Semi-Implicit Generative Model
arXiv:1905.12659
Abstract
To combine explicit and implicit generative models, we introduce semi-implicit generator (SIG) as a flexible hierarchical model that can be trained in the maximum likelihood framework. Both theoretically and experimentally, we demonstrate that SIG can generate high quality samples especially when dealing with multi-modality. By introducing SIG as an unbiased regularizer to the generative adversarial network (GAN), we show the interplay between maximum likelihood and adversarial learning can stabilize the adversarial training, resist the notorious mode collapsing problem of GANs, and improve the diversity of generated random samples.
arXiv commentThird workshop on Bayesian Deep Learning (NeurIPS 2018), Montreal, Canada
References in corpus (20)
- Auto-Encoding Variational Bayes
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- Pixel Recurrent Neural Networks
- NIPS 2016 Tutorial: Generative Adversarial Networks
- Density estimation using Real NVP
- Adversarial Feature Learning
- Adversarially Learned Inference
- VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning
- Unrolled Generative Adversarial Networks
- Mode Regularized Generative Adversarial Networks
- A Note on the Inception Score
- Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
- AdaGAN: Boosting Generative Models
- On distinguishability criteria for estimating generative models
- Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models
- On the Discrimination-Generalization Tradeoff in GANs
- Global optimality conditions for deep neural networks
- Improved generator objectives for GANs
- On the Limitations of First-Order Approximation in GAN Dynamics
- Theoretical limitations of Encoder-Decoder GAN architectures