Deep Representation Interface for Engineering Problems

Abstract: Using deep neural networks (DNNs) as elements of engineering solutions can potentially enhance the system’s overall performance. However, existing black-box practices with DNNs are incompatible with the modularized design of engineering systems. To tackle this problem, we propose using feature representations as the interface for integrating DNN modules into engineering solutions. We provide mathematical characterizations of such representations based on application requirements. Our development is built on a new information-theoretic framework that connects feature learning with statistical dependence. This fundamental connection allows a unified approach to analyzing and designing information extraction processes in learning. In particular, we leverage the connection to 1) introduce a new metric that measures the semantics of the information contents carried by features and 2) design a new DNN architecture that separates the contributions of different variables to their multivariate dependence. As an illustrating example, we consider symbol detection in wireless fading interference channels, a classical wireless communication problem requiring nonlinear and adaptive processing. We illustrate how to incorporate engineering domain knowledge in our learning-based receiver design and demonstrate its superior performance.

Bio: Xiangxiang Xu received the B.Eng. and Ph.D. degrees in electronic engineering from Tsinghua University, Beijing, China, in 2014 and 2020, respectively. He is a postdoctoral associate in the Department of EECS at MIT. His research focuses on information theory, statistical learning, representation learning, and their applications in understanding and developing learning algorithms. He is a recipient of the 2016 IEEE PES Student Prize Paper Award in Honor of T. Burke Hayes and the 2024 ITA (Information Theory and Applications) Workshop Sand Award.

*** The event will take place in a hybrid format. The location for in-person attendance will be room 3427 EECS. Attendance will also be available via Zoom.

Join Zoom Meeting: https://umich.zoom.us/j/93679028340

Meeting ID: 936 7902 8340

Passcode: XXX (Will be sent via email to attendees)

Zoom Passcode information is available upon request to Kristi Rieger ([email protected]).