ALICE: Towards Understanding Adversarial Learning for Joint Distribution Matching

TL;DR

This paper explores the challenges of adversarial training for joint distribution matching, proposing unified methods that improve stability and extend to semi-supervised learning, validated through theoretical and empirical results.

Contribution

It introduces a unified framework for adversarial and non-adversarial joint distribution matching, enhancing stability and extending applicability to semi-supervised learning.

Findings

Stabilized bidirectional adversarial learning methods.

Unified a broad family of adversarial models.

Validated approaches on synthetic and real-world data.

Abstract

We investigate the non-identifiability issues associated with bidirectional adversarial training for joint distribution matching. Within a framework of conditional entropy, we propose both adversarial and non-adversarial approaches to learn desirable matched joint distributions for unsupervised and supervised tasks. We unify a broad family of adversarial models as joint distribution matching problems. Our approach stabilizes learning of unsupervised bidirectional adversarial learning methods. Further, we introduce an extension for semi-supervised learning tasks. Theoretical results are validated in synthetic data and real-world applications.