Files in this item



application/pdfDESHPANDE-THESIS-2018.pdf (14MB)
(no description provided)PDF


Title:Generative modeling using the sliced Wasserstein distance
Author(s):Deshpande, Ishan
Advisor(s):Schwing, Alexander G.
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Machine Learning
Deep Learning
Abstract:Generative adversarial nets (GANs) are very successful at modeling distributions from given samples, even in the high-dimensional case. However, their formulation is also known to be hard to optimize and often unstable. While the aforementioned problems are particularly true for early GAN formulations, there has been significant empirically motivated and theoretically founded progress to improve stability, for instance, by using the Wasserstein distance rather than the Jenson-Shannon divergence. Here, we consider an alternative formulation for generative modeling based on random projections which, in its simplest form, results in a single objective rather than a saddlepoint formulation. By augmenting this approach with a discriminator we improve its accuracy. We found our approach to be significantly more stable compared to even the improved Wasserstein GAN. Further, unlike the traditional GAN loss, the loss formulated in our method is a good measure of the actual distance between the distributions and, for the first time for GAN training, we are able to show estimates for the same.
Issue Date:2018-04-11
Rights Information:Copyright 2018 Ishan Deshpande
Date Available in IDEALS:2018-09-04
Date Deposited:2018-05

This item appears in the following Collection(s)

Item Statistics