Mode Collapse and Regularity of Optimal Transportation Maps

TL;DR

This paper links optimal transportation theory with GANs to explain mode collapse and convergence issues, showing that non-convex data supports cause discontinuities in mappings, and proposes a new method to address these problems.

Contribution

It provides a theoretical explanation for GAN drawbacks based on optimal transport regularity and introduces a novel Brenier potential approximation method to mitigate mode collapse.

Findings

Discontinuous optimal maps occur with non-convex supports.

Autoencoder with discrete optimal transport confirms support non-convexity.

Proposed Brenier potential approximation improves mode collapse issues.

Abstract

This work builds the connection between the regularity theory of optimal transportation map, Monge-Amp\`{e}re equation and GANs, which gives a theoretic understanding of the major drawbacks of GANs: convergence difficulty and mode collapse. According to the regularity theory of Monge-Amp\`{e}re equation, if the support of the target measure is disconnected or just non-convex, the optimal transportation mapping is discontinuous. General DNNs can only approximate continuous mappings. This intrinsic conflict leads to the convergence difficulty and mode collapse in GANs. We test our hypothesis that the supports of real data distribution are in general non-convex, therefore the discontinuity is unavoidable using an Autoencoder combined with discrete optimal transportation map (AE-OT framework) on the CelebA data set. The testing result is positive. Furthermore, we propose to approximate the continuous Brenier potential directly based on discrete Brenier theory to tackle mode collapse. Comparing with existing method, this method is more accurate and effective.