Semi-Implicit Generative Model
This addresses the mode collapse and training instability issues in generative adversarial networks (GANs) for researchers and practitioners in generative modeling.
The paper tackles the problem of combining explicit and implicit generative models by introducing a semi-implicit generator (SIG) that can be trained with maximum likelihood. The result shows SIG generates high-quality multi-modal samples and, when used as a regularizer for GANs, stabilizes training, resists mode collapse, and improves sample diversity.
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.