Gated Information Bottleneck for Generalization in Sequential Environments

Francesco Alesiani, Shujian Yu, Xi Yu

Deep neural networks suffer from poor generalization to unseen environments when the underlying data distribution is different from that in the training set. By learning minimum su…