtyrVp5ytqmANq4mESqSxtvOkg/gjUMOqH6JO9Z6+PGRacalW LA01p1A1qc5sMlckrLGwfbUAk4yKGKQilpxfTewUFmvNsh4vDGSyqmRToI22NhGvfA1pHGqk2c2e AKFPvetOBZrPrprWr7KI6SwH21rXvtZ1IiCrA8742CPI9bawZQdl34RRYZ41ssD9qEsTmtui/wWG t2Ug7V6Ti9W/ela1uqXtTdBK3Qmpz5uz3U9tB9hdzo50t58taV7/EMbe0tWp8wxBdrWbvjoctry4 ZZZzlSna+kYXufjFAVIBKN/9AmWK7qXW8qYbYG5Rr7O8A+1XBoxZADcYncx1RUZ5yVO9kvTCMFju hkFr2T7cF8QmmOIFKfzGxi4jwUY4MYpLAF+GNpe4cE3lGUBU2hlv57sFVsqBX4wGHjPYxyle5YjV y1cWf8HISF4rW1NrVhLvtqdXaG+UaTw/5kaYySXmg5a3XDrh3ji0P4RxjNOqZhD7iFdoSXORF0xm XQZ5IEPW8Y7pXOchLHm1ed5u147c5xCnd/+1xfWDjAttSLtWGcxXZu9e26xWbzhaX6G1rmDlMGZG gwCDXf7Ac54LDEqrdNCeFgKH4DzhUs8Z1akGAllCgrNWi5LPsW4qqAn8aEQneg+LzrUUd83rDND6 rU5eQ7CFfb/zwlTUDbL1QuewbGaf9s4AifS0x+ChHlt7SdiW9qYrTAFvf/uZu541h9GM5SJ0m9Dn ju1yqWzb2SY7DdWOd0LnrbNQ17ud98b3pPUN5BRWVL/sXiocfktwHxJ41biJs6v3XG54NzyOmVvx YkYd6HGD4d0Xnxxgew1nbbMB5CFPcrq/DGnxinngKVeszEkO8HZHgeExH/l1Mxhtk8PRDfn/zvnJ EN7kiQsG10KfKNH5m0JTwwDKSUdiOnVO0pLneA1Qj3qPXO7lf4cS2GzWenynPHavh/kOQY86Li6d rp7T9+dtSLvaib0eq3Md7l6tuNiVDu0mNf2/sN67oWnud5sXUO+C306N0Uv4s9MB54nvMtspuMeA 23K6TjcthOnddlzaMPM7gHzkQ2pmTHve6ORG/Oira3pl/v2GFl69yB9t9/6+WvWy58DUd87Kyhue CaJfvaWpTl7f5yH4wldy30+vcGVjPvdJA3V6aq9pvF9ButDX/aCAsfnOZ/vtmcYC9rNvo+1fnSv/ GLXPuc1d8tsoa7wQKbHdYny0P9/9rC02/3l9fXdq35/8i/dsHEB9DwR6OIB8wodTvBciERcTBfhx sYd/MsdvNsZzhdd819d+Elhwk2dFO+WAv4cD47eBzXZGVJF+vWR5YCB3JCiAxZeCIWhYMNeCqtR6 ILhtasCCNOhvC+N2/WeAj4R0OyhQHWgBBMgLQFhLuDeE+6Z/Feha6gd+T7ZSTMhTpQeFDaSCPvV/ Vch5l8V/1RcHCAh9AdhvFrg5r8cGY+h+RVgBHCeFSahTQtiFGLVyAxhtIZGGGbhNdLheZTdARwiB fNiH6DZzXWU8WsgFayh7GmaI9oOIMZgoXCiB89cXtdZ//HUGP8WEJniG3xeGmdhKg0iIn//mifug h1NwKuZGiCeoXnkYiSyAQ6RYimTnLZDIaQwWh1G2dlV2iwt3ZLqIZAtkipgYimKwibNIX86WMg1Y jMH4hZuUjFJzh/rgjKcDONIIbliIhrBIR9iYjUhkbFaWiEvEcuDIgx5YOeQIMoT2jH1miT6IimNw S+5YZ/BYjaBojEcAL6sojff4fZ8HBXZjcefIeI8Yj90YhNFYkOF4iKOzjlJAjwxpkID4kAmZA8lU j+9IjWEhjysIbxppjxwJLLZnfccCAv14jmShfvngke72ASmpkuIoU694jd84kdp4kL+GBuYUkiLp kDsJTHKIkyu5X2HokzBpcUhJZlwRhS3/iYRKwC8xWZBNGXFa4pInmZQ4eW06SToQOS7vs5UpBpTr N48EuZQ/WZFBeYxnKZZPeCkB9YAfaU9bKVLeR5JH6W4XM5XgaJcMqCXzgJVlgT0E6ZZYs0r5sC9S +AV8WZdkuZh+VJhoWWiExX9fuWa/ZJhmCJcbJJg0QpiZqZnoiCpWyQ8B+QMf05huSRVOeZpdAZqi OUhdyTIXGS6FGZteWF97c5kGpIS4ORKJGTe8CSa3GX6ayZoz5JmQdZP0g5ubuTGWwRDKeWrM6Zxc yZnSiXoEg5LFaZ1/eBCGU5OK2J3e+ZZWtJu1aQLeNpm5VhAoWJZBRJ7s2Z5/aRfTOTEe/9Bj80mf PdgwAuCan5mfqumdyOmfALpRAkqe5Zmb6KOYJcmOC8p6X1g1w2kDq7ifwsYOx3Z1EKqQETqazGif 2rklSamgGMpsR+Gg+YhY/Dig5dl9oKSiB6og3PmhweWB4VmMwOeiCwqj00ah6blJCmqc1il/DQWk Ilij3mij5gk3HEqkZKOVS8qkFOmGA0CbeoafrsSjVFqLggUZWIqBPeA0XQo94MlOYboEU3mi+lYN MYGkajqkOHYNelKndopaeGATOFOh7VJHCTQSu5eb3OeCxFcrnGmNsVid2uMKI8JWdmlp0yBiIoZT ftl1eQqXaepxzemnFcSojcofkTpUt/9iOpGqYY1YV1wUpMzkUA6SqSaZP0qqFu1QEKgaqqhKT/zh qH91qwtoqKwqI0+ZqJyzprPKLVX6YwfiX+rFpyVDplMqDqpag8lqpTrqm3Ukp2kBqBRIqLX4YIVa ByOxpwEAFG22P39qiI+Kq7sqXrwoqpTiUvwWraoGnbRJrqHHpcUqqggFqrv6MKTqru+qrrVaffIa BM/wple6H/Z6gKrJpn1gqv7irrc6sZoGf9s3senUiCA6Fb+asA9aKdyJrTjoIuZ1rF4qWvGHHFkq rrKaoH6RrZmisqQJmXSJVubqRNq6sU54nSJhhA1SDsXSa1uKrzBbAK0iVBpLqaO6ra3/ILMpg6VB 2zwDKbJt4wrngbRIW1cSO39S5LSoA6b9tbC2MzBPdyW0Kn/6SqnqyrOXqmPNko+2pSNtsiUOKwhD w5sFK2vpYx/VylzvNi3fyVkvAE0dVLiw9US+urN4xbUvhRhHqxiDG7mSe7gsUDh3erl2egeVyk64 8TAqkiHIKjjcKpuTW7mlmx+DJwku9bmse7U5CbJNKrini7oSOnu1CwlGirm6u7u827u++7vAG7zC O7zEuyqJS7uKmyeUK7vJO422m7flRwl8al34Y7kny7ayWb03Wrl366OPwLTgZqqmO75w67zIO73W +7E01r00uwjoW6rLm73vy70Aa6yTjAO/QlO+Zbq//Nu//vu/ykI3XAG9hEDAV7R2/sR98MuvwhCq KEpb8Fps6aqxBuwGvDi9iYXACqg42jCMzVsKHfzBG8muD4dep2qGFdwGF5zCOAvBvPY/aAt/GAzA Z8DCNOw/N5zDOrzDPNzDPvzDQBzEQjzERFzERnzESJzESrzETNzETvzEUBzFNJgAADs= ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/slide0039.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT)
Future Research
Is it a hack?
