Our understanding of GAN (generative adversarial net) training is still very limited since it is a non-convex-non-concave min-max optimization. As a result, most recent studies focused on local analysis. In this talk, we discuss how to perform a global analysis of GANs. We prove that the original JS-GAN has exponentially many bad strict local minima which are perceived as mode-collapse. We show that a 2-line modification to JS-GAN called relativistic standard GAN (RS-GAN) eliminates all bad basins. We extend the two results to a large class of losses as well: for separable GANs (including JS-GAN, hinge-GAN, LS-GAN) exponentially many bad basins exist, while for R-GANs no bad basins exist. The effectiveness of R-GANs were verified by a few empirical works before (e.g. ESR-GAN in super resolution). Based on theory, we predict that R-GANs has a bigger advantage for narrower neural nets, and our experiments verify that R-GANs (e.g. RS-GAN) can beat their separable counter-parts (e.g. JS-GAN) by 5-10 FID scores in narrower nets. Our theory also implies that the advantage is larger for higher-dimensional images; we show that for high-resolution images like LSUN, while JS-GAN only generates noise, RS-GAN can generate quite good images.
August 2020
10:30am - 11:30am
Department of Mathematics
Payment Details
Alumni, Faculty and Staff, PG Students, UG Students
Other Events
Seminar, Lecture, Talk
MATH - Seminar on Data Science - Theory of Deep Convolutional Neural Networks
Deep learning has been widely applied and brought breakthroughs in speech recognition, computer vision, and many other domains. The involved deep neural network architectures and computational issues ...
Nov 2020
Seminar, Lecture, Talk
MATH - Seminar on Applied Mathematics - Thirty Years of Applied Mathematics
The 50's to the 80's saw tremendous growth of applied math, driven mainly by PDEs and numerical algorithms. The integration of the two produced the "Courant School", which has had a far-reaching impac...
Nov 2020