Description:

This talk will be split into two parts. In the first half, I will discuss my work on developing a novel approach to record neural activity in vivo from hundreds of neurons across the transverse plane of the hippocampus. This has enabled experiments to probe the nature of the hippocampal circuit to an extent not possible with prior technology, and will allow for greater testing of theories regarding how the hippocampus supports crucial cognitive processes, such as episodic and spatial memory. In the second half of the talk, I will present work of mine that has developed a renormalization group framework for studying properties of sparse deep neural networks. This novel, but natural, viewpoint leverages the rich theory developed on the renormalization group to study universal behavior of “winning tickets”, a class of sparse deep neural networks that have been found to learn to perform well on a variety of tasks. The renormalization group inspired analysis provides new insight on existing experimental results regarding winning tickets and enables a better understanding of their capacity for task transfer.

 William Redman, PhD Student, University of California, Santa Barbara

Host:  Istvan Kovacs