Learning threshold neurons via the Edge of Stability

Описание к видео Learning threshold neurons via the Edge of Stability

Presentation by Kwangjun Ahn and Felipe Suarez of their work https://arxiv.org/abs/2212.07469 with S. Bubeck, S. Chewi, Y.T. Lee and Y. Zhang. This work presents an analysis of the non-convex dynamic in a toy model (the sparse coding model) that captures the essence of the emergence of edge detectors in convolutional neural networks. Surprisingly this emergence is connected to the well-documented phenomenon of instability in training neural networks (aka Edge of Stability).

Комментарии

Информация по комментариям в разработке

Похожие видео