Learning what works in a problem
Forgetting options
Large state space =
– Approximate clustering
– Local clustering
– POMDPs
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/master03_stylesheet.css Content-Transfer-Encoding: base64 Content-Type: text/css Ym9keQ0KCXt3aWR0aDo1MzRweDsNCgloZWlnaHQ6NDAwcHg7fQ0KLlRCDQoJe21zby1zcGVjaWFs LWZvcm1hdDpub2J1bGxldFwyMDIyO30NCi5UDQoJe3RleHQtYWxpZ246Y2VudGVyOw0KCWZvbnQt ZmFtaWx5OkFyaWFsOw0KCWZvbnQtd2VpZ2h0OmJvbGQ7DQoJY29sb3I6YmxhY2s7DQoJbXNvLWNv bG9yLWluZGV4OjM7DQoJZm9udC1zaXplOjE1MiU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1r aW5zb2t1LW92ZXJmbG93OjE7fQ0KLkJCDQoJe21zby1zcGVjaWFsLWZvcm1hdDpub2J1bGxldFwy MDIyO30NCi5CDQoJe3RleHQtYWxpZ246bGVmdDsNCglmb250LWZhbWlseTpBcmlhbDsNCgljb2xv cjpibGFjazsNCgltc28tY29sb3ItaW5kZXg6MTsNCglmb250LXNpemU6MTMzJTsNCgltc28tY2hh ci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouQjFCDQoJe21zby1zcGVjaWFs LWZvcm1hdDpidWxsZXRcMjAxMzt9DQouQjENCgl7dGV4dC1hbGlnbjpsZWZ0Ow0KCWZvbnQtZmFt aWx5OkFyaWFsOw0KCWNvbG9yOmJsYWNrOw0KCW1zby1jb2xvci1pbmRleDoxOw0KCWZvbnQtc2l6 ZToxMTQlOw0KCW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5C MkINCgl7bXNvLXNwZWNpYWwtZm9ybWF0OmJ1bGxldFwyMDIyO30NCi5CMg0KCXt0ZXh0LWFsaWdu OmxlZnQ7DQoJZm9udC1mYW1pbHk6QXJpYWw7DQoJY29sb3I6YmxhY2s7DQoJbXNvLWNvbG9yLWlu ZGV4OjE7DQoJZm9udC1zaXplOjExNCU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1 LW92ZXJmbG93OjE7fQ0KLkIzQg0KCXttc28tc3BlY2lhbC1mb3JtYXQ6YnVsbGV0XDIwMTM7fQ0K LkIzDQoJe3RleHQtYWxpZ246bGVmdDsNCglmb250LWZhbWlseTpBcmlhbDsNCgljb2xvcjpibGFj azsNCgltc28tY29sb3ItaW5kZXg6MTsNCglmb250LXNpemU6MTE0JTsNCgltc28tY2hhci13cmFw OjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouQjRCDQoJe21zby1zcGVjaWFsLWZvcm1h dDpidWxsZXRcMDBCQjt9DQouQjQNCgl7dGV4dC1hbGlnbjpsZWZ0Ow0KCWZvbnQtZmFtaWx5OkFy aWFsOw0KCWNvbG9yOmJsYWNrOw0KCW1zby1jb2xvci1pbmRleDoxOw0KCWZvbnQtc2l6ZToxMTQl Ow0KCW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5ODQoJe21z by1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5OMQ0KCXttc28tY2hh ci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouTjINCgl7bXNvLWNoYXItd3Jh cDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLk4zDQoJe21zby1jaGFyLXdyYXA6MTsN Cgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5ONA0KCXttc28tY2hhci13cmFwOjE7DQoJbXNv LWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouT0INCgl7bXNvLXNwZWNpYWwtZm9ybWF0Om5vYnVsbGV0 XDIwMjI7fQ0KLk8NCgl7dGV4dC1hbGlnbjpsZWZ0Ow0KCWZvbnQtZmFtaWx5OkFyaWFsOw0KCWNv bG9yOmJsYWNrOw0KCW1zby1jb2xvci1pbmRleDoxOw0KCWZvbnQtc2l6ZTo4NSU7DQoJbXNvLWNo YXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLk8xDQoJe21zby1jaGFyLXdy YXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5PMg0KCXttc28tY2hhci13cmFwOjE7 DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouTzMNCgl7bXNvLWNoYXItd3JhcDoxOw0KCW1z by1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLk80DQoJe21zby1jaGFyLXdyYXA6MTsNCgltc28ta2lu c29rdS1vdmVyZmxvdzoxO30NCi5DQg0KCXtmb250LXNpemU6MTMzJTsNCgltc28tY2hhci13cmFw OjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouQ0IxDQoJe2ZvbnQtc2l6ZToxMTQlOw0K CW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5DQjINCgl7bXNv LWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLkNCMw0KCXtmb250LXNp emU6MTE0JTsNCgltc28tY2hhci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQou Q0I0DQoJe2ZvbnQtc2l6ZToxMTQlOw0KCW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1v dmVyZmxvdzoxO30NCi5DVEINCgl7bXNvLXNwZWNpYWwtZm9ybWF0Om5vYnVsbGV0XDIwMjI7fQ0K LkNUDQoJe3RleHQtYWxpZ246Y2VudGVyOw0KCWZvbnQtZmFtaWx5OkFyaWFsOw0KCWZvbnQtd2Vp Z2h0OmJvbGQ7DQoJY29sb3I6YmxhY2s7DQoJbXNvLWNvbG9yLWluZGV4OjM7DQoJZm9udC1zaXpl OjE1MiU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLkhC Qg0KCXttc28tc3BlY2lhbC1mb3JtYXQ6bm9idWxsZXRcMjAyMjt9DQouSEINCgl7dGV4dC1hbGln bjpsZWZ0Ow0KCWZvbnQtZmFtaWx5OkFyaWFsOw0KCWNvbG9yOmJsYWNrOw0KCW1zby1jb2xvci1p bmRleDoxOw0KCWZvbnQtc2l6ZToxMTQlOw0KCW1zby1tYXJnaW4tbGVmdC1hbHQ6MDsNCgltc28t Y2hhci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouSEIxDQoJe2ZvbnQtc2l6 ZTo5NSU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7fQ0KLkhC Mg0KCXttc28tY2hhci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouSEIzDQoJ e2ZvbnQtc2l6ZTo5NSU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93 OjE7fQ0KLkhCNA0KCXtmb250LXNpemU6OTUlOw0KCW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2lu c29rdS1vdmVyZmxvdzoxO30NCi5RQkINCgl7bXNvLXNwZWNpYWwtZm9ybWF0Om5vYnVsbGV0XDIw MjI7fQ0KLlFCDQoJe2ZvbnQtc2l6ZTo5NSU7DQoJbXNvLW1hcmdpbi1sZWZ0LWFsdDowOw0KCW1z by1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5RQjENCgl7Zm9udC1z aXplOjg1JTsNCgltc28tY2hhci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQou UUIyDQoJe21zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5RQjMN Cgl7Zm9udC1zaXplOjg1JTsNCgltc28tY2hhci13cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZs b3c6MTt9DQouUUI0DQoJe2ZvbnQtc2l6ZTo4NSU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1r aW5zb2t1LW92ZXJmbG93OjE7fQ0KLlRibA0KCXtmb250LXNpemU6MTE0JTsNCgltc28tY2hhci13 cmFwOjE7DQoJbXNvLWtpbnNva3Utb3ZlcmZsb3c6MTt9DQouVGJsMQ0KCXtmb250LXNpemU6OTUl Ow0KCW1zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5UYmwyDQoJ e21zby1jaGFyLXdyYXA6MTsNCgltc28ta2luc29rdS1vdmVyZmxvdzoxO30NCi5UYmwzDQoJe2Zv bnQtc2l6ZTo5NSU7DQoJbXNvLWNoYXItd3JhcDoxOw0KCW1zby1raW5zb2t1LW92ZXJmbG93OjE7 fQ0KLlRibDQNCgl7Zm9udC1zaXplOjk1JTsNCgltc28tY2hhci13cmFwOjE7DQoJbXNvLWtpbnNv a3Utb3ZlcmZsb3c6MTt9DQouZGVmYXVsdEINCgl7bXNvLXNwZWNpYWwtZm9ybWF0Om5vYnVsbGV0 XDIwMjI7fQ0KLmRlZmF1bHQNCgl7dGV4dC1hbGlnbjpjZW50ZXI7DQoJZm9udC1mYW1pbHk6QXJp YWw7DQoJZm9udC13ZWlnaHQ6bm9ybWFsOw0KCWZvbnQtc3R5bGU6bm9ybWFsOw0KCXRleHQtZGVj b3JhdGlvbjpub25lOw0KCXRleHQtc2hhZG93Om5vbmU7DQoJdGV4dC1lZmZlY3Q6bm9uZTsNCglt c28tZmFyZWFzdC1oaW50Om5vOw0KCWxheW91dC1mbG93Omhvcml6b250YWw7DQoJY29sb3I6Ymxh Y2s7DQoJbXNvLWNvbG9yLWluZGV4OjM7DQoJZm9udC1zaXplOjE1MiU7DQoJbXNvLXRleHQtcmFp c2U6MCU7DQoJbXNvLWxpbmUtc3BhY2luZzoiMTUwIDAgMCI7DQoJbXNvLW1hcmdpbi1sZWZ0LWFs dDowOw0KCW1zby10ZXh0LWluZGVudC1hbHQ6MDsNCgltc28tY2hhci13cmFwOjE7DQoJbXNvLWtp bnNva3Utb3ZlcmZsb3c6MTsNCglkaXJlY3Rpb246bHRyOw0KCW1zby13b3JkLXdyYXA6MTsNCglt c28tdmVydGljYWwtYWxpZ24tc3BlY2lhbDpiYXNlbGluZTsNCgltc28tYW5zaS1sYW5ndWFnZTpF Ti1VUzt9DQphOmxpbmsNCgl7Y29sb3I6IzAwOTk5OSAhaW1wb3J0YW50O30NCmE6YWN0aXZlDQoJ e2NvbG9yOiMzMzMzOTkgIWltcG9ydGFudDt9DQphOnZpc2l0ZWQNCgl7Y29sb3I6Izk5Q0MwMCAh aW1wb3J0YW50O30NCn== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/script.js Content-Transfer-Encoding: quoted-printable Content-Type: application/javascript; charset="us-ascii" function LoadSld() { var sld=3DGetObj("SlideObj") if( !g_supportsPPTHTML ) { sld.style.visibility=3D"visible" return } if( MakeNotesVis() ) return runAnimations =3D _InitAnimations(); if( IsWin("PPTSld") ) parent.SldUpdated(GetSldId()) g_origSz=3DparseInt(SlideObj.style.fontSize) g_origH=3Dsld.style.posHeight g_origW=3Dsld.style.posWidth g_scaleHyperlinks=3D(document.all.tags("AREA").length>0) if( g_scaleHyperlinks ) InitHLinkArray() if( g_scaleInFrame||(IsWin("PPTSld") && parent.IsFullScrMode() ) ) document.body.scroll=3D"no" _RSW() if( IsWin("PPTSld") && parent.IsFullScrMode() ) FullScrInit(); =09 MakeSldVis(); ChkAutoAdv() if( runAnimations ) { if( document.all("NSPlay") ) document.all("NSPlay").autoStart =3D false; if( sld.filters && sld.filters.revealtrans ) setTimeout( "document.body.start()", sld.filters.revealtrans.duration * = 1000 ); else document.body.start(); } } function MakeSldVis() { var fTrans=3Dg_showAnimation && SldHasTrans() if( fTrans ) { if( g_bgSound ) { idx=3Dg_bgSound.indexOf(","); pptSound.src=3Dg_bgSound.substr( 0, idx ); pptSound.loop=3D -(parseInt(g_bgSound.substr(idx+1))); } SlideObj.filters.revealtrans.Apply() } SlideObj.style.visibility=3D"visible" if( fTrans ) SlideObj.filters.revealtrans.Play() } function MakeNotesVis() { if( !IsNts() ) return false SlideObj.style.display=3D"none" nObj =3D document.all.item("NotesObj") parent.SetHasNts(0) if( nObj ) { nObj.style.display=3D"" parent.SetHasNts(1) } return 1 } function ChkAutoAdv() { if(SldHasTrans()) SlideObj.onfilterchange=3DAutoAdv else AutoAdv() } function AutoAdv() { if(!IsWin("PPTSld") || !gUseSldTimings )return var sld=3DGetCurSld() if( (sld.mAdvDelay>0) && !parent.IsFramesMode() ) setTimeout("parent.GoToNextSld()",sld.mAdvDelay) } function GetObj(id) { if(g_supportsPPTHTML) return document.all(id); else return document.getElementById(id); } function SldHasTrans() { return SlideObj.style.getAttribute("filter")!=3D""= } function GetSldId() { return sId=3Dlocation.href.substring(location.href.la= stIndexOf('/')+1) } function HideMenu() { if( frames["PPTSld"] && PPTSld.document.all.item("ctx= tmenu") && PPTSld.ctxtmenu.style.display!=3D"none" ) { PPTSld.ctxtmenu.styl= e.display=3D'none'; return true } return false } function IsWin( name ) { return window.name =3D=3D name } function IsNts() { return IsWin("PPTNts") } function IsSldOrNts() { return( IsWin("PPTSld")||IsWin("PPTNts") ) } function SupportsPPTAnimation() { return( navigator.platform =3D=3D "Win32"= && navigator.appVersion.indexOf("Windows")>0 ) } function SupportsPPTHTML() { var appVer=3Dnavigator.appVersion, msie=3DappVer.indexOf("MSIE "), ver=3D0 if( msie >=3D 0 ) ver=3DparseFloat( appVer.substring( msie+5, appVer.indexOf(";",msie) ) ) else ver=3DparseInt(appVer) return( ver >=3D 4 && msie >=3D 0 ) } function _RSW() { if( !g_supportsPPTHTML || IsNts() || ( !g_scaleInFrame && (!IsWin("PPTSld") || !parent.IsFullScrMode()) ) ) return var padding=3D0; if( IsWin("PPTSld") && parent.IsFramesMode() ) padding=3D6 cltWidth=3Ddocument.body.clientWidth-padding cltHeight=3Ddocument.body.clientHeight-padding factor=3D(1.0*cltWidth)/g_origW if( cltHeight < g_origH*factor ) factor=3D(1.0*cltHeight)/g_origH newSize =3D g_origSz * factor if( newSize < 1 ) newSize=3D1 s=3DSlideObj.style s.fontSize=3DnewSize+"px" s.posWidth=3Dg_origW*factor s.posHeight=3Dg_origH*factor s.posLeft=3D(cltWidth-s.posWidth+padding)/2 s.posTop=3D(cltHeight-s.posHeight+padding)/2 if( g_scaleHyperlinks ) ScaleHyperlinks( factor ) } function _InitAnimations() { animRuntimeInstalled =3D ''+document.body.localTime !=3D 'undefined'; isFullScreen =3D (window.name =3D=3D "PPTSld") && !parent.IsFramesMode(); g_animUseRuntime =3D g_showAnimation && animRuntimeInstalled && !(isFullSc= reen && parent.IsSldVisited()); if( g_animUseRuntime ) { collSeq =3D document.all.tags("seq"); if( collSeq !=3D null ) { for(ii=3D0;ii numSlds ) gSldJumpIdx =3D numSlds; if ( gSldJumpIdx >=3D 0 ) { if ( gSldJumpIdx =3D=3D 0 ) gSldJumpIdx =3D 1; var jumpTo =3D parseInt(gSldJumpIdx); gSldJump =3D 0; gSldJumpIdx =3D ""; win.GoToSld( parent.GetSldList().mList[jumpTo-1].mSldHref ) } } } function _KDH() { if( event.keyCode =3D=3D 8 ) { event.returnValue =3D 0; parent.GoToPrevSld(); } } function DocumentOnClick() { if( IsNts() || parent.HideMenu() ) return; if( ( g_allowAdvOnClick && !parent.IsFramesMode() ) || (event && (event.keyCode=3D=3D32) ) ) parent.GoToNextSld(); } var g_supportsPPTHTML =3D SupportsPPTHTML(), g_scaleInFrame =3D 1, gId=3D""= , g_bgSound=3D"", g_scaleHyperlinks =3D false, g_allowAdvOnClick =3D 1, g_showInBrowser = =3D 0, gLoopCont =3D 0, gUseSldTimings =3D 1; var g_showAnimation =3D g_supportsPPTHTML && SupportsPPTAnimation() && ( (w= indow.name=3D=3D"PPTSld" && !parent.IsFramesMode()) || g_showInBrowser );va= r g_animManager =3D null; var g_animUseRuntime =3D false; var g_animItemsToHide, g_animInteractiveItems, g_animSlideTime; var g_animMainSequence =3D null; var ENDSHOW_MESG=3D"End of slide show, click to exit.", SCREEN_MODE=3D"Fram= es", gIsEndShow=3D0, NUM_VIS_SLDS=3D17, SCRIPT_HREF=3D"script.js", FULLSCR_= HREF=3D"fullscreen.htm"; var gCurSld =3D gPrevSld =3D 1, g_offset =3D 0, gNtsOpen =3D gHasNts =3D gO= tlTxtExp =3D 0, gHasNarration =3D 0, gOtlOpen =3D true window.gPPTHTML=3DSupportsPPTHTML() var gMainDoc=3Dnew Array(new hrefList("slide0002.htm",1,-1,1),new hrefList(= "slide0051.htm",1,-1,1),new hrefList("slide0052.htm",1,-1,1),new hrefList("= slide0053.htm",1,-1,1),new hrefList("slide0025.htm",1,-1,1),new hrefList("s= lide0015.htm",1,-1,1),new hrefList("slide0018.htm",1,-1,1),new hrefList("sl= ide0021.htm",1,-1,1),new hrefList("slide0027.htm",1,-1,1),new hrefList("sli= de0022.htm",1,-1,1),new hrefList("slide0055.htm",1,-1,1),new hrefList("slid= e0056.htm",1,-1,1),new hrefList("slide0057.htm",1,-1,1),new hrefList("slide= 0058.htm",1,-1,1),new hrefList("slide0059.htm",1,-1,1),new hrefList("slide0= 060.htm",1,-1,1),new hrefList("slide0039.htm",1,-1,1)); function FullScrInit() { g_allowAdvOnClick =3D GetCurSld().mAdvOnClk document.body.style.backgroundColor=3D"black" document.oncontextmenu=3Dparent._CM; document.onkeydown =3D _KDH; document.ondragstart=3DCancel document.onselectstart=3DCancel self.focus() } function Redirect( frmId ) { var str=3Ddocument.location.hash,idx=3Dstr.indexOf('#'), sId=3DGetSldId() if(idx>=3D0) str=3Dstr.substr(1); if( window.name !=3D frmId && ( sId !=3D str) ) { obj =3D GetObj("Main-File") window.location.href=3Dobj.href+"#"+sId return 1 } return 0 } var MHTMLPrefix =3D CalculateMHTMLPrefix(); function CalculateMHTMLPrefix() { if ( document.location.protocol =3D=3D 'mhtml:') { href=3Dnew String(document.location.href) Start=3Dhref.indexOf('!')+1 End=3Dhref.lastIndexOf('/')+1 if (End < Start) return href.substring(0, Start) else return href.substring(0, End) } return ''; } function GetTags(base,tag) { if(g_supportsPPTHTML) return base.all.tags(tag); else return base.getElementsByTagName(tag); } function UpdNtsPane(){ if(frames["PPTNts"]) PPTNts.location.replace( MHTMLP= refix+GetHrefObj( gCurSld ).mNtsHref ) } function UpdNavPane( sldIndex ){ if(gNavLoaded) PPTNav.UpdNav() } function UpdOtNavPane(){ if(gOtlNavLoaded) PPTOtlNav.UpdOtlNav() } function UpdOtlPane(){ if(gOtlLoaded) PPTOtl.UpdOtl() } function SetHasNts( fVal ) { if( gHasNts !=3D fVal ) { gHasNts=3DfVal UpdNavPane() } } function ToggleOtlText() { gOtlTxtExp=3D!gOtlTxtExp UpdOtlPane() } function ToggleOtlPane() { frmset=3Ddocument.all("PPTHorizAdjust") frm=3Ddocument.all("PPTOtl") if( gOtlOpen ) frmset.cols=3D"*,100%" else frmset.cols=3D"25%,*" gOtlOpen=3D!gOtlOpen frm.noResize=3D!frm.noResize UpdOtNavPane() } function ToggleNtsPane() { frmset=3Ddocument.all("PPTVertAdjust") frm=3Ddocument.all("PPTNts") if( gNtsOpen ) frmset.rows=3D"100%,*" else frmset.rows=3D"*,20%" gNtsOpen=3D!gNtsOpen UpdNtsPane() } function ClearMedia() { if (PPTSld.pptSound) PPTSld.pptSound.loop =3D 0; } function FullScreen() { if ( PPTSld.g_animUseRuntime ) PPTSld.document.body.pause(); ClearMedia(); var href =3D ( document.location.protocol =3D=3D 'mhtml:') ? FULLSCR_HREF = : FULLSCR_HREF+"#"+GetHrefObj(gCurSld).mSldHref; if(PPTNav.event.ctrlKey) { var w =3D (window.screen.availWidth * 1.0) / 2.0 var h =3D w * (PPTSld.g_origH * 1.0) / PPTSld.g_origW win =3D window.open( MHTMLPrefix+href,null,"toolbar=3D0,resizable=3D1,top= =3D0,left=3D0," + "width=3D"+ w + ",height=3D" + h ); if( PPTSld.g_animUseRuntime ) win.document.body.PPTSldFrameset=3Dwindow; } else { win =3D window.open( MHTMLPrefix+href,null,"fullscreen=3Dyes" ); if( PPTSld.g_animUseRuntime ) win.document.body.PPTSldFrameset=3Dwindow; } } function ToggleVNarration() { rObj=3DPPTSld.document.all("NSPlay") if( rObj && !PPTSld.g_animUseRuntime ) { if( (rObj.playState =3D=3D 1)||(rObj.playState =3D=3D 0) ) rObj.Play() else if( rObj.playState =3D=3D 2 ) rObj.Pause() else return; } else if( PPTSld.g_animUseRuntime ) { narObj =3D PPTSld.document.all("narrationID") if( narObj ) narObj.togglePause() } } function GetCurSldNum() { obj=3DGetHrefObj(gCurSld) if( obj.mOrigVis =3D=3D 1 ) return obj.mSldIdx else return gCurSld } function GetNumSlds() { if( GetHrefObj(gCurSld).mOrigVis =3D=3D 1 ) return GetSldList().mNumVisSlds; else return GetSldList().mList.length } function GetSldNum( href ) { for(ii=3D0; ii 1 ) PopSldList(); else if( !IsFramesMode() ) { if( gLoopCont ) GoToFirst() else EndShow() } } function GoToPrevSld() { ii=3DgCurSld-1 if( ii > 0 ) { obj=3DGetHrefObj(ii) while ( obj && ( obj.mVis =3D=3D 0 ) && ( ii>0 ) ) obj=3DGetHrefObj(--ii) if( ii =3D=3D 0 ) ii=3D1 GoToSldNum(ii) } } function GoToFirst(){ GoToSld( GetHrefObj(1).mSldHref ) } function GoToLast() { ii=3DGetSldList().mList.length if( ii !=3D gCurSld ) GoToSld( GetHrefObj(ii).mSldHref ) } function GoToSldNum( num ) { if( PPTSld.event ) PPTSld.event.cancelBubble=3Dtrue obj =3D GetHrefObj( num ) obj.mVis=3D1 gPrevSld=3DgCurSld gCurSld =3D num; PPTSld.location.replace(MHTMLPrefix+obj.mSldHref) if( IsFramesMode() ) { UpdNavPane(); UpdOtlPane(); UpdNtsPane() } } function GoToSld( href ) { if( PPTSld.event ) PPTSld.event.cancelBubble=3Dtrue GetHrefObj( GetSldNum(href) ).mVis=3D1 PPTSld.location.replace(MHTMLPrefix+href) } function SldUpdated( id ) { if( id =3D=3D GetHrefObj(gCurSld).mSldHref ) return gPrevSld=3DgCurSld gCurSld=3DGetSldNum(id) if( IsFramesMode() ) { UpdNavPane(); UpdOtlPane(); UpdNtsPane() } } function PrevSldViewed(){ GoToSld( GetHrefObj(gPrevSld).mSldHref ) } function HasPrevSld() { return ( gIsEndShow || ( gCurSld !=3D 1 && GetHrefO= bj( gCurSld-1 ).mVis =3D=3D 1 )||( GetCurSldNum() > 1 ) ) } function HasNextSld() { return (GetCurSldNum() !=3D GetNumSlds()) } function CloseWindow() { if( HideMenu() ) return; var event =3D PPTSld.event; if( !IsFramesMode() && event && (event.keyCode=3D=3D27 || event.keyCode=3D= =3D32 || event.type=3D=3D"click" ) ) window.close( self ); CatchNumKeys( self, event ); } function Unload() { gIsEndShow=3D0; } function SetupEndShow() { gIsEndShow=3D1; PPTSld.document.body.scroll=3D"no"; PPTSld.document.onkeypress=3DCloseWindow; PPTSld.document.onclick=3DCloseWindow; PPTSld.document.oncontextmenu=3D_CM; } function EndShow() { if( IsFramesMode() ) return if( PPTSld.event ) PPTSld.event.cancelBubble=3Dtrue doc=3DPPTSld.document doc.open() doc.writeln('


' + ENDSHOW_MESG + '

