McGan: Mean and Covariance Feature Matching GAN

Abstract

We introduce new families of Integral Probability Metrics (IPM) for training Generative Adversarial Networks (GAN). Our IPMs are based on matching statistics of distributions embedded in a finite dimensional feature space. Mean and covariance feature matching IPMs allow for stable training of GANs, which we will call McGan. McGan minimizes a meaningful loss between distributions.

15 pages; published at ICML 2017

References in corpus (3)

• Conditional Generative Adversarial Nets
• BEGAN: Boundary Equilibrium Generative Adversarial Networks
• Generative Moment Matching Networks

Cited by in corpus (14)

• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• CVAE-GAN: Fine-Grained Image Generation through Asymmetric Training
• Activation Maximization Generative Adversarial Nets
• Low-shot Learning via Covariance-Preserving Adversarial Augmentation Networks
• Memorization Precedes Generation: Learning Unsupervised GANs with Memory Networks
• Global Convergence to the Equilibrium of GANs using Variational Inequalities
• Robust Estimation and Generative Adversarial Nets
• Learning Implicit Generative Models with the Method of Learned Moments
• Correcting differences in multi-site neuroimaging data using Generative Adversarial Networks
• Reciprocal Adversarial Learning via Characteristic Functions
• Stabilizing Training of Generative Adversarial Nets via Langevin Stein Variational Gradient Descent
• Semi-Supervised Learning with IPM-based GANs: an Empirical Study
• Adversarial Code Learning for Image Generation
• Linear Discriminant Generative Adversarial Networks