This repository is the official implementation of Paraphrasing Complex Network: Network Compression via Factor Transfer (FT). The source code is for the experiment of ResNet on CIFAR-10. In this experiment, we use ResNet56 as a teacher network and ResNet20 as a student network on CIFAR-10 Dataset.
Before training the student network, pre-trained teacher network and paraphraser are needed. More details are in the paper. We published FT in NeurIPS 2018. See our paper here.
To install requirements using environment.yml refer to the documentation.
train_student.py is the code for training student with pretrained teacher and paraphraser using Factor Transfer (FT). To train the model(s), run this command:
python train_student.py --cu_num 0
Please refer to the following citation if this work is useful for your research.
@inproceedings{kim2018paraphrasing,
title={Paraphrasing complex network: network compression via factor transfer},
author={Kim, Jangho and Park, SeongUk and Kwak, Nojun},
booktitle={Proceedings of the 32nd International Conference on Neural Information Processing Systems},
pages={2765--2774},
year={2018}
}