Model and Feature Diversity for Bayesian Neural Networks in Mutual Learning

Van Cuong Pham · Cuong C Nguyen · Cuong Nguyen · Trung Le · Dinh Phung · Gustavo Carneiro · Thanh-Toan Do

Great Hall & Hall B1+B2 (level 1) #1208
[ ]
Tue 12 Dec 8:45 a.m. PST — 10:45 a.m. PST


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.

Chat is not available.