Code supports Python 2.7 and will later support Python 3.6.
pip install -r requirements.txt
bash datasets/cifar10.sh
bash datasets/cifar100.sh
Scripts to reproduce results are located in scripts/
. Currently, we only release an example for the inference stage ResNet18 with cifar100 using 2-bit weights and activations. And we will release the training codes when the paper is published. To reproduce the example result:
bash scripts/cifar_KD_eval.sh ${gpu_id} ResNet18 cifar100 MHGD-RKD-SVD 2 adam 0.4
The result will be shown at
results/cifar100_ResNet18_Student_2_1e-05_200_0.001_128_MHGD-RKD-SVD_adam_0.4_0_KD_eval/progress.csv
- Augmentation policy
- Quantization
- Knolwedge Distillation
@article{role-kd,
Author = {Jie Fu and Xue Geng and Zhijian Duan and Bohan Zhuang and Xingdi Yuan and Adam Trischler and Jie Lin and Chris Pal and Hao Dong},
Title = {Role-Wise Data Augmentation for Knowledge Distillation},
Year = {2020},
Eprint = {arXiv:2004.08861},
}