Towards Principled Methods for Training Generative Adversarial Networks

Abstract

The goal of this paper is not to introduce a single algorithm or method, but to make theoretical steps towards fully understanding the training dynamics of generative adversarial networks. In order to substantiate our theoretical analysis, we perform targeted experiments to verify our assumptions, illustrate our claims, and quantify the phenomena. This paper is divided into three sections. The first section introduces the problem at hand. The second section is dedicated to studying and proving rigorously the problems including instability and saturation that arize when training generative adversarial networks. The third section examines a practical and theoretically grounded direction towards solving these problems, while introducing new tools to study them.

References in corpus (3)

• Explaining and Harnessing Adversarial Examples
• Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks
• Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling

Cited by in corpus (20)

• Towards the Automatic Anime Characters Creation with Generative Adversarial Networks
• ChemGAN challenge for drug discovery: can AI reproduce natural chemical diversity?
• Model-based Deep Reinforcement Learning for Dynamic Portfolio Optimization
• On denoising autoencoders trained to minimise binary cross-entropy
• Spatial Evolutionary Generative Adversarial Networks
• Tutorial: Deriving the Standard Variational Autoencoder (VAE) Loss Function
• Comparison of Maximum Likelihood and GAN-based training of Real NVPs
• Face Transfer with Generative Adversarial Network
• Machine Learning Methods Economists Should Know About
• High-Quality Face Image SR Using Conditional Generative Adversarial Networks
• eCommerceGAN : A Generative Adversarial Network for E-commerce
• An Improved Self-supervised GAN via Adversarial Training
• LGAN: Lung Segmentation in CT Scans Using Generative Adversarial Network
• On Relativistic $f$-Divergences
• Metric Learning-based Generative Adversarial Network
• Artist Style Transfer Via Quadratic Potential
• Efficient Super Resolution For Large-Scale Images Using Attentional GAN
• Classification of sparsely labeled spatio-temporal data through semi-supervised adversarial learning
• Diversity Regularized Adversarial Learning
• A Self-Training Method for Semi-Supervised GANs