Scaling Properties of Deep Residual Networks

This repository replicates the experiments of the paper Scaling properties of Deep Residual Networks. This research is published at ICML 2021. This is a joint collaboration between InstaDeep and the University of Oxford.

Fully-connected experiments

Run the following command

python scaling_run.py

It will create experiment result folders under ./scaling/.

Convolutional experiments

Run the following command

python cifar_run.py

The CIFAR-10 code is inspired from kuangliu/pytorch-cifar.