Pairing Regularization for Mitigating Many-to-One Collapse in GANs

arXiv cs.LG / 4/23/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses a less-studied GAN failure case: intra-mode (many-to-one) collapse, where different latent codes produce the same or very similar outputs.
It introduces a pairing regularizer that is jointly optimized with the generator to enforce local consistency between latent variables and generated samples.
The authors show that the regularizer’s benefit depends on the training failure regime: it promotes structured local exploration when exploration is limited, improving coverage/recall.
In more stable settings with enough exploration, the method improves precision by discouraging redundant mappings, while maintaining recall.
Experiments across toy distributions and real-image benchmarks indicate the regularizer complements existing GAN stabilization methods by directly targeting intra-mode collapse.

Abstract

Mode collapse remains a fundamental challenge in training generative adversarial networks (GANs). While existing works have primarily focused on inter-mode collapse, such as mode dropping, intra-mode collapse-where many latent variables map to the same or highly similar outputs-has received significantly less attention. In this work, we propose a pairing regularizer jointly optimized with the generator to mitigate the many-to-one collapse by enforcing local consistency between latent variables and generated samples. We show that the effect of pairing regularization depends on the dominant failure mode of training. In collapse-prone regimes with limited exploration, pairing encourages structured local exploration, leading to improved coverage and higher recall. In contrast, under stabilized training with sufficient exploration, pairing refines the generator's induced data density by discouraging redundant mappings, thereby improving precision without sacrificing recall. Extensive experiments on both toy distributions and real-image benchmarks demonstrate that the proposed regularizer effectively complements existing stabilization techniques by directly addressing intra-mode collapse.