/BSS_distillation

Knowledge Distillation with Adversarial Samples Supporting Decision Boundary (AAAI 2019)

Primary LanguagePythonMIT LicenseMIT

Knowledge Distillation with Adversarial Samples Supporting Decision Boundary

Official Pytorch implementation of paper:

Knowledge Distillation with Adversarial Samples Supporting Decision Boundary (AAAI 2019).

Sporlight and poster are available on homepage

Environment

Python 3.6, Pytorch 0.4.1, Torchvision

Knowledge distillation (CIFAR-10)

python train_BSS_distillation.py

Distillation from ResNet 26 (teacher) to ResNet 10 (student) on CIFAR-10 dataset.

Pre-trained teacher network (ResNet 26) is included.

Citation

@inproceedings{BSSdistill,
	title = {Knowledge Distillation with Adversarial Samples Supporting Decision Boundary},
	author = {Byeongho Heo, Minsik Lee, Sangdoo Yun, Jin Young Choi},
	booktitle = {AAAI Conference on Artificial Intelligence (AAAI)},
	year = {2019}
}