Uncertainty Decomposition in Bayesian Neural Networks with Latent Variables

TL;DR

This paper introduces a method to decompose predictive uncertainty in Bayesian neural networks with latent variables into epistemic and aleatoric parts, enhancing active learning and safe reinforcement learning applications.

Contribution

It presents a natural uncertainty decomposition in BNNs with latent variables and applies it to improve active learning and risk-sensitive reinforcement learning.

Findings

Effective uncertainty decomposition for BNNs with latent variables

Improved active learning performance

Enhanced safety in reinforcement learning environments

Abstract

Bayesian neural networks (BNNs) with latent variables are probabilistic models which can automatically identify complex stochastic patterns in the data. We describe and study in these models a decomposition of predictive uncertainty into its epistemic and aleatoric components. First, we show how such a decomposition arises naturally in a Bayesian active learning scenario by following an information theoretic approach. Second, we use a similar decomposition to develop a novel risk sensitive objective for safe reinforcement learning (RL). This objective minimizes the effect of model bias in environments whose stochastic dynamics are described by BNNs with latent variables. Our experiments illustrate the usefulness of the resulting decomposition in active learning and safe RL settings.