Скачать или смотреть Lecture 4: Spectral signatures and efficient certifiability

Lecture notes available here: https://jerryzli.github.io/robust-ml-...

The main lemma has appeared in many very similar forms in the literature. A similar form was first proved in [DKKLMS17], which was then refined in a long series of papers. To my knowledge the first time this exact form appeared was in [DHL19], a paper we will cover later on, but similar forms appear earlier as well. See [DK19] for a more extensive literature review.

References:

[TLM18] Tran, Brandon, Jerry Li, and Aleksander Madry. "Spectral signatures in backdoor attacks." Advances in Neural Information Processing Systems. 2018.

[DKKLMS17] Diakonikolas, Ilias, et al. "Being robust (in high dimensions) can be practical." Proceedings of the 34th International Conference on Machine Learning-Volume 70. JMLR. org, 2017.

[DK19] Diakonikolas, Ilias, and Daniel M. Kane. "Recent Advances in Algorithmic High-Dimensional Robust Statistics." arXiv preprint arXiv:1911.05911 (2019).

[DHL19] Dong, Yihe, Samuel Hopkins, and Jerry Li. "Quantum entropy scoring for fast robust mean estimation and improved outlier detection." Advances in Neural Information Processing Systems. 2019.