Improving GAN Training with Probability Ratio Clipping and Sample Reweighting
arXiv:2006.06900
Abstract
Despite success on a wide range of problems related to vision, generative adversarial networks (GANs) often suffer from inferior performance due to unstable training, especially for text generation. To solve this issue, we propose a new variational GAN training framework which enjoys superior training stability. Our approach is inspired by a connection of GANs and reinforcement learning under a variational perspective. The connection leads to (1) probability ratio clipping that regularizes generator training to prevent excessively large updates, and (2) a sample re-weighting mechanism that improves discriminator training by downplaying bad-quality fake samples. Moreover, our variational GAN framework can provably overcome the training issue in many GANs that an optimal discriminator cannot provide any informative gradient to training generator. By plugging the training approach in diverse state-of-the-art GAN architectures, we obtain significantly improved performance over a range of tasks, including text generation, text style transfer, and image generation.
arXiv commentNeurIPS 2020 camera ready version (citations updated)
References in corpus (36)
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- Large Scale GAN Training for High Fidelity Natural Image Synthesis
- Spectral Normalization for Generative Adversarial Networks
- Categorical Reparameterization with Gumbel-Softmax
- Trust Region Policy Optimization
- Conditional Image Synthesis With Auxiliary Classifier GANs
- The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
- Energy-based Generative Adversarial Network
- BEGAN: Boundary Equilibrium Generative Adversarial Networks
- Style Transfer from Non-Parallel Text by Cross-Alignment
- Reinforcement Learning and Control as Probabilistic Inference: Tutorial and Review
- MaskGAN: Better Text Generation via Filling in the______
- Maximum a Posteriori Policy Optimisation
- An empirical study on evaluation metrics of generative adversarial networks
- Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
- Style Transfer as Unsupervised Machine Translation
- The Unusual Effectiveness of Averaging in GAN Training
- On Unifying Deep Generative Models
- Deep Directed Generative Models with Energy-Based Probability Estimation
- Learning Data Manipulation for Augmentation and Weighting
- Bias Correction of Learned Generative Models using Likelihood-Free Importance Weighting
- Multiple-Attribute Text Style Transfer
- Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect
- Your GAN is Secretly an Energy-based Model and You Should use Discriminator Driven Latent Sampling
- AutoGAN: Neural Architecture Search for Generative Adversarial Networks
- Deep Generative Models with Learnable Knowledge Constraints
- Lipschitz Generative Adversarial Nets
- Style Transfer for Texts: Retrain, Report Errors, Compare with Rewrites
- Structured Content Preservation for Unsupervised Text Style Transfer
- Cold-Start Reinforcement Learning with Softmax Policy Gradient
- Smoothness and Stability in GANs
- Adversarial Lipschitz Regularization
- Residual Energy-Based Models for Text Generation
- Connecting the Dots Between MLE and RL for Sequence Prediction
- Understanding the Effectiveness of Lipschitz-Continuity in Generative Adversarial Nets