Boundary-Seeking Generative Adversarial Networks

Abstract

Generative adversarial networks (GANs) are a learning framework that rely on training a discriminator to estimate a measure of difference between a target and generated distributions. GANs, as normally formulated, rely on the generated samples being completely differentiable w.r.t. the generative parameters, and thus do not work for discrete data. We introduce a method for training GANs with discrete data that uses the estimated difference measure from the discriminator to compute importance weights for generated samples, thus providing a policy gradient for training the generator. The importance weights have a strong connection to the decision boundary of the discriminator, and we call our method boundary-seeking GANs (BGANs). We demonstrate the effectiveness of the proposed algorithm with discrete image and character-based natural language generation. In addition, the boundary-seeking objective extends to continuous data, which can be used to improve stability of training, and we demonstrate this on Celeba, Large-scale Scene Understanding (LSUN) bedrooms, and Imagenet without conditioning.

References in corpus (7)

• Improved Training of Wasserstein GANs
• NIPS 2016 Tutorial: Generative Adversarial Networks
• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• Adversarial Learning for Neural Dialogue Generation
• Blocks and Fuel: Frameworks for deep learning
• Hierarchical Implicit Models and Likelihood-Free Variational Inference
• Variational Inference using Implicit Distributions

Cited by in corpus (49)

• Improved Training of Wasserstein GANs
• Style Transfer from Non-Parallel Text by Cross-Alignment
• Objective-Reinforced Generative Adversarial Networks (ORGAN) for Sequence Generation Models
• How Generative Adversarial Networks and Their Variants Work: An Overview
• MaskGAN: Better Text Generation via Filling in the______
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• NetGAN: Generating Graphs via Random Walks
• Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
• Differentially Private Synthetic Medical Data Generation using Convolutional GANs
• Generative Adversarial Networks (GANs) in Networking: A Comprehensive Survey & Evaluation
• A survey on text generation using generative adversarial networks
• Generating and designing DNA with deep generative models
• Generative Models for Automatic Chemical Design
• Language Generation with Recurrent Generative Adversarial Networks without Pre-training
• A review of Generative Adversarial Networks for Electronic Health Records: applications, evaluation measures and data sources
• Robust Imitation of Diverse Behaviors
• Implicit Maximum Likelihood Estimation
• Learning Independent Features with Adversarial Nets for Non-linear ICA
• Generative adversarial networks in time series: A survey and taxonomy
• Human Motion Modeling using DVGANs
• Adversarially Regularized Autoencoders
• Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields
• Low-Resource Text Classification using Domain-Adversarial Learning
• Unsupervised Cipher Cracking Using Discrete GANs
• Comparative Study on Generative Adversarial Networks
• Neural Language Generation: Formulation, Methods, and Evaluation
• Gang of GANs: Generative Adversarial Networks with Maximum Margin Ranking
• CorGAN: Correlation-Capturing Convolutional Generative Adversarial Networks for Generating Synthetic Healthcare Records
• The Six Fronts of the Generative Adversarial Networks
• Exploring Bias in GAN-based Data Augmentation for Small Samples
• Tail of Distribution GAN (TailGAN): Generative-Adversarial-Network-Based Boundary Formation
• Boundary of Distribution Support Generator (BDSG): Sample Generation on the Boundary
• Pulsar Candidate Identification with Artificial Intelligence Techniques
• Discriminator Contrastive Divergence: Semi-Amortized Generative Modeling by Exploring Energy of the Discriminator
• Training Generative Adversarial Networks with Binary Neurons by End-to-end Backpropagation
• Cycle-Consistent Adversarial Autoencoders for Unsupervised Text Style Transfer
• RankGAN: A Maximum Margin Ranking GAN for Generating Faces
• GAN-based Generation and Automatic Selection of Explanations for Neural Networks
• Fonts-2-Handwriting: A Seed-Augment-Train framework for universal digit classification
• Variance Regularizing Adversarial Learning
• Synthesizing brain tumor images and annotations by combining progressive growing GAN and SPADE
• MMGAN: Manifold Matching Generative Adversarial Network
• Implementing a Hierarchical Deep Learning Approach for Simulating Multi-Level Auction Data
• Training Generative Adversarial Networks in One Stage
• Adversarial Neural Networks for Cross-lingual Sequence Tagging
• DepthwiseGANs: Fast Training Generative Adversarial Networks for Realistic Image Synthesis
• Prb-GAN: A Probabilistic Framework for GAN Modelling
• ReshapeGAN: Object Reshaping by Providing A Single Reference Image
• Unaligned Image-to-Image Translation by Learning to Reweight