The k-tied Normal Distribution: A Compact Parameterization of Gaussian Mean Field Posteriors in Bayesian Neural Networks
This work addresses the challenge of parameter efficiency in variational Bayesian inference for deep learning practitioners, though it is incremental as it builds on existing Gaussian mean-field methods.
The authors found that posterior standard deviations in Bayesian neural networks trained with Gaussian mean-field variational inference exhibit strong low-rank structure, enabling a more compact parameterization without performance loss, and this approach improves stochastic gradient signal-to-noise ratio for faster convergence.
Variational Bayesian Inference is a popular methodology for approximating posterior distributions over Bayesian neural network weights. Recent work developing this class of methods has explored ever richer parameterizations of the approximate posterior in the hope of improving performance. In contrast, here we share a curious experimental finding that suggests instead restricting the variational distribution to a more compact parameterization. For a variety of deep Bayesian neural networks trained using Gaussian mean-field variational inference, we find that the posterior standard deviations consistently exhibit strong low-rank structure after convergence. This means that by decomposing these variational parameters into a low-rank factorization, we can make our variational approximation more compact without decreasing the models' performance. Furthermore, we find that such factorized parameterizations improve the signal-to-noise ratio of stochastic gradient estimates of the variational lower bound, resulting in faster convergence.