Principled Pruning of Bayesian Neural Networks through Variational Free Energy Minimization
This work addresses pruning challenges for Bayesian neural networks in signal processing, offering a principled approach with a clear stopping criterion, though it appears incremental as it builds on existing Bayesian model reduction techniques.
The paper tackles the problem of pruning Bayesian neural networks by applying Bayesian model reduction with variational free energy minimization, resulting in a novel iterative pruning algorithm that shows better model performance compared to state-of-the-art methods on UCI datasets.
Bayesian model reduction provides an efficient approach for comparing the performance of all nested sub-models of a model, without re-evaluating any of these sub-models. Until now, Bayesian model reduction has been applied mainly in the computational neuroscience community on simple models. In this paper, we formulate and apply Bayesian model reduction to perform principled pruning of Bayesian neural networks, based on variational free energy minimization. Direct application of Bayesian model reduction, however, gives rise to approximation errors. Therefore, a novel iterative pruning algorithm is presented to alleviate the problems arising with naive Bayesian model reduction, as supported experimentally on the publicly available UCI datasets for different inference algorithms. This novel parameter pruning scheme solves the shortcomings of current state-of-the-art pruning methods that are used by the signal processing community. The proposed approach has a clear stopping criterion and minimizes the same objective that is used during training. Next to these benefits, our experiments indicate better model performance in comparison to state-of-the-art pruning schemes.