Drs. Atul Prakash (Computer Science and Engineering) and Huei Peng (Mechanical Engineering) aim to improve the robustness of machine learning models by taking advantage of environmental context, particular laws of physics and semantic background information. This new approach to machine learning can lead to reduced training needs and more robust learning even with significant perturbations.
Efficient Adversarial Training with Transferable Adversarial Examples
(CVPR 2020 publication on a fast adversarial training method called ATTA)
Code release for ATTA
This team has also received an NSF grant.