TabularBench: Benchmarking Adversarial Robustness for Tabular Deep Learning in Real-world Use-cases

We evaluated adversarial robustness with CAA, an ensemble of gradient and search attacks which was recently demonstrated as the most effective attack against a tabular model.