Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors
Abstract
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.
Cite
Text
Ghosh et al. "Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors." International Conference on Machine Learning, 2018.Markdown
[Ghosh et al. "Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors." International Conference on Machine Learning, 2018.](https://mlanthology.org/icml/2018/ghosh2018icml-structured/)BibTeX
@inproceedings{ghosh2018icml-structured,
title = {{Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors}},
author = {Ghosh, Soumya and Yao, Jiayu and Doshi-Velez, Finale},
booktitle = {International Conference on Machine Learning},
year = {2018},
pages = {1744-1753},
volume = {80},
url = {https://mlanthology.org/icml/2018/ghosh2018icml-structured/}
}