Hidden-Fold Networks (HFN): Random Recurrent Residuals Using Sparse Supermasks

by Ángel López García-Arias, Masanori Hashimoto, Masato Motomura, and Jaehoon Yu

Accepted at The British Machine Vision Conference (BMVC) 2021.

arXiv link

The code in this repository was originally developed from a fork of hidden-networks.

Requirements

Python 3.6 or higher. CUDA 11.0.
For library requirements see requirements.txt.

Usage

Experiment configurations are defined in yaml files under configs/.
These configurations can be overridden by setting parameters directly through arguments (see args.py).

Example 1: HFN-ResNet50, top-k%=30%, CIFAR100

python3 main.py --config configs/CIFAR100/HFN_ResNet50_3_4.yaml --multigpu 0

Example 2: HFN-ResNet50, top-k%=50%, CIFAR100, 100 epochs

python3 main.py --config configs/CIFAR100/HFN_ResNet50_3_4.yaml --multigpu 0 --top_k=0.5 --epochs 100