Adversarial Examples Are Not Bugs, They Are Features: NeurIPS 2019 Video

Andrew Ilyas · 4K views · 2019 · 3:11

Open on YouTube

Machine learning & deep learningother

Why the open web links it

Argues that adversarial examples in machine learning are not bugs but reflect model reliance on high-level features.

Linked by 2 independent sites, 2 of them widely read in the 2026-10 crawl of the open web, including neurips.cc, nips.cc.