Bayesian models of inductive learning and reasoning

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 operating over intuitive theories of the structure of specific domains. This talk will focus on learning about the structure of biological species domains, including the properties and relationships between kinds. I will show how domain theories generate the hypothesis spaces necessary for Bayesian generalization, and how these theories may themselves be acquired through higher-order statistical inferences in hierarchical Bayesian models. Relations to alternative modeling approaches based on exemplar-similarity and connectionist networks will also be discussed. Time permitting, I will show how our approach to modeling human learning motivates new machine learning techniques for semi-supervised classification: generalizing concepts from very few labeled examples with the aid of a large sample of unlabeled data.

(You can also visit this page to check events on additional dates.)