In this work, we employ ideas from metric learning based on deep neural features and from recent advances that augment neural networks with external memories. Our framework learns a network that maps a small labelled support set and an unlabelled example to its label, obviating the need for fine-tuning to adapt to new class types.
Memos
distance 기반으로 하는 learning들을 모티브 삼음 (1997, 2004 논문 등 전통 논문들...)
Local Weighted Learning (1997)
Neighbourhood component analysis (2004)
목적은, parametric / non-parametric 방법들로부터의 장점을 모두 활용하는 것.
저자의 관점: parametric 방법은 천천히 sample들로부터 parameter 업데이트. / non-parametric은 빠르게 새로운 샘플에 adapt.
contribution: define model & training criterion for one-shot learning / benchmark task 제시
2016 paper(arXiv): https://arxiv.org/pdf/1606.04080.pdf from Google DeepMind
Short
episode learning 방식을 제안한 논문
Key Idea (from Abstract)
In this work, we employ ideas from metric learning based on deep neural features and from recent advances that augment neural networks with external memories. Our framework learns a network that maps a small labelled support set and an unlabelled example to its label, obviating the need for fine-tuning to adapt to new class types.
Memos