A Theory of Non-Linear Feature Learning with One Gradient Step in Two-Layer Neural Networks.

Behrad Moniri, Donghwan Lee, Hamed Hassani, Edgar Dobriban

Browse the full ICML paper archive.