/Filter_Pruning_Using_MR_Paths

(KSC 2020) 사전 학습된 신경망에서 최대 적합성 경로를 이용한 구조화된 프루닝 기법

Primary LanguagePython

Structured pruning using max relevance paths from a pretrained neural networks

comarison mehtods

Thinet(option : greedy) paper : ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression(ICCV,2017)

LASSO(option : lasso) paper: Channel Pruning for Accelerating Very Deep Neural Networks(ICCV,2017)

Prerequistes

pytorch, python : pytorch 1.6 ↑, python 3.7 ↑ package : numpy, os, torchsummaryX, tqdm, networkx

have to download VGG16_BN weight and then move model weights to directory './experiments/vgg16_exp_cifar100_0/checkpoints/'

https://drive.google.com/drive/folders/1D0zxgCMg3nDUGGxCt8n2PDtcU6FHVysP?usp=sharing

Experiments

VGG16-BN / CIFAR-100