This paper studies the geometrical properties of the representations learned by neural networks. Published at NeurIPS 2019.

This paper studies the geometrical properties of the representations learned by neural networks. Published at NeurIPS 2019.
A study on spurious correlations which is induced by a causal pespective of stress tests, published at NeurIPS 2021.
This is an inspiring work on ICLR 2020 about how to realize the potential of generative models on downstream discriminative problems.