Sparsity-Inducing Categorical Prior Improves Robustness of the Information Bottleneck

Publication
Proceedings of the 26th International Conference on Artificial Intelligence and Statistics