← Back to projects

Grokking

An investigation of how a neural network can shift from memorization to generalization after extended training.

Loading project report…