Wlodek Zadrozny - Building Watson: An overview of the Deep Question Answering project

14:00
Jeudi
6
Oct
2011
Organisé par : 

Dominique Vaufreydaz

Intervenant : 

Wlodek Zadrozny d’IBM


Information détaillée : 

Dr. Wlodek Zadrozny is a Research Staff Member in the IBM Thomas J. Watson Research Center. He received his M.S. degree in mathematics from Warsaw University in 1976, and a Ph.D. degree in mathematics from Polish Academy of Science in 1980. In 1982-1983 he visited the University of Heidelberg and RWTH Aachen as an Alexander von Humboldt fellow. He joined IBM Research in 1985 and worked mostly on applications of natural language processing. In the Watson/DeepQA project he was responsible for acquiring and preparing textual resources. He also worked on applying this question answering technology to help desk support and medical diagnosis. Dr. Zadrozny is an author or co-author of about 40 patents and 50 technical papers. He is a member of the IEEE, ACM and AAAI.

This talk is organised collaboratively by XRCE and LIG and is part of the LIG keynote series. It will be given at ENSIMAG and simultaneously video-streamed to XRCE.

Résumé : 

In the early 2011, an IBM computer named Watson won against two human champions in Jeopardy!, an American quiz show covering broad domains of knowledge. Watson proved capable of answering questions about history, literature, pop culture, science, geography, and more. It showed that a machine is capable of engaging in an open domain question answering.

Attaining champion-level performance in Jeopardy! requires a massively parallel computer with an access to a large body of knowledge and a novel architecture integrating natural language processing, information retrieval, knowledge representation, reasoning, and machine learning. Parallel processing allows Watson to evaluate many competing hypotheses and combine sets of evidence using an array of complex algorithms.

This presentation will introduce the Jeopardy! challenge, cover Watson’s architecture and performance, as well as discuss some future applications.