VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning

Abstract

Deep generative models provide powerful tools for distributions over complicated manifolds, such as those of natural images. But many of these methods, including generative adversarial networks (GANs), can be difficult to train, in part because they are prone to mode collapse, which means that they characterize only a few modes of the true distribution. To address this, we introduce VEEGAN, which features a reconstructor network, reversing the action of the generator by mapping from data to noise. Our training objective retains the original asymptotic consistency guarantee of GANs, and can be interpreted as a novel autoencoder loss over the noise. In sharp contrast to a traditional autoencoder over data points, VEEGAN does not require specifying a loss function over the data, but rather only over the representations, which are standard normal by assumption. On an extensive set of synthetic and real world image datasets, VEEGAN indeed resists mode collapsing to a far greater extent than other recent GAN variants, and produces more realistic samples.

Published as a conference paper at NIPS, 2017

References in corpus (10)

• Auto-Encoding Variational Bayes
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
• Improved Techniques for Training GANs
• Adversarial Feature Learning
• Adversarially Learned Inference
• Mode Regularized Generative Adversarial Networks
• Amortised MAP Inference for Image Super-resolution
• Likelihood-free inference via classification

Cited by in corpus (101)

• ICLabel: An automated electroencephalographic independent component classifier, dataset, and website
• How Generative Adversarial Networks and Their Variants Work: An Overview
• Multimodal Unsupervised Image-to-Image Translation
• BAGAN: Data Augmentation with Balancing GAN
• Multi-Modal Self-Supervised Learning for Recommendation
• Using GANs for Sharing Networked Time Series Data: Challenges, Initial Promise, and Open Questions
• Scale- and Context-Aware Convolutional Non-intrusive Load Monitoring
• Fast and Accurate Optical Fiber Channel Modeling Using Generative Adversarial Network
• Enhanced Balancing GAN: Minority-class Image Generation
• Modeling Tabular data using Conditional GAN
• Learning Predictive Representations for Deformable Objects Using Contrastive Estimation
• Lifelong Teacher-Student Network Learning
• Machine Learning for Physics-Informed Generation of Dispersed Multiphase Flow Using Generative Adversarial Networks
• IB-GAN: Disentangled Representation Learning with Information Bottleneck Generative Adversarial Networks
• Distribution Matching in Variational Inference
• Generative Image Inpainting with Segmentation Confusion Adversarial Training and Contrastive Learning
• Exploration and Regularization of the Latent Action Space in Recommendation
• BourGAN: Generative Networks with Metric Embeddings
• Generative Adversarial Network Training is a Continual Learning Problem
• Approximability of Discriminators Implies Diversity in GANs
• DiverGAN: An Efficient and Effective Single-Stage Framework for Diverse Text-to-Image Generation
• MGGAN: Solving Mode Collapse using Manifold Guided Training
• High signal-to-noise ratio reconstruction of low bit-depth optical coherence tomography using deep learning
• Adversarial score matching and improved sampling for image generation
• Deep Ensemble Bayesian Active Learning : Addressing the Mode Collapse issue in Monte Carlo dropout via Ensembles
• Audio inpainting with generative adversarial network
• Mode Collapse and Regularity of Optimal Transportation Maps
• Investigating Cosmological GAN Emulators Using Latent Space Interpolation
• Responsible Disclosure of Generative Models Using Scalable Fingerprinting
• Generative Latent Flow
• Closed-Loop Data Transcription to an LDR via Minimaxing Rate Reduction
• PTP: Parallelized Tracking and Prediction with Graph Neural Networks and Diversity Sampling
• Imperfect ImaGANation: Implications of GANs Exacerbating Biases on Facial Data Augmentation and Snapchat Selfie Lenses
• Generative Adversarial Networks (GANs): What it can generate and What it cannot?
• CMB-GAN: Fast Simulations of Cosmic Microwave background anisotropy maps using Deep Learning
• Self-Diagnosing GAN: Diagnosing Underrepresented Samples in Generative Adversarial Networks
• DLow: Diversifying Latent Flows for Diverse Human Motion Prediction
