AnalogySpace: Reducing the Dimensionality of Commonsense Knowledge

Henry Lieberman

(with Rob Speer and Catherine Havasi)

Media Laboratory

Massachusetts Institute of Technology

Friday, March 14, 2008
11:00 a.m. - 12:00 p.m.
Fuller Labs 320

We are interested in the problem of reasoning over very large common sense knowledge bases. Rather than absolute truth, we seek a reasoning method that computes similarity, analogies, tendencies, and relevance. We present AnalogySpace, which accomplishes this by forming the analogical closure of a semantic network through dimensionality reduction. It "self-organizes" concepts around dimensions such as good vs. bad, or easy vs. hard, delivering judgments about where concepts lie along these dimensions. Unlike logical chaining of assertions, it is more tolerant of knowledge that is imprecise or contradictory. Unlike statistical techniques such as Hidden Markov Models and Bayesian reasoning, it can provide Commonsense justifications for its decisions. An evaluation shows that users often agree with the predicted knowledge, and that its accuracy is an improvement over previous techniques.

http://www.media.mit.edu/~lieber/

http://analogyspace.media.mit.edu/

_____

Henry Lieberman is a Research Scientist at the MIT Media Laboratory. He works with the Agents Group. He is especially interested in combining artificial intelligence with interactive graphics and human interface ideas.

He is working on building software agents for interactive graphical applications that can learn from examples demonstrated by a user.

Host: Charles Rich

Refreshments will be served.

Maintained by webmaster@cs.wpi.edu
Last modified: March 11, 2008