Model and Feature Diversity for Bayesian Neural Networks in Mutual Learning

TL;DR

This paper introduces a deep mutual learning approach for Bayesian Neural Networks that increases model and feature diversity, leading to improved classification accuracy and better uncertainty calibration.

Contribution

It proposes a novel method to enhance BNN performance by promoting diversity in parameters and features during mutual learning, which is a new approach in this context.

Findings

Significant improvements in classification accuracy.

Reduced negative log-likelihood.

Lower expected calibration error.

Abstract

Bayesian Neural Networks (BNNs) offer probability distributions for model parameters, enabling uncertainty quantification in predictions. However, they often underperform compared to deterministic neural networks. Utilizing mutual learning can effectively enhance the performance of peer BNNs. In this paper, we propose a novel approach to improve BNNs performance through deep mutual learning. The proposed approaches aim to increase diversity in both network parameter distributions and feature distributions, promoting peer networks to acquire distinct features that capture different characteristics of the input, which enhances the effectiveness of mutual learning. Experimental results demonstrate significant improvements in the classification accuracy, negative log-likelihood, and expected calibration error when compared to traditional mutual learning for BNNs.