Abstract: Selecting informative features is a central problem in learning. In this talk, we define the information measure for feature functions based on the geometry of functional space. We show that neural networks can be configured to perform some basic geometric operations on feature functions such as decompositions and projections. By connecting such basic building blocks, we demonstrate some new ways to incorporate domain knowledge or structural constraints in the learning process, and to solve a “learning with side information” problem, which is an attempt to the multi-task learning problem with insights from multi-terminal information theory.

Bio: Lizhong Zheng received the B.S and M.S. degrees from the Department of Electronic Engineering, Tsinghua University, China, and the Ph.D. degree from the Department of Electrical Engineering and Computer Sciences, University of California, Berkeley. Since graduation, he has been working at MIT, where he is currently a professor of Electrical Engineering. His research interests include information theory, statistical inference, communications, and networks theory. He received the IEEE Information Theory Society Paper Award, NSF CAREER award, the AFOSR Young Investigator Award.