HEMP

HEMP: High order Entropy Minimizationfor neural network comPression

Please cite this work as

@article{TARTAGLIONE2021,
title = {HEMP: High-order Entropy Minimization for neural network comPression},
journal = {Neurocomputing},
year = {2021},
issn = {0925-2312},
doi = {https://doi.org/10.1016/j.neucom.2021.07.022},
url = {https://www.sciencedirect.com/science/article/pii/S0925231221010663},
author = {Enzo Tartaglione and Stephane Lathuiliere and Attilio Fiandrotti and Marco Cagnazzo and Marco Grangetto}
}

Requirements

• PyTorch >= 1.8.1
• CUDA >= 11.1
• scipy >= 1.5.4
• numpy >= 1.20.2
• torchvision >= 0.9.1
• py7zr >= 0.16.0
• matplotlib >= 3.2.2
• tqdm >= 4.56.0

Running code

python3 main.py \
-model [architecture] \
-dataset [training dataset] \
-device [cuda:id or cpu] \
-batch_size [batch size for training] \
-test-batch-size [batch size for test] \
-epochs [wall epochs]\
-lr [learning rate] \
-lamb_H [weight on HEMP] \
-lamb_RMSE [weight on RMSE term] \
-entropy_order [entropy order to be evaluated] \
-N [number of bins]