Error-Driven Stochastic Search for Theories and Concepts

Owen LewisMIT, Cambridge, MA, USA
Santiago PerezMIT, Cambridge, MA, USA
Josh TenenbaumMIT, Cambridge, MA, USA

Abstract

Bayesian models have been strikingly successful in a wide range of domains. However, the stochastic search algorithms generally used by these models have been criticized for not capturing the error-driven nature of human learning. Here, we incorporate error-driven proposals into a stochastic search algorithm and evaluate its performance on concept and theory learning problems. Compared to a model with random proposals, we find that error-driven search requires fewer proposals and fewer evaluations against labelled data.

Files

Error-Driven Stochastic Search for Theories and Concepts (1.1 MB)



Back to Table of Contents