Adversarially Robust Multi-task Representation Learning
Austin Watkins, Thanh Nguyen-Tang, Enayat Ullah, Raman Arora
Abstract
We study adversarially robust transfer learning, wherein, given labeled data on multiple (source) tasks, the goal is to train a model with small robust error on a previously unseen (target) task. In particular, we consider a multi-task representation learning (MTRL) setting, i.e., we assume that the source and target tasks admit a simple (linear) predictor on top of a shared representation (e.g., the final hidden layer of a deep neural network). In this general setting, we provide rates on~the excess adversarial (transfer) risk for Lipschitz losses and smooth nonnegative losses. These rates show that learning a representation using adversarial training on diverse tasks helps protect against inference-time attacks in data-scarce environments. Additionally, we provide novel rates for the single-task setting.
BibTeX
@inproceedings{
watkins2024adversarially,
title={Adversarially Robust Multi-task Representation Learning},
author={Austin Watkins and Thanh Nguyen-Tang and Enayat Ullah and Raman Arora},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024},
url={https://openreview.net/forum?id=w2L3Ll1jbV}
}