Scaling-based Data Augmentation for Generative Models and its Theoretical Extension

TL;DR

This paper introduces Scale-GAN, a novel data scaling approach for stable and high-quality generative model training, supported by theoretical analysis and empirical validation on benchmark datasets.

Contribution

It reveals data scaling as a key factor for stable learning and proposes Scale-GAN, integrating data scaling and variance regularization, with theoretical bias-variance trade-off analysis.

Findings

Scale-GAN improves stability and data quality in generative models.

Data scaling controls the bias-variance trade-off in estimation error.

Empirical results outperform existing methods on benchmark datasets.

Abstract

This paper studies stable learning methods for generative models that enable high-quality data generation. Noise injection is commonly used to stabilize learning. However, selecting a suitable noise distribution is challenging. Diffusion-GAN, a recently developed method, addresses this by using the diffusion process with a timestep-dependent discriminator. We investigate Diffusion-GAN and reveal that data scaling is a key component for stable learning and high-quality data generation. Building on our findings, we propose a learning algorithm, Scale-GAN, that uses data scaling and variance-based regularization. Furthermore, we theoretically prove that data scaling controls the bias-variance trade-off of the estimation error bound. As a theoretical extension, we consider GAN with invertible data augmentations. Comparative evaluations on benchmark datasets demonstrate the effectiveness of our method in improving stability and accuracy.