Abstract: Deep learning revolutionized computer vision, speech recognition, natural language understanding, and more. However, from the theoretical perspective we know that training neural networks is computationally hard and one can construct distributions on which deep learning fails. The talk will propose several parameters of distributions that can move them from being easy-to-train to being hard-to-train.
Shai Shalev-Shwartz (Mobileye &amp; HUJI)
More from the Same Authors
2022 Poster: Knowledge Distillation: Bad Models Can Be Good Role Models »
Gal Kaplun · Eran Malach · Preetum Nakkiran · Shai Shalev-Shwartz
2021 : Q&A with Shai Shalev-Shwartz »
2015 Poster: Beyond Convexity: Stochastic Quasi-Convex Optimization »
Elad Hazan · Kfir Y. Levy · Shai Shalev-Shwartz