Sign inSign up
How Watson Learns Superhuman Jeopardy! Strategies thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

How Watson Learns Superhuman Jeopardy! Strategies

Published on Nov 07, 20132837 Views

Gerry Tesauro

Major advances in Question Answering technology were needed for Watson to play Jeopardy! at championship level -- the show requires rapid-fire answers to challenging natural language questions, broad

LSOLDM 2013 - Cumberland Lodge

Related categories

Decision SupportOn-line Learning

Chapter list

How Watson Learns Superhuman Jeopardy! Strategies00:00
Outline00:49
Jeopardy! Gameplay Basics03:44
The Jeopardy! Challenge05:13
DeepQA11:30
Watson’s Competitive Record17:18
Four Challenging Aspects of Jeopardy! Strategy20:02
Motivating Quantitative J! Strategy27:07
Why Build a Jeopardy Simulator?29:32
Approach to Modeling Jeopardy!33:57
Components of a J! Game Simulator36:43
Modeling Daily Double Placement39:08
Modeling Human Final Jeopardy40:39
Stochastic Process Model of Regular Clues44:55
Watson’s Daily Double Betting Strategy46:34
Simplified List of “State” Variables47:56
Computing Optimal DD Bet48:49
Illustrative Example of NN DD betting51:03
Demo51:51
NNDD Analysis - 152:52
NNDD Analysis - 253:33
NN DD Performance Metrics55:14
Watson’s Final Jeopardy Betting Strategy57:04
Watson’s Clue Selection Strategy58:43
Daily Double Seeking Animation01:00:07
Endgame Buzz Threshold01:00:57
Dynamic Programming Solution01:01:57
Buzz Threshold -- Illustrative Examples01:03:01
Conclusions01:05:40