Skip to content

Latest commit

 

History

History
8 lines (5 loc) · 604 Bytes

200325 Rethinking Few-Shot Image Classification.md

File metadata and controls

8 lines (5 loc) · 604 Bytes

https://arxiv.org/abs/2003.11539

Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (Yonglong Tian, Yue Wang, Dilip Krishnan, Joshua B. Tenenbaum, Phillip Isola)

메타러닝 문제(few shot learning)를 메타트레이닝셋에 이미지 분류기를 학습시키고 그 학습된 모델에서 추출한 임베딩으로 few shot 학습을 시키는 것으로 풀기. 베이스라인에 가까운 방법인데 기존의 복잡한 방법들보다 성능이 더 낫다...요즘 few shot learning과 metric learning에 폭탄이 떨어지고 있는 듯.

#meta_learning #review