Speaker: Erin Grant, Ph.D. candidate, Dept. of Electrical Engineering & Computer Sciences at UC Berkeley which is affiliated with the Berkeley Artificial Intelligence Research (BAIR) Lab at UC Berkeley and the Computational Cognitive Science (CoCoSci) Lab at Princeton University.
She has been a research intern and a student researcher on the Google Brain Team and a research intern at OpenAI and received a B.Sc. from the University of Toronto in Computer Science and Statistics. Her research interests lie in understanding prior knowledge in human and machine learning using the lens of meta-learning.
Title: Investigating Inductive Biases in Humans and Machines
Abstract:
Meta-learning describes how an intelligent agent leverages prior learning episodes as a basis for quickly improving performance on a novel task. Bayesian hierarchical modeling provides a theoretical framework for formalizing this capability as statistical inference for a set of parameters that are shared as an initialization for domain-specific learners. She will demonstrate how to reformulate recent algorithms for meta-learning as methods for statistical inference in a hierarchical Bayesian model, thus bridging these two independent approaches. This connection provides a means to understand and improve on the computational underpinnings of algorithms for meta-learning in machine learning, as well as to adapt cognitive models expressed in the language of hierarchical modeling to the modern machine learning toolbox and to naturalistic stimuli.
Video: Youtube