Variational Generative Stochastic Networks with Collaborative Shaping

TL;DR

This paper introduces a novel training method for generative models that unrolls a variational auto-encoder into a Markov chain, shaping its trajectories to better reproduce target distributions, with empirical results demonstrating state-of-the-art performance.

Contribution

It proposes a new approach combining variational auto-encoders and Markov chains with trajectory shaping and regularization, advancing generative modeling techniques.

Findings

Achieves state-of-the-art results on MNIST and TFD datasets.

Provides a new framework for training generative models with trajectory shaping.

Demonstrates improved qualitative and quantitative performance.

Abstract

We develop an approach to training generative models based on unrolling a variational auto-encoder into a Markov chain, and shaping the chain's trajectories using a technique inspired by recent work in Approximate Bayesian computation. We show that the global minimizer of the resulting objective is achieved when the generative model reproduces the target distribution. To allow finer control over the behavior of the models, we add a regularization term inspired by techniques used for regularizing certain types of policy search in reinforcement learning. We present empirical results on the MNIST and TFD datasets which show that our approach offers state-of-the-art performance, both quantitatively and from a qualitative point of view.