A law of robustness and the importance of overparametrization in deep learning

Microsoft Research Senior Principal Researcher Sebastien Bubeck answers several questions about the NeurIPS 2021 paper, “A Universal Law of Robustness via Isoperimetry.” He discusses what overparametrization is, the law of robustness, and how these two concepts can impact the future of large deep neural network research.

Read the paper in full here: https://www.microsoft.com/en-us/research/publication/a-universal-law-of-robustness-via-isoperimetry/

Видео A law of robustness and the importance of overparametrization in deep learning канала Microsoft Research