') doc.close() } function SetSldVisited(){ GetSldList().mList[gCurSld-1].mVisited=3Dtrue } function IsSldVisited(){ return GetSldList().mList[gCurSld-1].mVisited } function hrefList( sldHref, visible, advDelay, advClk ) { this.mSldHref=3D this.mNtsHref =3D sldHref this.mOrigVis=3D this.mVis =3D visible this.mVisited=3D false this.mAdvDelay=3D advDelay this.mAdvOnClk=3D advClk } function SldList(arr,curSld,fEnd) { this.mCurSld =3D curSld; this.mList =3D new Array(); var idx =3D 1; for(ii=3D0;ii 0) { PushSldList(sldList,fEnd); gCurSld =3D 1; } else if( PPTSld.event ) PPTSld.event.cancelBubble=3Dtrue } function PushSldList(arr,fEnd) { var ii =3D gSldStack.length; gSldStack[ii] =3D new SldList(arr,gCurSld,fEnd); GoToSld( gSldStack[ii].mList[0].mSldHref ); } function PopSldList() { if (gSldStack[gSldStack.length-1].fEndShow) EndShow() else { gCurSld =3D gSldStack[gSldStack.length-1].mCurSld; gSldStack[gSldStack.length-1] =3D null; gSldStack.length--; var sldList =3D gSldStack[gSldStack.length-1]; GoToSld( sldList.mList[gCurSld - 1].mSldHref ); } } var custShowList=3Dnew Array(); function ImgBtn( oId,bId,w,action ) { var t=3Dthis t.Perform =3D _IBP t.SetActive =3D _IBSetA t.SetInactive=3D _IBSetI t.SetPressed =3D _IBSetP t.SetDisabled=3D _IBSetD t.Enabled =3D _IBSetE t.ChangeIcon =3D null t.UserAction =3D action t.ChgState =3D _IBUI t.mObjId =3D oId t.mBorderId=3D bId t.mWidth =3D w t.mIsOn =3D t.mCurState =3D 0 } function _IBSetA() { if( this.mIsOn ) { obj=3Dthis.ChgState( gHiliteClr,gShadowClr,2 ) obj.style.posTop=3D0 } } function _IBSetI() { if( this.mIsOn ) { obj=3Dthis.ChgState( gFaceClr,gFaceClr,1 ) obj.style.posTop=3D0 } } function _IBSetP() { if( this.mIsOn ) { obj=3Dthis.ChgState( gShadowClr,gHiliteClr,2 ) obj.style.posLeft+=3D1; obj.style.posTop+=3D1 } } function _IBSetD() { obj=3Dthis.ChgState( gFaceClr,gFaceClr,0 ) obj.style.posTop=3D0 } function _IBSetE( state ) { var t=3Dthis GetObj( t.mBorderId ).style.visibility=3D"visible" if( state !=3D t.mIsOn ) { t.mIsOn=3Dstate if( state ) t.SetInactive() else t.SetDisabled() } } function _IBP() { var t=3Dthis if( t.mIsOn ) { if( t.UserAction !=3D null ) t.UserAction() if( t.ChangeIcon ) { obj=3DGetObj(t.mObjId) if( t.ChangeIcon() ) obj.style.posLeft=3Dobj.style.posLeft+(t.mCurState-4)*t.mWidth else obj.style.posLeft=3Dobj.style.posLeft+(t.mCurState-0)*t.mWidth } t.SetActive() } } function _IBUI( clr1,clr2,nextState ) { var t=3Dthis SetBorder( GetObj( t.mBorderId ),clr1,clr2 ) obj=3DGetObj( t.mObjId ) obj.style.posLeft=3Dobj.style.posLeft+(t.mCurState-nextState)*t.mWidth-obj= .style.posTop t.mCurState=3DnextState return obj } function TxtBtn( oId,oeId,action,chkState ) { var t=3Dthis t.Perform =3D _TBP t.SetActive =3D _TBSetA t.SetInactive=3D _TBSetI t.SetPressed =3D _TBSetP t.SetDisabled=3D _TBSetD t.SetEnabled =3D _TBSetE t.GetState =3D chkState t.UserAction =3D action t.ChgState =3D _TBUI t.mObjId =3D oId t.m_elementsId=3D oeId t.mIsOn =3D 1 } function _TBSetA() { var t=3Dthis if( t.mIsOn && !t.GetState() ) t.ChgState( gHiliteClr,gShadowClr,0,0 ) } function _TBSetI() { var t=3Dthis if( t.mIsOn && !t.GetState() ) t.ChgState( gFaceClr,gFaceClr,0,0 ) } function _TBSetP() { if( this.mIsOn ) this.ChgState( gShadowClr,gHiliteClr,1,1 ) } function _TBSetD() { this.ChgState( gFaceClr,gFaceClr,0,0 ) this.mIsOn =3D 0 } function _TBSetE() { var t=3Dthis if( !t.GetState() ) t.ChgState( gFaceClr,gFaceClr,0,0 ) else t.ChgState( gShadowClr,gHiliteClr,1,1 ) t.mIsOn =3D 1 } function _TBP() { var t=3Dthis if( t.mIsOn ) { if( t.UserAction !=3D null ) t.UserAction() if( !t.GetState ) return if( t.GetState() ) t.SetPressed() else t.SetActive() } } function _TBUI( clr1,clr2,lOffset,tOffset ) { SetBorder( GetObj( this.mObjId ),clr1,clr2 ) Offset( GetObj( this.m_elementsId ),lOffset,tOffset ) } function Offset( obj, top, left ){ obj.style.top=3Dtop; obj.style.left=3Dle= ft } function SetBorder( obj, upperLeft, lowerRight ) { s=3Dobj.style; s.borderStyle =3D "solid" s.borderWidth =3D 1 s.borderLeftColor =3D s.borderTopColor =3D upperLeft s.borderBottomColor=3D s.borderRightColor =3D lowerRight } function GetBtnObj(){ return gBtnArr[window.event.srcElement.id] } function BtnOnOver(){ b=3DGetBtnObj(); if( b !=3D null ) b.SetActive() } function BtnOnDown(){ b=3DGetBtnObj(); if( b !=3D null ) b.SetPressed() } function BtnOnOut(){ b=3DGetBtnObj(); if( b !=3D null ) b.SetInactive() } function BtnOnUp() { b=3DGetBtnObj() if( b !=3D null ) b.Perform() else Upd() } function GetNtsState(){ return parent.gNtsOpen } function GetOtlState(){ return parent.gOtlOpen } function GetOtlTxtState(){ return parent.gOtlTxtExp } function NtsBtnSetFlag( fVal ) { s=3Ddocument.all.item( this.m_flagId ).style s.display=3D"none" if( fVal ) s.display=3D"" else s.display=3D"none" } function _BSetA_Border(){ b =3D gBtnArr[this.mObjId]; if( b !=3D null ) b.S= etActive() } function _BSetI_Border(){ b =3D gBtnArr[this.mObjId]; if( b !=3D null ) b.S= etInactive() } var gHiliteClr=3D"THREEDHIGHLIGHT",gShadowClr=3D"THREEDSHADOW",gFaceClr=3D"= THREEDFACE" var gBtnArr =3D new Array() gBtnArr["nb_otl"] =3D new TxtBtn( "nb_otl","nb_otlElem",parent.ToggleOtlPan= e,GetOtlState ) gBtnArr["nb_otlElem"] =3D new TxtBtn( "nb_otl","nb_otlElem",parent.ToggleOt= lPane,GetOtlState ) gBtnArr["nb_nts"] =3D new TxtBtn( "nb_nts","nb_ntsElem",parent.ToggleNtsPan= e,GetNtsState ) gBtnArr["nb_prev"]=3D new ImgBtn( "nb_prev","nb_prevBorder",30,parent.GoToP= revSld ) gBtnArr["nb_next"]=3D new ImgBtn( "nb_next","nb_nextBorder",30,parent.GoToN= extSld ) gBtnArr["nb_sldshw"]=3D new ImgBtn( "nb_sldshw","nb_sldshwBorder",18,parent= .FullScreen ) gBtnArr["nb_sldshwBorder"] =3D new TxtBtn( "nb_sldshw","nb_sldshwBorder",pa= rent.FullScreen,null ) gBtnArr["nb_sldshwBorder"].SetActive =3D _BSetA_Border; gBtnArr["nb_sldshwBorder"].SetInactive =3D _BSetI_Border; gBtnArr["nb_sldshwText"] =3D new TxtBtn( "nb_sldshw","nb_sldshwText",parent= .FullScreen,null ) gBtnArr["nb_sldshwText"].SetActive =3D _BSetA_Border; gBtnArr["nb_sldshwText"].SetInactive =3D _BSetI_Border; gBtnArr["nb_voice"] =3D new ImgBtn( "nb_voice","nb_voiceBorder",18,parent.