R2D2

Meta-learning with differentiable closed-form solvers

Introduction

Name:	R2D2
Embed.:	Conv64F
Type:	Metric
Venue:	ICLR'19
Codes:	MetaOptNet

Cite this work with:

@inproceedings{DBLP:conf/iclr/BertinettoHTV19,
  author    = {Luca Bertinetto and
               Jo{\\~{a}}o F. Henriques and
               Philip H. S. Torr and
               Andrea Vedaldi},
  title     = {Meta-learning with differentiable closed-form solvers},
  booktitle = {7th International Conference on Learning Representations, {ICLR} 2019,
               New Orleans, LA, USA, May 6-9, 2019},
  year      = {2019},
  url       = {https://openreview.net/forum?id=HyxnZh0ct7}
}

---

Results and Models

Classification

	Embedding	📖 miniImageNet (5,1)	💻 miniImageNet (5,1)	📖miniImageNet (5,5)	💻 miniImageNet (5,5)	📝 Comments
1	Conv64F	-	51.19 ± 0.36 ⬇️ 📋	-	67.29 ± 0.31 ⬇️ 📋	Table.2
2	ResNet12	-	59.52 ± 0.39 ⬇️ 📋	-	74.61 ± 0.30 ⬇️ 📋	Table.2
3	ResNet18	-	58.36 ± 0.38 ⬇️ 📋	-	75.69 ± 0.29 ⬇️ 📋	Table.2

	Embedding	📖 tieredImageNet (5,1)	💻 tieredImageNet (5,1)	📖tieredImageNet (5,5)	💻 tieredImageNet (5,5)	📝 Comments
1	Conv64F	-	52.18 ± 0.40 ⬇️ 📋	-	69.19 ± 0.36 ⬇️ 📋	Table.2
2	ResNet12	-	65.07 ± 0.44 ⬇️ 📋	-	83.04 ± 0.30 ⬇️ 📋	Table.2
3	ResNet18	-	64.73 ± 0.44 ⬇️ 📋	-	83.40 ± 0.31 ⬇️ 📋	Table.2