Poster: Abstract Knowledge Guides Search and Prediction in Novel Situations

Quite a mouthful, this was the title of a paper (co-authored with my advisor Patrick Shafto of the University of Louisville, as well as Chris Baker and Joshua Tenenbaum of the MIT Computational Cognitive Science Group) accepted into the Cognitive Science Society 2009 Conference Proceedings. Posted at left is a small version of the poster presentation that was given. You can see a larger version of the image by clicking on the one at right or you can download the full PDF of the image (432kb – 1.5 x 2 meters).

Abstract Knowledge Guides Search and Prediction in Novel Situations

The elevator pitch for the experiment and paper is that if individuals can formulate structured knowledge about a situation that gives rise to probabilistically determined events, then individuals should also be able to adjust their search strategies to account for the differences in the likelihood of certain evidence. I had a really great time presenting these findings in the poster session at the conference and fielding lots of good questions. I’m really enjoying this line of research and can’t wait to submit a full paper for the conference next year. This year’s conference was in Amsterdam and you can see my photos of that on my photostream at Flickr.

For those interesting in some of the technical details, the full PDF version of the poster is fairly readable and here is the abstract:

The Abstract

People combine their abstract knowledge about the world with data they have gathered in order to guide search and prediction in everyday life. We present a Bayesian model that formalizes knowledge transfer. Our model consists of two components: a hierarchical Bayesian model of learning and a Markov Decision Process modeling planning and search. An experiment tests qualitative predictions of the model, showing a strong fit between human data and model predictions. We conclude by discussing relations to previous work and future directions.