[TNNLS2023] This code is the implementation of ASTER

Adaptive Filter Pruning via Sensitivity Feedback

Author: Yuyao Zhang and Nikolaos M. Freris

The paper is accepted at the IEEE Transactions of Neural Networks and Learning Systems (TNNLS), 2023

Usage

First, clone the repository locally:

git clone https://github.com/JuttaZhang/ASTER.git

Then, install PyTorch == 1.9.1+cu111\ prchvision == 0.10.1+cu111 \ ax-platform == 0.2.4 (Please check Ax)\ tensorboardx == 2.4\ tensorboard_logger == 0.1.0:

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install ax-platform == 0.2.4
pip install tensorboardx == 2.4
pip install tensorboard_logger == 0.1.0

For example:

Using Bayesian Optimization, VGG-16 and CIFAR-10, run:

python BayesianMain.py --model vgg16 --depth 16 --s 1e-4 --exp_flops 0.4 --batch_size 64 --test-batch-size 128 --epochs 320 --pec 0.5 --lb 0.7 --ub 1 

For ResNets, run:

# using ResNet-56 and CIFAR-100 as an example
python BayesianMaincifar100Res.py --model resnet56 --depth 56 --s 1e-4 --batch_size 64 --test-batch-size 128 --epochs 320 --pec 0.5 --lb 0.1 --ub 1 

Publications

If you find this repository useful for your research, please cite the following paper:

@article{zhang2023adaptive,
  title={Adaptive Filter Pruning via Sensitivity Feedback},
  author={Zhang, Yuyao and Freris, Nikolaos M},
  journal={IEEE Transactions on Neural Networks and Learning Systems},
  year={2023},
  publisher={IEEE}
}