Saehyung-Lee/OAT

This repository is the official implementation of "Removing Undesirable Feature Contributions using Out-of-Distribution Data", published as a conference paper at ICLR 2021.

Removing Undesirable Feature Contributions using Out-of-Distribution Data

This repository is the official implementation of "Removing Undesirable Feature Contributions using Out-of-Distribution Data", published as a conference paper at ICLR 2021.

You can download the OOD dataset and pre-trained models on CIFAR-10 here:

• OOD dataset

• OAT+PGD

• OAT+TRADES

• OAT+RST

You can also create your own OOD datasets using the work of Carmon et al.