Smile meeting on "double descent", October 17, 3pm at Inria (2, rue Simone Iff)

[Francis Bach]

Hi!

The next “Smile in Paris”  meeting will take place on October 17th at Inria (2, rue Simone Iff, Paris 12) between 3pm and 4.30pm. Going back to the origins of the group, I will present recent advances in learning theory, namely the “double descent” phenomenon, which is claimed to explain some of the recent successes of deep learning (references below).

So if you want to learn more about this phenomenon, to make your own opinion about its generality and relevance to deep learning, please attend!

In order to enter the Inria building, you need to register here first (simply put first name / last name): https://forms.gle/H7VTjNfnLGFKQpYf7

Best regards

Francis

[A] Reconciling modern machine learning practice and the bias-variance trade-off. Mikhail Belkin, Daniel Hsu, Siyuan Ma, Soumik Mandal. https://arxiv.org/pdf/1812.11118

[B] Two models of double descent for weak features. Mikhail Belkin, Daniel Hsu, Ji Xu. https://arxiv.org/pdf/1903.07571

[C] The generalization error of random features regression: Precise asymptotics and double descent curve. Song Mei, Andrea Montanari. https://arxiv.org/pdf/1908.05355

[Robert Gower]

Hi all,

just to remind you, tomorrow Francis Bach will give a "Smile in Paris” seminar in the Jacques Louis Lions room on the ground floor at Inria (2, rue Simone Iff, Paris 12) between 3pm and 4.30pm. Come if you want to find out exactly what is this "double descent" phenomenon behind some recent successes of deep learning. Details below.