Stochastic Block-ADMM for Training Deep Networks

TL;DR

This paper introduces Stochastic Block-ADMM, a novel method for training deep neural networks by splitting them into blocks and using auxiliary variables, enabling non-differentiable constraints and improving parallelization.

Contribution

It proposes a new stochastic optimization approach that allows training deep networks with non-differentiable constraints and integrates NMF layers for feature disentangling.

Findings

Proves convergence of the method.

Demonstrates effectiveness in supervised and weakly-supervised tasks.

Alleviates vanishing gradients and enables parallel training.

Abstract

In this paper, we propose Stochastic Block-ADMM as an approach to train deep neural networks in batch and online settings. Our method works by splitting neural networks into an arbitrary number of blocks and utilizes auxiliary variables to connect these blocks while optimizing with stochastic gradient descent. This allows training deep networks with non-differentiable constraints where conventional backpropagation is not applicable. An application of this is supervised feature disentangling, where our proposed DeepFacto inserts a non-negative matrix factorization (NMF) layer into the network. Since backpropagation only needs to be performed within each block, our approach alleviates vanishing gradients and provides potentials for parallelization. We prove the convergence of our proposed method and justify its capabilities through experiments in supervised and weakly-supervised settings.