A Framework for Variational Inference of Lightweight Bayesian Neural Networks with Heteroscedastic Uncertainties
This work addresses the need for efficient uncertainty estimation in lightweight BNNs for applications requiring heteroscedastic uncertainties, presenting an incremental improvement over existing methods.
The paper tackles the problem of obtaining heteroscedastic predictive uncertainties from Bayesian Neural Networks (BNNs) without adding extra parameters, by embedding both aleatoric and epistemic variances into learned BNN parameters, which improves predictive performance for lightweight networks.
Obtaining heteroscedastic predictive uncertainties from a Bayesian Neural Network (BNN) is vital to many applications. Often, heteroscedastic aleatoric uncertainties are learned as outputs of the BNN in addition to the predictive means, however doing so may necessitate adding more learnable parameters to the network. In this work, we demonstrate that both the heteroscedastic aleatoric and epistemic variance can be embedded into the variances of learned BNN parameters, improving predictive performance for lightweight networks. By complementing this approach with a moment propagation approach to inference, we introduce a relatively simple framework for sampling-free variational inference suitable for lightweight BNNs.