Alternating Back-Propagation for Generator Network

TL;DR

This paper introduces an alternating back-propagation algorithm for training generator networks, which are nonlinear factor analysis models using convolutional neural networks to generate natural images, videos, and sounds.

Contribution

It presents a novel iterative learning algorithm that infers latent factors and updates network parameters using back-propagation, applicable to complex data and incomplete training sets.

Findings

Successfully learned realistic generators for images, videos, and sounds

Can handle incomplete or indirect training data

Shares code for inference and learning steps

Abstract

This paper proposes an alternating back-propagation algorithm for learning the generator network model. The model is a non-linear generalization of factor analysis. In this model, the mapping from the continuous latent factors to the observed signal is parametrized by a convolutional neural network. The alternating back-propagation algorithm iterates the following two steps: (1) Inferential back-propagation, which infers the latent factors by Langevin dynamics or gradient descent. (2) Learning back-propagation, which updates the parameters given the inferred latent factors by gradient descent. The gradient computations in both steps are powered by back-propagation, and they share most of their code in common. We show that the alternating back-propagation algorithm can learn realistic generator models of natural images, video sequences, and sounds. Moreover, it can also be used to learn from incomplete or indirect training data.