| Building WatsonAn overview of the DeepQA Project Stelling1 #246911 David Ferrucci
Watson Principal Investigator and IBM Fellow
DeepQA Team @ IBM Research |
Topics A Grand Challenge Opportunity Want to Play Chess or Just Chat? Easy Questions? Hard Questions? The Jeopardy! Challenge The Possibilities Multiply Broad Domain Inducing Frames Evaluating Possibilities and Their Evidence Keyword Evidence Deeper Evidence Not Just for Fun Divide and Conquer The Missing Link What It Takes to compete against Top Human Jeopardy! Players Early Experiments on Focused Content A Few Guiding Principles - Specific Large Hand-Crafted Models Won’t Cut It
- Intelligence from many diverse methods
- Massive Parallelism is a Key Enabler
DeepQA: The Technology Behind WatsonMassively WatsonMassively Parallel Probabilistic Evidence-Based Architecture A METHODOLOGY FOR RAPID ADVANCEMENT Methodology Goal-Oriented System-Level Metrics and Longer-Term Incentives Extreme Collaboration Diverse Skills ALL IN One Room Disciplined Engineering and Evaluation Grouping Features to produce Evidence Profiles Evidence: Time, Popularity, Source, Classification etc. Evidence: Puns In-Category Learning DeepQA: Incremental Progress in Answering Precision NLP Components Technology Performance Question Processing
Relation Extraction
Linguistic Frame Extraction
Passage Matching Ensemble
KAFE: Knowledge from Extracted Content One Jeopardy! question can take 2 hours on a single 2.6Ghz Core Run-Time Stack Game Strategy With Precision, Accurate Confidence and Speed, the rest was History Potential Business Applications Evidence Profiles from disparate data is a powerful idea DeepQA in Continuous Evidence-Based Diagnostic Analysis The Core Technical Team TORONTO? Categories are not as simple as they seem Toronto vs. Chicago
|