Stabilizing Training of Generative Adversarial Networks through Regularization

Abstract

Deep generative models based on Generative Adversarial Networks (GANs) have demonstrated impressive sample quality but in order to work they require a careful choice of architecture, parameter initialization, and selection of hyper-parameters. This fragility is in part due to a dimensional mismatch or non-overlapping support between the model distribution and the data distribution, causing their density ratio and the associated f-divergence to be undefined. We overcome this fundamental limitation and propose a new regularization approach with low computational cost that yields a stable GAN training procedure. We demonstrate the effectiveness of this regularizer across several architectures trained on common benchmark image generation tasks. Our regularization turns GAN models into reliable building blocks for deep learning.

References in corpus (5)

• Improved Training of Wasserstein GANs
• Generative Moment Matching Networks
• Mode Regularized Generative Adversarial Networks
• Amortised MAP Inference for Image Super-resolution
• On integral probability metrics, ϕ-divergences and binary classification

Cited by in corpus (46)

• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• How to GAN LHC Events
• Towards the Automatic Anime Characters Creation with Generative Adversarial Networks
• Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers
• Review: Deep Learning in Electron Microscopy
• Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step
• ChatPainter: Improving Text to Image Generation using Dialogue
• Learning quantum data with the quantum Earth Mover's distance
• QuGAN: A Quantum State Fidelity based Generative Adversarial Network
• Stabilizing Generative Adversarial Networks: A Survey
• On gradient regularizers for MMD GANs
• Dropout-GAN: Learning from a Dynamic Ensemble of Discriminators
• Probability-Density-Based Deep Learning Paradigm for the Fuzzy Design of Functional Metastructures
• A Convex Duality Framework for GANs
• Understanding GANs: the LQG Setting
• Synthetic flow-based cryptomining attack generation through Generative Adversarial Networks
• Random Matrix Theory Proves that Deep Learning Representations of GAN-data Behave as Gaussian Mixtures
• Towards a Better Understanding and Regularization of GAN Training Dynamics
• Wasserstein-2 Generative Networks
• Some Theoretical Insights into Wasserstein GANs
• O-GAN: Extremely Concise Approach for Auto-Encoding Generative Adversarial Networks
• Negative Momentum for Improved Game Dynamics
• On the Effectiveness of Least Squares Generative Adversarial Networks
• Adversarially Robust Training through Structured Gradient Regularization
• Generative Adversarial Networks (GANs): What it can generate and What it cannot?
• Generative Convolution Layer for Image Generation
• GraN-GAN: Piecewise Gradient Normalization for Generative Adversarial Networks
• Dynamics of Fourier Modes in Torus Generative Adversarial Networks
• Training Generative Adversarial Networks via Primal-Dual Subgradient Methods: A Lagrangian Perspective on GAN
• Learning about an exponential amount of conditional distributions
• MR-GAN: Manifold Regularized Generative Adversarial Networks
• Autoencoding Generative Adversarial Networks
• Generating artificial digital image correlation data using physics-guided adversarial networks
• Generative Adversarial Forests for Better Conditioned Adversarial Learning
• K-Beam Minimax: Efficient Optimization for Deep Adversarial Learning
• Unpaired Multi-Domain Image Generation via Regularized Conditional GANs
• Alleviation of Gradient Exploding in GANs: Fake Can Be Real
• Adversarial Latent Autoencoder with Self-Attention for Structural Image Synthesis
• Adversarially Learned Mixture Model
• Functional Space Analysis of Local GAN Convergence
• Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution
• Adaptive Divergence for Rapid Adversarial Optimization
• Predicting Goal-directed Human Attention Using Inverse Reinforcement Learning
• Local Stability and Performance of Simple Gradient Penalty mu-Wasserstein GAN
• Diversity Regularized Adversarial Learning
• Solving Min-Max Optimization with Hidden Structure via Gradient Descent Ascent