A PyTorch implementation of this paper.
I'm currently in the process of updating this to work with the latest version of PyTorch! Currently the only network type that works is ResNet - other networks coming soon.
To run, try:
python train.py --model='resnet34' --checkpoint='resnet34'
python prune.py --model='resnet34' --checkpoint='resnet34'
Given a family of ResNets, we can construct a Pareto frontier of the tradeoff between accuracy and number of parameters:
Han et al. posit that we can beat this Pareto frontier by leaving network structures fixed, but removing individual parameters: