Communication-Efficient stochastic gradient mcmc for neural networks

Conference Paper

Learning probability distributions on the weights of neural networks has recently proven beneficial in many applications. Bayesian methods such as Stochastic Gradient Markov Chain Monte Carlo (SG-MCMC) offer an elegant framework to reason about model uncertainty in neural networks. However, these advantages usually come with a high computational cost. We propose accelerating SG-MCMC under the master-worker framework: workers asynchronously and in parallel share responsibility for gradient computations, while the master collects the final samples. To reduce communication overhead, two protocols (downpour and elastic) are developed to allow periodic interaction between the master and workers. We provide a theoretical analysis on the finite-time estimation consistency of posterior expectations, and establish connections to sample thinning. Our experiments on various neural networks demonstrate that the proposed algorithms can greatly reduce training time while achieving comparable (or better) test accuracy/log-likelihood levels, relative to traditional SG-MCMC. When applied to reinforcement learning, it naturally provides exploration for asynchronous policy optimization, with encouraging performance improvement.

Duke Authors

Cited Authors

  • Li, C; Chen, C; Pu, Y; Henao, R; Carin, L

Published Date

  • January 1, 2019

Published In

  • 33rd Aaai Conference on Artificial Intelligence, Aaai 2019, 31st Innovative Applications of Artificial Intelligence Conference, Iaai 2019 and the 9th Aaai Symposium on Educational Advances in Artificial Intelligence, Eaai 2019

Start / End Page

  • 4173 - 4180

International Standard Book Number 13 (ISBN-13)

  • 9781577358091

Citation Source

  • Scopus