In this talk, I will discuss whether overfitted DNNs in adversarial training can generalize from an approximation viewpoint. We prove by construction the existence of infinitely many adversarial training classifiers on over-parameterized DNNs that obtain arbitrarily small adversarial training error (overfitting), whereas achieving good robust generalization error under certain conditions concerning the data quality, well separated, and perturbation level. This construction is optimal and thus points out the fundamental limits of DNNs under adversarial training with statistical guarantees. Part of this talk comes from our recent work.
Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint full movie can be streamed on Popcorn Time. Enjoy unlimited access to movies and series with Popcorntime online including popular titles like Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint. Join millions of people who can watch Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint online for free with Popcorntime online's full HD movie streaming service. Get unlimited, worldwide access to Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint in HD on your laptop, phone and tablet. The Popcorn Time app is available on smart TVs and set-top boxes, making it easy to watch Can overfitted deep neural networks in adversarial training generalize? – An approximation viewpoint online on your big-screen devices for the full cinema experience.
Starring: Fanghui Liu, Engineering Research Building, Room 514, Yuchen Zeng
Genres: Crime, Documentary
Share this movie:
