The article discusses the importance of heteroscedastic predictive uncertainties in Bayesian Neural Networks (BNNs). The authors illustrate how both heteroscedastic aleatoric and epistemic variance can be integrated into the variances of learned BNN parameters, improving predictive performance for lightweight networks. They introduce a simple framework for sampling-free variational inference suitable for lightweight BNNs. The paper highlights the benefits of BNNs in quantifying output uncertainty, making them ideal for applications requiring competency assessments of their performance.
Publication date: 22 Feb 2024
Project Page: https://arxiv.org/abs/2402.14532v1
Paper: https://arxiv.org/pdf/2402.14532
