Train ImageNet without Hyperparameters with Automatic Gradient Descent

Towards architecture-aware optimisation

TL;DR We’ve derived an optimiser called automatic gradient descent (AGD) that can train ImageNet without hyperparameters. This removes the need for expensive and time-consuming learning rate tuning, selection of learning rate decay schedulers, etc. Our paper can be found here.

I worked on this project with Jeremy Bernstein, Kevin Huang, Navid Azizan and Yisong Yue. See Jeremy’s GitHub for a clean Pytorch implementation, or my GitHub for an experimental version with more features. …

gradient hyperparameter-tuning imagenet machine learning optimization-algorithms python thoughts-and-theory