Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks

Name: Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks
Uploaded: 2026-08-20T17:36:03.895319+00:00
Duration: 642 s
Description: The episode discusses 'Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks,' which maps out how training a simple network behaves under different scaling conditions. Hosts explain that learning success depends critically on the relationship between the learning rate, hidden unit width, and input dimension.

The episode discusses 'Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks,' which maps out how training a simple network behaves under different scaling conditions. Hosts explain that learning success depends critically on the relationship between the learning rate, hidden unit width, and input dimension.

Open the full episode