= ToggleVNarration ) gBtnArr["nb_otlTxt"]=3D new ImgBtn( "nb_otlTxt","nb_otlTxtBorder",23,parent= .ToggleOtlText ) gBtnArr["nb_nts"].m_flagId=3D "notes_flag" gBtnArr["nb_nts"].SetFlag =3D NtsBtnSetFlag gBtnArr["nb_otlTxt"].ChangeIcon=3D GetOtlTxtState var sNext=3D"Next",sPrev=3D"Previous",sEnd=3D"End Show",sFont=3D"Arial",sAr= row=3D"Arrow",sFreeform=3D"Freeform",sRect=3D"Rectangle",sOval=3D"Oval" function ShowMenu() { BuildMenu(); var doc=3DPPTSld.document.body,x=3DPPTSld.event.clientX+doc.scrollLeft,y= =3DPPTSld.event.clientY+doc.scrollTop m =3D PPTSld.document.all.item("ctxtmenu") m.style.pixelLeft=3Dx if( (x+m.scrollWidth > doc.clientWidth)&&(x-m.scrollWidth > 0) ) m.style.pixelLeft=3Dx-m.scrollWidth m.style.pixelTop=3Dy if( (y+m.scrollHeight > doc.clientHeight)&&(y-m.scrollHeight > 0) ) m.style.pixelTop=3Dy-m.scrollHeight m.style.display=3D"" } function _CM() { if( !parent.IsFullScrMode() ) return; if(!PPTSld.event.ctrlKey) { ShowMenu() return false } else HideMenu() } function BuildMenu() { if( PPTSld.document.all.item("ctxtmenu") ) return var mObj=3DCreateItem( PPTSld.document.body ) mObj.id=3D"ctxtmenu" mObj.style.visibility=3D"hidden" var s=3DmObj.style s.position=3D"absolute" s.cursor=3D"default" s.width=3D"120px" SetCMBorder(mObj,"menu","black") var iObj=3DCreateItem( mObj ) SetCMBorder( iObj, "threedhighlight","threedshadow" ) iObj.style.padding=3D2 CreateMenuItem( iObj,sNext,M_GoNextSld,M_True ) CreateMenuItem( iObj,sPrev,M_GoPrevSld,M_HasPrevSld ) CreateSeparator( iObj ) CreateMenuItem( iObj,sEnd,M_End,M_True ) mObj.style.visibility=3D"visible" } function Cancel() { window.event.cancelBubble=3Dtrue; window.event.returnVa= lue=3Dfalse } function Highlight() { ChangeClr("activecaption","threedhighlight") } function Deselect() { ChangeClr("threedface","menutext") } function Perform() { e=3DPPTSld.event.srcElement if( e.type=3D=3D"menuitem" && e.IsActive() ) e.Action() else PPTSld.event.cancelBubble=3Dtrue } function ChangeClr( bg,clr ) { e=3DPPTSld.event.srcElement if( e.type=3D=3D"menuitem" && e.IsActive() ) { e.style.backgroundColor=3Dbg e.style.color=3Dclr } } function M_HasPrevSld() { return( parent.HasPrevSld() ) } function M_GoNextSld() { if( gIsEndShow ) M_End(); else GoToNextSld() } function M_GoPrevSld() { if( gIsEndShow ) { history.back(); PPTSld.event.ca= ncelBubble=3Dtrue; } else GoToPrevSld() } function M_True() { return true } function M_End() { window.close( self ) } function CreateMenuItem( node,text,action,eval ) { var e=3DCreateItem( node ) e.type=3D"menuitem" e.Action=3Daction e.IsActive=3Deval e.innerHTML=3Dtext if( !e.IsActive() ) e.style.color=3D"threedshadow" e.onclick=3DPerform e.onmouseover=3DHighlight e.onmouseout=3DDeselect s=3De.style; s.fontFamily=3DsFont s.fontSize=3D"9pt" s.paddingLeft=3D2 } function CreateSeparator( node ) { var sObj=3DCreateItem( node ) SetCMBorder(sObj,"menu","menu") var s=3DsObj.style s.borderTopColor=3D"threedshadow" s.borderBottomColor=3D"threedhighlight" s.height=3D1 s.fontSize=3D"0px" } function CreateItem( node ) { var elem=3DPPTSld.document.createElement("DIV") node.insertBefore( elem ) return elem } function SetCMBorder( o,ltClr,rbClr ) { var s=3Do.style s.backgroundColor=3D"menu" s.borderStyle=3D"solid" s.borderWidth=3D1 s.borderColor=3DltClr+" "+rbClr+" "+rbClr+" "+ltClr } ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/fullscreen.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/buttons.gif Content-Transfer-Encoding: base64 Content-Type: image/gif R0lGODlhWAESAPf4AAAAAIAAAACAAICAAAAAgIAAgACAgICAgAQEBISEBASEBISEhAQEhMTExAQE /KTM9Pz8/ERERPz8BAT8/KSkpGRkhMTcxCRkxAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAMDAwP8AAAD/AP//AAAA//8A/wD//////yH5BAEAAPgALAAAAABYARIA QAj/APEJHEiwoMGDCBMqXMiwocOHECNKnEixosWLGCEiQKCRo8YDIEHig2DxAD4LKD1avDDhgUuV GUOGbHBR5QGYFBG4fHAB58EKECJEwCfUYAWPFGAygLCRKQIIUKM6WAohKUGgCikcFWh1INWoYKVS 7SrQ41GbIhGYLAuVQVOoC+IyXUuUAkG1HE3KFDhAwkgJc5/qDZmxsOHDiBMrXsy4sePDGyP7lChT 5MULmDPnlDw5YmW6FTNrfgi0aNGEZPEtbfo2bNTUpYeeRph69dPWrqvCNFuBIGG1A59CYPAVQtwF cw0OxWdTIOiBCQAjaAB17fOFII8voGmRc+eOkilm/z9OvTtnjN4hYk1YoT3H2rnBOphPAB9skgq3 2lcaX/5YnJVt9BloAar12UEIRGCZdQghMABI1Tn32IQUVthYexhmqGGGMBHAgIfEhSiiiATAtOGJ GnYI4ogsEldicAYOGNJGZaml3Y0L3NTZUBz1xlB0PUoUI0jlVWRSVN89xBKSGB1YU1TMoRfVBEle hR9HQs1GEHxQ3RZZbveZthxt/LEmXHyp5XWABSCpaRJwbM13gAACPHVccgJFgF+NEhqUAEkQSCfc ddhVhh5dVTaEAEsTUNlkZUWKN9BNGDE6QU8OrSdbllzpxyVTTrU21Uj3scfhp7nNpyoDaTIEJ3P9 If8J2kbLNYegg9JRBwGhFvbq66+ZUiDssMQWK2yHxiZLrInKNotss8l+B5yBrr5p2UIJvortANoC 6+234IbLla/elRtZjeamh5CC3Sb0YLvixivvYkDFGlZqBNh7r4n6goVvv6/ZlCicN+kL71ALTnpQ dCIZp7CihkLA3UQmoWRBogyxtBPGCTnZncVR1rSTo5nuydC/Eo8kUH+lOoTyyvjoCuZu+PjYIKLD rZybtrM1d10C0QXKEVz48GrQkEeWdFJKlbb0UmEeV0SzyC6RvNB646LWIVjUNeC1zAFb+RDKXksM 9r531bwmpQW9+hQDDsx5pqwDRWBXcFES2hc+gEn/wJGdEEWdE6LoWcoxQweaTNmkhydkKaYMYY2Q p1unbLl/pPKbn3v7EZRvrKvqlja2ONvrtoIw9tn2g32/tQBlhtYkU+MNikZ7QYlPDPuMldpOmuJX nVo5VF3HjI+q89mteX5IDd/fVFClma5kiE4f2XUJhlz09gj2NbTDRifkmu7zXjR++RVJ/lPznuvr QH2iD6S+UewP9Dnon6d2NLwN8n90BB6hC68cFMCHoe+ACASP9c4DowUyEF0OPFcDI0ij/VXPaAKC UfhS95AKGoQ7IBRICI1HwhGaUIQoLGEKT6jCFrLwhSuMoQtlCMMZ2rCGOKShDm+4wxzy8Ic+DGIP B4cIRCKGMCAAOy== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/frame.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii" Bias and Variance in Value Function Estimates Shie Mannor (MIT) Du= ncan Simester (MIT) Peng Sun (Duke) John Tsitsiklis (MIT) ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/outline.htm Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="us-ascii"
Slide Show
Outline
= = = = = = =
