Skip to content

Latest commit

 

History

History
6 lines (4 loc) · 464 Bytes

200711 Adversarially-Trained Deep Nets Transfer Better.md

File metadata and controls

6 lines (4 loc) · 464 Bytes

https://arxiv.org/abs/2007.05869

Adversarially-Trained Deep Nets Transfer Better: Illustration on Image Classification (Francisco Utrera, Evan Kravitz, N. Benjamin Erichson, Rajiv Khanna, Michael W. Mahoney)

adversarial training이 된 모델이 다른 데이터셋에 더 잘 transfer되고 학습이 빠르다는 결과. 그런데 테스트 데이터셋을 패치에 가까운 데이터셋들만 선정했는지 약간 의문. #transfer #adversarial_training