On distinguishability criteria for estimating generative models

Abstract

Two recently introduced criteria for estimation of generative models are both based on a reduction to binary classification. Noise-contrastive estimation (NCE) is an estimation procedure in which a generative model is trained to be able to distinguish data samples from noise samples. Generative adversarial networks (GANs) are pairs of generator and discriminator networks, with the generator network learning to generate samples by attempting to fool the discriminator network into believing its samples are real data. Both estimation procedures use the same function to drive learning, which naturally raises questions about how they are related to each other, as well as whether this function is related to maximum likelihood estimation (MLE). NCE corresponds to training an internal data model belonging to the {\em discriminator} network but using a fixed generator network. We show that a variant of NCE, with a dynamic generator network, is equivalent to maximum likelihood estimation. Since pairing a learned discriminator with an appropriate dynamically selected generator recovers MLE, one might expect the reverse to hold for pairing a learned generator with a certain discriminator. However, we show that recovering MLE for a learned generator requires departing from the distinguishability game. Specifically: (i) The expected gradient of the NCE discriminator can be made to match the expected gradient of MLE, if one is allowed to use a non-stationary noise distribution for NCE, (ii) No choice of discriminator network can make the expected gradient for the GAN generator match that of MLE, and (iii) The existing theory does not guarantee that GANs will converge in the non-convex case. This suggests that the key next step in GAN research is to determine whether GANs converge, and if not, to modify their training algorithm to force convergence.

This version adds a figure that appeared on the poster at ICLR, changes the template to say that the paper was accepted as a workshop contribution (previously it was under a review as a conference submission), and fixes some typos

Cited by in corpus (32)

• Generative Adversarial Networks: An Overview
• Improved Training of Wasserstein GANs
• NIPS 2016 Tutorial: Generative Adversarial Networks
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• Variational Approaches for Auto-Encoding Generative Adversarial Networks
• Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning
• Adversarial Ranking for Language Generation
• Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities
• Censoring Representations with an Adversary
• Hierarchical Implicit Models and Likelihood-Free Variational Inference
• Connecting Generative Adversarial Networks and Actor-Critic Methods
• Contrastive Representation Distillation
• Maximum Entropy Generators for Energy-Based Models
• Improved generator objectives for GANs
• Generative Adversarial Nets from a Density Ratio Estimation Perspective
• Distribution Matching in Variational Inference
• Face Super-Resolution Through Wasserstein GANs
• Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields
• Skill Rating for Generative Models
• Fast Variational AutoEncoder with Inverted Multi-Index for Collaborative Filtering
• Adversarial Contrastive Estimation
• On the Implicit Assumptions of GANs
• Sampling-Decomposable Generative Adversarial Recommender
• Rare Disease Detection by Sequence Modeling with Generative Adversarial Networks
• Self-Adversarial Learning with Comparative Discrimination for Text Generation
• Super-Resolution via Conditional Implicit Maximum Likelihood Estimation
• Infomax Neural Joint Source-Channel Coding via Adversarial Bit Flip
• A Novel Framework for Selection of GANs for an Application
• Semi-Implicit Generative Model
• Sequential Adversarial Anomaly Detection for One-Class Event Data
• Evaluating a Generative Adversarial Framework for Information Retrieval
• Multimodal Image Synthesis with Conditional Implicit Maximum Likelihood Estimation