------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/oledata.mso Content-Transfer-Encoding: base64 Content-Type: application/x-mso 0M8R4KGxGuEAAAAAAAAAAAAAAAAAAAAAPgADAP7/CQAGAAAAAAAAAAAAAAABAAAAAQAAAAAAAAAA EAAAAgAAAAEAAAD+////AAAAAAAAAAD///////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////9 ////DwAAAP7///8EAAAABQAAAAYAAAAHAAAACAAAAAkAAAAKAAAACwAAAAwAAAANAAAADgAAABAA AAD+////EQAAABIAAAD+//////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////1IA bwBvAHQAIABFAG4AdAByAHkAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAWAAUA//////////8CAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABCEuthZhsQB AwAAAAAeAAAAAAAAOAA5AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAYAAgH///////////////8AAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAsgcAAAAAAAA5ADIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABgACAQEAAAADAAAA/////wAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAB8AAAAdBwAAAAAAADEAMAA0AAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAIAAIB/////wQA AAD/////AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAPAAAAKgHAAAAAAAAAQAA AAIAAAADAAAABAAAAAUAAAAGAAAABwAAAAgAAAAJAAAACgAAAAsAAAAMAAAADQAAAA4AAAAPAAAA EAAAABEAAAASAAAAEwAAABQAAAAVAAAAFgAAABcAAAAYAAAAGQAAABoAAAAbAAAAHAAAAB0AAAAe AAAA/v///yAAAAAhAAAAIgAAACMAAAAkAAAAJQAAACYAAAAnAAAAKAAAACkAAAAqAAAAKwAAACwA AAAtAAAALgAAAC8AAAAwAAAAMQAAADIAAAAzAAAANAAAADUAAAA2AAAANwAAADgAAAA5AAAAOgAA ADsAAAD+////PQAAAD4AAAA/AAAAQAAAAEEAAABCAAAAQwAAAEQAAABFAAAARgAAAEcAAABIAAAA SQAAAEoAAABLAAAATAAAAE0AAABOAAAATwAAAFAAAABRAAAAUgAAAFMAAABUAAAAVQAAAFYAAABX AAAAWAAAAFkAAABaAAAA/v///1wAAABdAAAAXgAAAF8AAABgAAAAYQAAAGIAAABjAAAAZAAAAGUA AABmAAAAZwAAAGgAAABpAAAAagAAAGsAAABsAAAAbQAAAG4AAABvAAAAcAAAAHEAAAByAAAAcwAA AHQAAAB1AAAAdgAAAHcAAAD+//////////////////////////////////////////////8ALgAA eJztWV1sVEUUPnPvXdq9u6XtttuW0p+7pRRKS2lBRDFoUUsg0iI/L4qJLLrUNWy3P4tQTEhpCRrx B38CZJ+qD/qA0Qi8iJqU6IuS8BNe1BiBoDEYQYwkaJXWMzP3Lncuu2W2tALa05zce2bOnTlz5nxn zmxPHM898/ZHxWfBQfeDCkPDbphkayPIiiXkAOhm29Dw8DBtUpGHJ+iOolUQxb8YGNAEbfjshG5n KIxIBeBK7Dndf+UYj5AB3r3Ertu+7711c1Z8QWhMNU62YmoFbIRQWnPaSQeF0LkVM/ZkvvEj54M1 /0O4/gi0ox3r4dm05/chIuz+lP1ug/lUzPU/ip4PQRfUsz95KsL1UwxqIL9+6quYzt/pd1dt9g+Z 7AOO62xIjWlMAVCInIGcmUQvd4RvJ+j2oH8T//UXVj1N8e+hwgT+Gd0q/D9h4t+y2w1iHrCTlROo z7KAb50frsc21fFie16Svgm6PakJOmATBBH5YcRBG+aBFlN6TgqVxijin8bHaeWGatKU7vxjTXfy /Fb9TnM3zSEuZFqfWWc6zQs0VdCcTbFt4Z/WBfT8p2c8rRVoHUDzQ4E53lSguQlgCnIxcgC5BLkU uQy5HGjsANyNXIE8DbkSeTpyFfIM5JnI1cizkGuQa5FnI9chz0GmebIBeS7yPOS7kOeb8y/A5z3m +wSlJoIeUnUeQ86zm+5pc/ipzmhXdEPMaOrYFIyFo23GvLp6FgoPr060sfCwhLp5cPneAx2OQNN9 fatAxbDx5BJlx4BqC1wr9ojZZKWGt3Djv8SAnKvTPoxEdYqLnl4lrJdGqlux3vgpSN+IgkOwtzpS ZLZ5EnpepdHVX0UjtmpSNkYmnY9H8gC+fYa8DsPvqsYjnup4oTkYe2ZNdzumQ4PQuf9SXEPnmA27 2ZVYUyBndXdkfXQjbGukMVmmK15F3+ZHNk54t9FIpeubTTim5ur7SDOzd2a3zHhFhfr3yccj7HnJ Nu5ejQ5RLTUu2qkq28OgbI8gh0DZ6atQXmygoNaV3mr2ZZI+7NwzE5Q9TMHN5u0n1UV0FyPBLaFM s+WDfDqpEbQ0enzU1s5gLKFR5eMaHlM+5qU9y0PBzrZwW6ul1eKmWrWJrxpcVF6YkE8qvJ+v2DAz YPIVt07Tt6baGe5JYnoSlEML6BCzQ5b88nwq15jyZfhhnti/tdbeD8qmUi5zu3qGRrIrUKlfkNvh ftJcb5+nnxycLcrZs0T55AxRHgjY7V4TjoS6jJbQZmNVNBJsAznvxLVT03mcuU15bz7VMJAsjVAu 2xdDN+UezyRTwzC8ZttAhp5os7eDlry9EcT2TJYLFFhy5HMzaVxyFUDJIwXsXSVE1XWNEOLrW4kS 5gZfXxcmIQIefIuZgV4Lvr6o2GoLcaV3IQb/IkgxBLH9JDYqkjFhZw1icDnH4fYgcidyG1DsMr+Q 8+Xc1+ZukMctNFq7QRQ/yzdBS7bQeavqv4MT9d9tMf9/uf6bqP1uTLdz/ZdZxeu/4kT9lzUm9V9P xUj1X2PFzdZ/pFK/Mpb13/kZ+svjUf+hnX+M9vw/GBDP/805zvO/KUs8/wcyXKaGAemf86nOOev8 32sGTf+DBXBo6f/n/Jc95y2POavwuKMKjyetwuOOKjzuqMLjI1XhQONOLs6c8TyoVbM4awi5TTlQ TjViOBrX4H5oRTZMv1D/xExfYVsvOnHHQdXKPKzr2pUH3axifUx8vRHw9QaRtwDbgyDbmRSbzjfE MVJiI/HTaxckwg0bZ8bL2c2EOFuOOKSvr9M5CtVCT4o9F9XH2D2nIcx36KK6cqq4Q4Pa7367xiCZ X2ZpmN9Ae4HVInOLw7z1t1yeGtQOG+y2E7bkN8rYbciUL6o/l4j9PX57/yB5l9/iwjK3OMzTr8vF dVxbPcU+T1z7uECU/XmivD9blONuu91xcsgj9JNdDJ+Lwk5/Egce79S8sGw6zwsT+L+1ucLKy4un iai/qHYHxLzQUe7MC3nFY5sXGsr0DNm88G2liPsDFWJe0ANif/+U0eeFokL9nGxe6CoVcf5NsSjP LxTlo448cTgrvbywlsYBRBP9R1mG6ErIdawqX3ndOkly/7tl68fnWR5uTdi9qEiQyVY/l1P5t3mN 0bQl1hkE7t8fZf17fCodYnmY1pe0uteB34Io8bly2N0X4FNW1xFFqVjd3RULRaik4z32Ja6M9P4V H0wecm9zrlGF5NSyuGUZG9Omkc5Zy1qt32fp+65fKyzNFPkt8bMtzWXXjYXKcvu1j3yVIX9fwf04 LXtfyffI31cwvjRZexeVpGXvGVl7VwbSstclaa92Njste89K2qvV5NntFXDEzt/R4BzrryFZnK8t FnFemj96nGN99aYszj8sHDuc//LT6HCeql3A/zhg+gb/0xl3/OM+vTIe+Me4+3M88I/2vjoe+Ed7 B2Xxf8STlr2vyeLfkz0y/tP73WdQu4+d27FEXVTtF+WgT5RfmCzKp7yCTFaU2uUeJcLqrGCif7cu 6n+XKcp5mig/qXA5SR67ifsG1lnMzs5r97V8Uf4tV5SPZwl1lfaAV6irtHd0nm+tPNht2NcdJ5+U iHVY2VRR/2u3WLctzRT1908S9VPVef8AC5P84AAAAAAAAAAAAAAAAAAAACIAAHic7RhtbFRFcHbf 9evdAeX6QYECrwWxkFKOAhoQCAjFkvDRlEpiVMKVHnjQXttry3GgDbVIGvFHf/EVMGdC/UiMgBKI CYlFDZIQkR8khpho8YvEmEgMTUSldWZ33/HuuLbvgIg1nWb6dmb3zczOzszOuytfje1569SE6xAH S0CDvv4MSLXwGCI3iUwAXfH6+vv7iaUh9o/AsIJKqMe/ZjCgDAL4DEI4PhQGhVxIiZ45nT+/LCOk W06vtK7dtPGPxvnrPmcUU3tSzJhaB7XgS0qnFXTgjHRzFXt23slWKPUvx/3XQQPaUQ3bktbvxoyw +tPue1vUk6v9V6DnfdAEHvFnH/Jw/5SDDrC/fzqneQ45pscdi/19CtNB5nUajOT0/xn+zfy/slB/ +QnMfyeOl6ZL3kj+P5r8P6nqr2l3BsTWASuYNWEMrhmN6AJZG0bBSG0Y7lAGjdACXsx8P+ZBAOvA WkXtsJWVxn3EP+XqK3YW2oRk9T9sGM76zf7dDbKGUFmg/ozufSrRVBeoz6eaTXlPOU81gGoBfgLA WMQcJcPELBipC8MFGJ6cpsvzir+7x+G/Nf7Nwfqm+i3NRllji7fZXx8w5pZ4RCisWB/lifAwiZK5 cGvBh40xYeZk7nY/uNvxgmV8bzfeQHpp+3OQmpEPVA8qP/WqhT07cqDmtrydNcY0Xcc3NwDftxhJ fNndvgoJdyHfe0pzt1cIuzUVu0zFnfmNuhyt/BGJ3zWaw0jWbjPSli9mKdIzuDmStyiNGEcRYlTC 8hTPGV3n4hFuOClbpqeOgQnqzqVM6MbRccQe9OYKLjOG1rhgjbf5papwA5bTCqH7L374zg/Chk7x Se3gkLk+XFddXwuhpci45tSLn796K1SDyMQ6j9jfLCZzslTvZ2c5zRSF7cjrzNYPJ5bHlLw0NjWd ZmYpeXv6YHB5RweSR8+bUTuBZzukXJPuFH5eHNaVdzQsJM6xKiZA5weKgB+YAVE+HXXHHJpyt1fG rYY430jdDO7OxcLaZWtXxfOY5SeN+wEdZBYRSN9litoJcI7knmKcF64PNzX76uRqaWVUd4zP7Ngd L68AXmfm3JUbbvj4/BehZPwRe1797JM0EjdDxYGh7hNxaonidNZAcWrqlnKrWLFOKxrCJn0jVdLx 8ZZID8bbEbvx5kihFd5EcoHkVvnrfE3GWl/IqKyv8wbAnv3AF2TQispoHN8Seiqi9EyqTrAqSueK 091wjx8hKTvi95fGjok8nZNIbqLzmT3U/h5V/7ePDbnMNiSr/2HDcNb/sPq/XMSJShb1f9Q70O05 HnGC4lPMTkKcjDgFKHYkfwQeHfw3+r+V57epetCxLBdOl+eKser/tMcZvtRWB7zNi7gT0Q+S9g/a /31WALALgzmQYvZ/Z/mD9389KZHJA/V/XYibMLxfS0nc/0WYrNu9fw/Qr7UuRcZ7M3Xu4nprDqJx xdV6b/93lE/LtvZ/g8sLT9XfTiyPoQ83I9YitiA2ITYj+hCDgsaTg1F839hCasC2mj34cuzEGA22 Uz+GbdurxRDl8v1ZBRZu2d21fN/VAnPW0usNBXkfdEHm4SwxPh4543EckuNLK26HTjh+FjeiEdCF ZyK8MpXOySDwuQRvScoiRnWswRts9otwTVf870XMGIZJzxO0V9EXtDfErV4cpT9iknYqXRuFpDrv Tn+df5fP7l0/0FnE3vWzHZrQXuo16ZlM0mbvPETvOWiv+Cv2iqP7MlqLhfe2h80dponviuVRugQk neC7IIm9xfdTEf5+nvBkmLftEvEGTgdjTDeDyMl4RyndLjGhFcvl+38rFL2tN+hrknJPOFa7xIkG TfoFp6QzFL0wQ7wRqFFvPGj/b+7nxVy5H6fSc2Y0zWyubUH3B6PWVY8W1tSa1kRcTqbLsoaNVFuD KnFBlYZ+y7MeMQAq3cTqWNQmYZKpNBMsr3iBt9VIGtPuLj+octxI+ttpdRzfvEcqcFwEsupR9Xap +cfUGHuGPUVqTA6vRGxQzv+uP7Zv6TSjdCkULoMg9qJeqE3iTOKB9mao8TPTCsqt9aPzYj58ub83 VJ6WA0929oa+nq4fPIQ8wkubboRkdwPwLY5rdkC6KTM7qLOaa8xTdLnK4+pmHuNYFkQQPYe74Nwk XdT50wdJzxkPvWRiJs4df7dLyLiK809VnxE/99NcxZEu8LzZBSyBzSD8RDM5UH5Q2r8I93Gz489Q R35viM6A7B/IZvMcyN/m+eQBxNWco3zKZOv3372/A6ypMsp2NgfFzzV4T2n28n23tmwSiagNm3TP