• Effect of The Latent Structure on Clustering with GANs
• Hierarchical Patch VAE-GAN: Generating Diverse Videos from a Single Sample
• 3D Brain and Heart Volume Generative Models: A Survey
• Deception for Cyber Defence: Challenges and Opportunities
• Private Post-GAN Boosting
• Learning disconnected manifolds: a no GANs land
• Autoencoder Image Interpolation by Shaping the Latent Space
• Training Generative Adversarial Networks via Primal-Dual Subgradient Methods: A Lagrangian Perspective on GAN
• Continuous Conditional Generative Adversarial Networks: Novel Empirical Losses and Label Input Mechanisms
• GDPP: Learning Diverse Generations Using Determinantal Point Process
• MCL-GAN: Generative Adversarial Networks with Multiple Specialized Discriminators
• Neural Architecture Search for Generative Adversarial Networks: A Comprehensive Review and Critical Analysis
• Self-Supervised GANs with Label Augmentation
• VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models
• Style Transfer with Time Series: Generating Synthetic Financial Data
• Discriminator Feature-based Inference by Recycling the Discriminator of GANs
• Judge the Judges: A Large-Scale Evaluation Study of Neural Language Models for Online Review Generation
• Cycle-Consistent Adversarial Learning as Approximate Bayesian Inference
• Out-of-Sample Extrapolation with Neuron Editing
• First Order Generative Adversarial Networks
• Bidirectional Generative Modeling Using Adversarial Gradient Estimation
• Towards Better Long-range Time Series Forecasting using Generative Adversarial Networks
• Dual Contrastive Loss and Attention for GANs
• Learning Implicit Generative Models by Teaching Explicit Ones
• Sparse Flows: Pruning Continuous-depth Models
• Generative Ratio Matching Networks
• Semi-supervised Conditional GAN for Simultaneous Generation and Detection of Phishing URLs: A Game theoretic Perspective
• Improving the Reconstruction of Disentangled Representation Learners via Multi-Stage Modeling
• Multimodal Image-to-Image Translation via Mutual Information Estimation and Maximization
• Inclusive GAN: Improving Data and Minority Coverage in Generative Models
• Improved BiGAN training with marginal likelihood equalization
• Mode matching in GANs through latent space learning and inversion
• REDS: Rule Extraction for Discovering Scenarios
• Ensembles of GANs for synthetic training data generation
• Tempered Adversarial Networks
• Diverse Image Generation via Self-Conditioned GANs
• not-so-BigGAN: Generating High-Fidelity Images on Small Compute with Wavelet-based Super-Resolution
• Semi-Implicit Generative Model
• AdvNF: Reducing Mode Collapse in Conditional Normalising Flows using Adversarial Learning
• Combining AI Control Systems and Human Decision Support via Robustness and Criticality
• Collaborative Sampling in Generative Adversarial Networks
• Exploiting Spatial Dimensions of Latent in GAN for Real-time Image Editing
• Adversarial sampling of unknown and high-dimensional conditional distributions
• HRVGAN: High Resolution Video Generation using Spatio-Temporal GAN
• Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution
• Conditioning Trick for Training Stable GANs
• MaskAAE: Latent space optimization for Adversarial Auto-Encoders
• Hierarchy of GANs for learning embodied self-awareness model
• Exploiting Chain Rule and Bayes' Theorem to Compare Probability Distributions
• IID-GAN: an IID Sampling Perspective for Regularizing Mode Collapse
• SetGAN: Improving the stability and diversity of generative models through a permutation invariant architecture
• Normalized Diversification
• Learning to infer in recurrent biological networks
• Toward Privacy and Utility Preserving Image Representation
• Escaping from Collapsing Modes in a Constrained Space
• High Diversity Attribute Guided Face Generation with GANs
• Quantifying point cloud realism through adversarially learned latent representations
• HydraGAN A Multi-head, Multi-objective Approach to Synthetic Data Generation
• Forward Operator Estimation in Generative Models with Kernel Transfer Operators
• LiRA: Light-Robust Adversary for Model-based Reinforcement Learning in Real World
• Coulomb Autoencoders
• Generative Adversarial Learning via Kernel Density Discrimination
• One-shot domain adaptation for semantic face editing of real world images using StyleALAE
• Variational Hetero-Encoder Randomized GANs for Joint Image-Text Modeling