Deep generative models via explicit Wasserstein minimization

This thesis provides a procedure to fit generative networks to target distributions, with the goal of a small Wasserstein distance (or other optimal transport costs). The approach is based on two principles: (a) if the source randomness of the network is a continuous distribution (the “semi-discrete” setting), then the Wasserstein distance is realized by a deterministic optimal transport mapping; (b) given an optimal transport mapping between a generator network and a target distribution, the Wasserstein distance may be decreased via a regression between the generated data and the mapped target points. The procedure here therefore alternates these two steps, forming an optimal transport and regressing against it, gradually adjusting the generator network towards the target distribution. Mathematically, this approach is shown to minimize the Wasserstein distance to both the empirical target distribution, and also its underlying population counterpart. Empirically, good performance is demonstrated on the training and testing sets of the MNIST and Thin-8 data. As a side product, the thesis proposes several effective metrics of measure performance of deep generative models. The thesis closes with a discussion of the unsuitability of the Wasserstein distance for certain tasks, as has been identified in prior work.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Chen, Yucheng

Contributors dc:contributor

• Peng, Jian

• Deep Generative Models
• Generative Adversarial Networks
• Optimal Transport

Statement dc:rights

• Copyright 2019 Yucheng Chen

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/104932

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/104932