Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors
This work addresses model selection for BNNs, which is crucial for practitioners needing well-calibrated uncertainties, but it is incremental as it builds on prior horseshoe prior methods.
The paper tackled the problem of model selection in Bayesian Neural Networks (BNNs) by using horseshoe priors to turn off unnecessary nodes, resulting in improved model compactness while maintaining predictive performance, especially in small-sample settings like reinforcement learning.
Bayesian Neural Networks (BNNs) have recently received increasing attention for their ability to provide well-calibrated posterior uncertainties. However, model selection---even choosing the number of nodes---remains an open question. Recent work has proposed the use of a horseshoe prior over node pre-activations of a Bayesian neural network, which effectively turns off nodes that do not help explain the data. In this work, we propose several modeling and inference advances that consistently improve the compactness of the model learned while maintaining predictive performance, especially in smaller-sample settings including reinforcement learning.