Discriminator Contrastive Divergence: Semi-Amortized Generative Modeling by Exploring Energy of the Discriminator

By Yuxuan Song, Qiwei Ye, Minkai Xu, Tie-Yan Liu. 2020

8 citations recorded by the source corpus.

Generative Adversarial Networks (GANs) have shown great promise in modeling high dimensional data. The learning objective of GANs usually minimizes some measure discrepancy, \textit{e.g.}, $f$-divergence~($f$-GANs) or Integral Probability Metric~(Wasserstein GANs). With $f$-divergence as the objective function, the discriminator essentially estimates the density ratio, and the estimated ratio proves useful in further improving the sample quality of the generator. However, how to leverage the information contained in the discriminator of Wasserstein GANs (WGAN) is less explored. In this paper, we introduce the Discriminator Contrastive Divergence, which is well motivated by the property of WGAN's discriminator and the relationship between WGAN and energy-based model. Compared to standard GANs, where the generator is directly utilized to obtain new samples, our method proposes a semi-amortized generation procedure where the samples are produced with the generator's output as an initial state. Then several steps of Langevin dynamics are conducted using the gradient of the discriminator. We demonstrate the benefits of significant improved generation on both synthetic data and several real-world image generation benchmarks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Statistical and Nonlinear Physics, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Statistical and Nonlinear Physics

In Generative Models

Citations in the corpus

VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models (2020, background): “Zhao et al. (2017); Che et al. (2020); Song et al. (2020); Arbel et al. (2020) formulate EBMs with GANs, and use the discriminator to assign an energy.”
Discriminator optimal transport (2019, background): “In our experiments, DOT works only when ||x− y||2 is small enough for given y, and it may suggest DOT approximates their solution, however, note that it is not evident whether our practical gradient-based implementation realizes it as pointed out in [27].”
Learning Neural Generative Dynamics for Molecular Conformation Generation (2021, methodology): “Sampling from Eφ can be done with a variety of methods such as Markov chain Monte Carlo (MCMC) or Gibbs sampling (Hinton & Salakhutdinov, 2006), possibly accelerated using Langevin dynamics (Du & Mordatch, 2019; Song et al., 2020), which leverages the gradient of the EBM to conduct sampling: xk = xk”