Peter Chen

  • About
  • iOS
  • ML
  • Design

Andrew Karpathy's Recipe for Training Neural Networks

Apr 29, 2019

Inspect and understand the data Setup and verify each component of full end-to-end training with a simple model to verify data, model, training flow is working as expected Use metric from previous end-to-end training as baseline Overfit with a large model Refine model, regularize, etc Squeeze last few bits of performance

https://karpathy.github.io/2019/04/25/recipe/