A Bayesian Approach to Invariant Deep Neural Networks
Machine Learning
2021-11-04 v2 Machine Learning
Abstract
We propose a novel Bayesian neural network architecture that can learn invariances from data alone by inferring a posterior distribution over different weight-sharing schemes. We show that our model outperforms other non-invariant architectures, when trained on datasets that contain specific invariances. The same holds true when no data augmentation is performed.
Cite
@article{arxiv.2107.09301,
title = {A Bayesian Approach to Invariant Deep Neural Networks},
author = {Nikolaos Mourdoukoutas and Marco Federici and Georges Pantalos and Mark van der Wilk and Vincent Fortuin},
journal= {arXiv preprint arXiv:2107.09301},
year = {2021}
}
Comments
8 pages, 3 figures, To be published in ICML UDL 2021