A Bayesian View of Inductive Learning in Humans and Machines - 2004

Explore a comprehensive lecture on the Bayesian approach to inductive learning in humans and machines, delivered by Josh Tenenbaum from MIT in 2004 at the Center for Language & Speech Processing (CLSP), JHU. Delve into the fascinating world of human cognition and machine learning as Tenenbaum explains how people, even young children, can make successful generalizations from limited evidence. Discover the role of domain-general rational Bayesian inferences constrained by implicit theories in various task domains, including biological property generalization and word meaning acquisition. Examine the interaction between domain theories and everyday inductive leaps, and learn how these theories generate hypothesis spaces for Bayesian generalization. Investigate the potential for acquiring these theories through higher-order statistical inferences. Finally, uncover how this approach to modeling human learning inspires new machine learning techniques for semi-supervised learning, enabling generalizations from minimal labeled examples with the aid of large unlabeled datasets. This 1 hour and 26 minute talk offers valuable insights for researchers, students, and professionals interested in cognitive science, artificial intelligence, and machine learning.