OCt9wnHSaaUvaBdF7Vs91O8R5v32jr2ault7diKJ+EbJ7b4zmFzcn8Pe/m7zX0Td+cn2d6dde+dP sNbnCC8fT3RZlL6ea63Xu7WmnNj1gWzrfIQ/nUX01qjf/wECKzShAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAKgAAeJztWGtsVFUQnnvO3d32trZl+1gKLd3dVh4GgdT4rBpMaQNY HpFKYiDaapZabHfbsiIrsTag/PCBxijGoEYTYyBGRIL+wR8QjSYGgfDQRP1RYyIkkuAD8YWtM+ec e3vPsi13qZUae8hwZ+Y85jxmvpntkcOT+l9/d8o3kNZuBQ4Dg7ngd+kMJGYLRQABpRsYHBwkFUca nGj/qXYHJPBfEsLQAHH89kAq3RVGbGXgc97cJP84JD1kv+xudI/tuq68dOGyj4xC5Pss26eWQQfE srLpbhYwg2wz5Xte5sxAKgHbfj2evxO6cB/3wtqs7QcxItz36WVOGGmN4hugGx6EVrz5dtxHHPuW Kmm9p1sJX4L9KNJbhuR9SH+55g8ougKGj+UB9UUIgEniDkDgxOg8caJdjvZvxv/8tZ1tFP95xE/E v2iXK/4fUvFvz8sBHQeo2VhAT0XffPwWqDnFcGG852bQTbTx3eyajmo3il/KB4TlVNuRT9Cb0vtT zNL7X6F8gHK4G//JHyimSoEwASCENBmpHGkK0lSkCqRKpGlIVcqnIupbjd8apCuRpiPNVPpZ6jvR xqYxhb/LEfljsA7miX/e22TEXwOGan+v884FsjBykeY+zz+36v/DvoHRxS25Rnruphhe0n5fT2Jd Yk0y3ND9YGuyPREPXzNnnoCCBSscnYAHW5hzDZy9cU+3bucHgwk8qbXmm1XCY9YnCQ+IGDQeeEbl o+U3lUH+wjLBc8PglmUEN7cBLzQgD7n1hDgWfTlOQEW96ELmAcQo7GKbZoOjZU8WR1zahqGxbMvx iN2L3dtmAtuGUMO23AJkLrg5CbyADGzqhOCmdqQ4cLwdg5uZzNISgdNsW85pIFP6ILcdNQitbUR6 BLFzUwdSAqlNHUlYbEXa4MWi4fpJPtRqsSxaNYuuNBnLUfLb03X59xoDhn7Tj1f8n8D+sW2XC/+P TOD/uLA/XvB/1cXxH5mVEuTHFPvVuOYxNfWFBHuRY9jjz3Nl5uJTnXygho4W/zOZspMUpSlllhfS uS2DPXaAy44Mexl5LWqPpCi/lLA47KokrpjN45Ij3bEqWyc50j0dtnWSI93WqK3bGiX/+ZO9c/5b sf6zIp2ZDIpWpDrvTXRA73xU3FltPZHPrN5SpPCR/F5AhKNcdzX6pCl801C+OZstqKQlNqa8rNtQ Y8Uyr2uodZmz7s6o4Tr/bh4NyTP0+CVHumsrbJ3kSlgf314odX38w2q5p74BSNvTkuZww4ZkTyvI sz7p7axN/mVTDRH3du6nBWVOkO1NTMIRHLw1QH1YCfBPOCF9heilfeUym5NZgDiDEQwRN8eYrHR5 zrh81hLor6GKYbq/ECsDiVlUSexHbg/Sa1gGBAOy4qAx+bCkNXl/c6orBtBnyDv4ZcR3eSliHc18 B6DOvtpczGjqzJSX9U7UWItGfuehdRvyaYlZntbFfR4bbp/6W602raKs9rvY6359U7La73Fv+/2e NVZmtd/bve33e/ZoVO6X5D+Y9BxQ/kv7kHEzzzynYukn/7kKW5cbtXWSK8GQqFXjwDweTY8vI/Md nPAWX0ljTzXttSc1XNym3UGTNywZ8Jli3V9TUn7f1OVGrstn/VL+TcnLmZTPCTn7XywwbC66CP5n zLjpv/9u+K4Ujq+e+P3n9fffSPh9bAZANyLpXge/v7wAv8mfOqbSzJOOP+lyI9fls34pn3L8Scrf Kfmg//oqkr9S8qf8PU0+w5oqSF6g5BazS2DQXiXX+epC7vEtZr/A1Fuc/o25st9LPsR4/dxbvB70 V0doiQ5n369rcp1vRrkuf1ysy215uuz3SzkDrgDts7m9M7YuvDT2UPiORGdrHLzF/0F/aBqt8ICy 86PvKoG1Dc6+X9D6z7AdWn+db2rZUP9o/cfmvOf//f7wiPm/vxox0585/3d5yv+7otb2kfP/13xJ Fvkffeigt/z0NV+Y7yU/xeHRoF3DSo50b5TYOsmR7u5SWyc50t0TsnX3hLzUqXgfL3utf0uD3utf vJfPvNa/D4fc9e+/W//j+V8Zi/ofz39ovNX/eNZXs6n/h/DK4jsiupwqc8v7zLV5upw7DL4Nc1eH vd1Vk/++cjf+38yqtHqmlldp9cxuo8pVz9D4U+XufFXLpXzKGS9lO1/182+1/LLP7B0mv2SH2xfe d/1kNy53+1aG3Lg8l+0Mk1zv3P9hDcfnsk8r3f1z2cOl+vjXivXx24Pu/n4eKSAYbkva57w7R4xP 2v3dAbEfR7Z8Yr4z/mWuj3+Rufv7+Qvq7zq23CQySltS3mN48NLuMd0/9L8fpfndKN5nLjsj6pTZ yVwl75hCmbGzdUPM1pwPkaYj0Raz5+wvEXMS9oj3CsWc9njMvoUTIkpmt9vyFlPK9Fd1ym+WyIGy /JUnKgL5x9EPRNY0GIuuSK1LxjpJsiACT4jB9N9TPwehYCC3N/2mOGRuS29bukhy9vtRfTWn1O2H LeYnmt/U+e6a5ParFrO5SO//soDkZqd/mpDbnP5VOfr8eEC3F/Dp673E9fHPM73/vDj7Sqdf+Vkm HBqFP7SYz8m3dezUB3X5qF+XN8uXTf0TL3v65KW+bHbv/zdH+9Q1AAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAACoAAHic7VlrbBRVFD5zZ7vtTku73e1uS+ljdluwNbVtgB8gxaClREJbDK36R5TFLHWl 2+0LZSU0pSAQJcbEhB+aKP4xmpiIiCGCMTXww8QEUQgxSKQGFI0EDQ81PlrPmTt3dmf7mtIt1Li3 Odlzvvuac8+937kzPfVF7tCb7xd+BwnlPpBheMQB9jhMQmHCcAKk69jwyMgIQTLKSKr8p8o6iOBf L6jQAB342w3RxK0wYfFCmhFzG+2Pk3yHDPLqVfFtnT89WPvz2hNSDuorFLGn1kI7BKc0Z3xRgEk0 N9P3npU+KkoeiPnr0f8wdOJzbISnpzy/C09E/Hpa6VOCsknXG6ALtkAAVz6Ez9GBz9asW89YWhX1 Fub3obwhcT0N5Z+4/sO6zIHxz3I28DPv1Pzn7ewTtE+V2Vtu5/k/UO4xADEAMAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACAACAP////// /////////wAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAFsAAAAVBwAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAA////////////////AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAD///////////////8AAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAP///////////////wAAAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAF5L5z8T9f7U+dfKnTr/R/Tz L/plgJkHqAguoFC5UbJQcmH0OXfo2PA092Kq3P4i7nR0d6PzS/mAuJzudrQnKLYU/0w9/pQXiP8p hzv1/eDS9wedKQ8QJwDkoxSgzEUpRJmHUoRSjFKBUgo8h5D4US9DKUeZj7IA5a64+lSZucJ0/n0I mT8IPVCr/VkvBci/EsTu/lb7vZQ+hUkmKfH+JG/U/8v8DVE65wuVIfmmjSJZr9vHbHtksjcb9QPM XJ+hRb7VqG/U3gbbonTO/2L9w3z8l7VMY2PgbA2Fgz1qc/BZdV0kHOiALKb0eVDUU1l9gLuO+Oce HJN+f5UY8HFrWGsJDVEVdej2nLm028KBrUGBBLyEtEfagqLPfLfWJyJaFGZrfUIdQfG0g3atRUjY XTZuE9stsOcg69F68jTJPXIC37Qfa1cciTF/S7SnNxgmS8Gs+oLWmCqvXHZB9rCjj3vEdI8k7ZSY i4QsJys8hol3J+LQptCT3ZGeyKZetaFrS6A3FOlQF1XXalS8ssXANHoWRvUiuLH0UJe16M92/q/U f1NlZsqd4v+rKf6fJfPPfv5v8dMQalTY50rN9r5Snh+EfbGI1wvu/35Uvlian5gvGr1an4iwT7vN 9vN6PhH2eSevn35+ueP8zxKxg7t8Y76/d646aytr5u/vg/qHYYo1g1Wfvqa/T+695IFL672aLkuS rCiSa2cbyDkSoNKKiSQXFNpXaNULeDOH2UAVGCh70e2LQxtibdnuMz5Ri9X7K4DtxzTBdi8HOZsG HgiDayCE0gEyo0S9QwXZNtaU1D39CtufcQVomtEN4+cRDbERzrgNZTvmvoF2lAgKOikZswdQts78 7GB4D5LedYKF0oegcn/zavrhnCFyv6QjYkOcxiTchTz9bjrV4U6Xz8rE9EVaLd0UHExoPAuQJjGi IdKqpQIdyzTaZbFBu1ru0s9AoZ4z6CYxiNohlCG8CnTb+Y2D2mRBU6D3qdZoZxB3oMS55ebfF7Vn 0LiFEbe0RMMbI+3Qt4KcsyuHxuYW0DnlvLxE47KKqJXx9pYr907MVbFxF2XREJUaF/3JuGegry/N T2uQxzpgZS5pblYrc42wkFtgXCNsWZ7AuEZYnVdgdV7+/O9Nth4fjP38sZspf/4q9pzGa9uiVsbF dVlmjcOr2BGNX7dHhQ9rioUPXCOss1RgXCPsAVVgXCOs3i+wer9F/w9b9X9H0ZT8r7Pq/yf+eP8P yk/oMey2c42w7nkC41oe65chm2P98tdlY+RV7ZmaWtWGrb3dAeC+fmjN10b7vkIpafy/U+f/IckL RWtmF/9bYt9JmXcqnJ/sGW+Z58cpzQb/m20z/39ezvm/zuD/15PC/xt8E/G/6psu/5+qVK5Oxv/V knX+Ly5Twlb5vzLTGv+fyxEcxjXCRoycMGLkhI9cAuMaYUeNnHA0zwpP4Xr8Mh4nTJT/Mxbw+Dca 8X8rKfHv908U/xX+6cZ/y93KkmTmf3+u8mWy8/8lp4gh1whzGPnfYeT/4wZ23MBOeAR2wmMl/rge S2ci/+O6fGU1/13wmPO/u0j4wDXClpQIjGuEXTOwawZ23bgTXFct+j/O/W20/+/Mm5L/p636n+0z 5/89HpHruUbYq4UC45o5/1f4reR/9HWcO9no/H94Lg3Rrr83K/J6n9l+xBNvH7M9nGm2b6Rxe7zn SrgrLbe2Vo32AwU0xDf6PHVscRnZv+v2Qpnbf+j2QYnbvxnt52t+XTbac/tHoz23fzC+W5zUzuxy w6+Qg8+f6Beh1r9jjF7v/vz47yldaa9od+IG47vF42r89xVFbi+Jb1/D2orj62vYo3nm9kGXuf2G XPP3mtUZ5u85j6Wb4z1HNdvf5iUv/nhWzliN/2f55vjv8pnjz+1Y/Lkdi39TgTn+3I7Fn9u3P/5v e83xP+Yxxz+71BzPxcXmeNYUmeN/wW1un5YQf7AQ/5n+/wYvsftl6v43u+5//wKFIdPIAAAAAAAA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA== ------=_NextPart_01C48638.50519170 Content-Location: file:///C:/28B2BDB1/QClusterICML1_files/filelist.xml Content-Transfer-Encoding: quoted-printable Content-Type: text/xml; charset="utf-8" ------=_NextPart_01C48638.50519170--