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

2026-04-21

AI Summary

This work addresses the long-overlooked intra-mode many-to-one collapse problem in generative adversarial networks (GANs), where multiple latent codes map to similar outputs, thereby reducing sample diversity. To mitigate this issue, the authors propose an adaptive pairing regularization method that jointly optimizes the generator while enforcing local consistency constraints. This approach dynamically adjusts the training process—enhancing sample coverage during under-exploration phases and improving generation fidelity once training stabilizes. The proposed mechanism is compatible with existing GAN stabilization techniques and readily integrates into diverse architectures. Empirical evaluations on both synthetic distributions and real-world image datasets demonstrate that the method significantly improves both recall and precision of generated samples, effectively compensating for limitations in current anti-collapse strategies.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

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.

Problem

• mode collapse
• intra-mode collapse
• many-to-one mapping
• generative adversarial networks
• latent space

Innovation

• pairing regularization
• many-to-one collapse
• intra-mode collapse
• generative adversarial networks
• local consistency