ALICE: Towards Understanding Adversarial Learning for Joint Distribution Matching
This addresses a fundamental problem in adversarial learning for researchers, but it appears incremental as it builds on existing joint distribution matching frameworks.
The paper tackled non-identifiability issues in bidirectional adversarial training for joint distribution matching by proposing adversarial and non-adversarial approaches based on conditional entropy, stabilizing unsupervised learning and extending to semi-supervised tasks, with validation on synthetic and real-world data.
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.