******* Language, Cognition, and Computation Lecture Series *******

 

Title                    A Bayesian View of Inductive Learning in Humans and Machines

Speaker               Josh Tenenbaum

Affiliation           BCS

Date                    Tuesday, December 2, 2003

Time                    4:15pm

Location              E15-054

 

Abstract    

 

In everyday learning and reasoning, people routinely draw successful

generalizations from very limited evidence.  Even young children can infer

the meanings of words or the existence of hidden biological properties or

causal relations from just one or a few relevant observations -- far

outstripping the capabilities of conventional learning machines.  How do

they do it?  I will argue that the success of people's everyday inductive

leaps can be understood as the product of domain-general rational Bayesian

inferences constrained by people's implicit theories of the structure of

specific domains. This talk will explore the interactions between people's

domain theories and their everyday inductive leaps in several different

task domains, such as generalizing biological properties and learning word

meanings.  I will illustrate how domain theories generate the hypothesis

spaces necessary for Bayesian generalization, and how these theories may

themselves be acquired as the products of higher-order statistical

inferences.  I will also show how our approach to modeling human

learning motivates new machine learning techniques for semi-supervised

learning: generalizing from very few labeled examples with the aid of a

large sample of unlabeled data.

 Bio:

Josh Tenenbaum received his B.S. in Physics from Yale University in 1993,

and his Ph.D. in Brain and Cognitive Sciences from MIT in 1999. After a short

stint at Stanford Psych, he is back home at MIT, where he is an assistant professor

in BCS with a cross appointment in CSAIL.

 

 

*******************************************************************