Department of Electrical and Computer Engineering Seminar
Speaker: Dr. Maziar Sanjabi, Post-Doctoral Scholar, Department of Data Science and Operations, University of Southern California
Date: Friday, February 8th, 2019
Location: 238 SES
Title: An Optimization Perspective on GANs
Abstract: Generative Adversarial Networks (GANs) provide a popular way of learning generative models, especially for applications such as image generation where they are capable of generating very sharp fake images. On the other hand, GANs are notoriously difficult to train and suffer from issues such as non-convergence and mode collapse. In this talk, we draw upon tools from the optimization literature to design a new objective for learning GANs. We show, both in theory and practice, that using our proposed objective leads to convergent and stable training of GANs.
Bio: Dr. Maziar Sanjabi is a postdoctoral scholar with the Department of Data Science and Operations at USC. Prior to joining USC, he was a postdoctoral scholar with the Computer Science Department at UCLA. He obtained his PhD and Masters in Electrical Engineering, with minors in Computer Science and Math, at University of Minnesota. He was also a DSP firmware engineer in Starkey Hearing Technologies’ R&D division where he worked on developing the next generation of smart hearing aids. His general research interest is in designing scalable methods for statistical learning with broad applications in machine learning, artificial intelligence, signal processing and wireless communications.
Host: Dr. Mojtaba Soltanalian, Assistant Professor, firstname.lastname@example.org