Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision
(CS 528)

Bayesian models of human learning and reasoning

Josh Tenenbaum, MIT
March 7, 2005, 4:15PM
TCSeq 200


In the last decade, Bayesian methods have revolutionized major areas of artificial intelligence, machine learning, and natural language processing. In contrast, Bayesian methods have not yet achieved nearly the same success among cognitive scientists trying to explain how humans learn, reason and communicate. In this talk I will sketch some of the challenges and prospects for Bayesian models in cognitive science, and also draw some lessons for advancing the state of the art in probabilistic approaches to artificial intelligence.

I will focus on everyday reasoning tasks where people can routinely draw successful generalizations from very limited evidence. These generalizations can be modeled as Bayesian inferences constrained by people's intuitive theories about the causal structure of the world. I will present several case studies drawn from task domains such as diagnostic reasoning, predicting the duration of events, inferring the properties of biological species, and learning physical laws. Time permitting, I will also talk about some recent work on how people might learn their abstract theories about the structure of these domains, and some applications of our models to problems in machine learning such as semi-supervised classification and relational clustering.

About the Speaker

I study the computational basis of human learning and inference. Through a combination of mathematical modeling, computer simulation, and behavioral experiments, I try to uncover the logic behind our everyday inductive leaps: constructing perceptual representations, separating "style" and "content" in perception, learning concepts and words, judging similarity or representativeness, inferring causal connections, noticing coincidences, predicting the future. I approach these topics with a range of empirical methods -- primarily, behavioral testing of adults, children, and machines -- and formal tools -- drawn chiefly from Bayesian statistics and probability theory, but also from geometry, graph theory, and linear algebra. My work is driven by the complementary goals of trying to achieve a better understanding of human learning in computational terms and trying to build computational systems that come closer to the capacities of human learners.


Back to the Colloquium Page