MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization

Yangbin Chen; Yun Ma; Tom Ko; Jianping Wang; Qing Li

doi:10.1109/ICPR48806.2021.9413158

MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization

Yangbin Chen, Yun Ma, Tom Ko, Jianping Wang, Qing Li

Department of Computer Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

6 Citations (Scopus)

Abstract

Model-Agnostic Meta-Learning (MAML) and its variants are popular few-shot classification methods. They train an initializer across a variety of sampled learning tasks (also known as episodes) such that the initialized model can adapt quickly to new ones. However, current MAML-based algorithms have limitations in forming generalizable decision boundaries. In this paper, we propose an approach called MetaMix, which generates virtual feature-target pairs within each episode to regularize the backbone models. MetaMix can be integrated with any of the MAML-based algorithms and learn the decision boundaries generalizing better to new tasks. Experiments on the mini-ImageNet, CUB, and FC100 datasets show that MetaMix improves the performance of MAML-based algorithms and achieves state-of-the-art result when integrated with MetaTransfer Learning.

Original language	English
Title of host publication	Proceedings of ICPR 2020
Subtitle of host publication	25^th International Conference on Pattern Recognition
Publisher	IEEE
Pages	407-414
ISBN (Electronic)	978-1-7281-8808-9
ISBN (Print)	978-1-7281-8809-6
DOIs	https://doi.org/10.1109/ICPR48806.2021.9413158
Publication status	Published - Jan 2021
Event	25th International Conference on Pattern Recognition (ICPR2020) - Virtual, Milan, Italy Duration: 10 Jan 2021 → 15 Jan 2021 https://www.micc.unifi.it/icpr2020/

Publication series

Name	Proceedings - International Conference on Pattern Recognition
ISSN (Print)	1051-4651

Conference

Conference	25th International Conference on Pattern Recognition (ICPR2020)
Abbreviated title	ICPR 2020
Country/Territory	Italy
City	Milan
Period	10/01/21 → 15/01/21
Internet address	https://www.micc.unifi.it/icpr2020/

Access Output

10.1109/ICPR48806.2021.9413158

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@inproceedings{e8c9b503a98e4701bc10b53ae926b669,

title = "MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization",

abstract = "Model-Agnostic Meta-Learning (MAML) and its variants are popular few-shot classification methods. They train an initializer across a variety of sampled learning tasks (also known as episodes) such that the initialized model can adapt quickly to new ones. However, current MAML-based algorithms have limitations in forming generalizable decision boundaries. In this paper, we propose an approach called MetaMix, which generates virtual feature-target pairs within each episode to regularize the backbone models. MetaMix can be integrated with any of the MAML-based algorithms and learn the decision boundaries generalizing better to new tasks. Experiments on the mini-ImageNet, CUB, and FC100 datasets show that MetaMix improves the performance of MAML-based algorithms and achieves state-of-the-art result when integrated with MetaTransfer Learning.",

author = "Yangbin Chen and Yun Ma and Tom Ko and Jianping Wang and Qing Li",

year = "2021",

month = jan,

doi = "10.1109/ICPR48806.2021.9413158",

language = "English",

isbn = "978-1-7281-8809-6",

series = "Proceedings - International Conference on Pattern Recognition",

publisher = "IEEE",

pages = "407--414",

booktitle = "Proceedings of ICPR 2020",

address = "United States",

note = "25th International Conference on Pattern Recognition (ICPR2020), ICPR 2020 ; Conference date: 10-01-2021 Through 15-01-2021",

url = "https://www.micc.unifi.it/icpr2020/",

}

Chen, Y , Ma, Y, Ko, T, Wang, J & Li, Q 2021, MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization. in Proceedings of ICPR 2020: 25^th International Conference on Pattern Recognition., 9413158, Proceedings - International Conference on Pattern Recognition, IEEE, pp. 407-414, 25th International Conference on Pattern Recognition (ICPR2020), Milan, Italy, 10/01/21. https://doi.org/10.1109/ICPR48806.2021.9413158

MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization. / Chen, Yangbin ; Ma, Yun; Ko, Tom et al.
Proceedings of ICPR 2020: 25^th International Conference on Pattern Recognition. IEEE, 2021. p. 407-414 9413158 (Proceedings - International Conference on Pattern Recognition).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - MetaMix

T2 - 25th International Conference on Pattern Recognition (ICPR2020)

AU - Chen, Yangbin

AU - Ma, Yun

AU - Ko, Tom

AU - Wang, Jianping

AU - Li, Qing

PY - 2021/1

Y1 - 2021/1

N2 - Model-Agnostic Meta-Learning (MAML) and its variants are popular few-shot classification methods. They train an initializer across a variety of sampled learning tasks (also known as episodes) such that the initialized model can adapt quickly to new ones. However, current MAML-based algorithms have limitations in forming generalizable decision boundaries. In this paper, we propose an approach called MetaMix, which generates virtual feature-target pairs within each episode to regularize the backbone models. MetaMix can be integrated with any of the MAML-based algorithms and learn the decision boundaries generalizing better to new tasks. Experiments on the mini-ImageNet, CUB, and FC100 datasets show that MetaMix improves the performance of MAML-based algorithms and achieves state-of-the-art result when integrated with MetaTransfer Learning.

AB - Model-Agnostic Meta-Learning (MAML) and its variants are popular few-shot classification methods. They train an initializer across a variety of sampled learning tasks (also known as episodes) such that the initialized model can adapt quickly to new ones. However, current MAML-based algorithms have limitations in forming generalizable decision boundaries. In this paper, we propose an approach called MetaMix, which generates virtual feature-target pairs within each episode to regularize the backbone models. MetaMix can be integrated with any of the MAML-based algorithms and learn the decision boundaries generalizing better to new tasks. Experiments on the mini-ImageNet, CUB, and FC100 datasets show that MetaMix improves the performance of MAML-based algorithms and achieves state-of-the-art result when integrated with MetaTransfer Learning.

UR - http://www.scopus.com/inward/record.url?scp=85110535799&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85110535799&origin=recordpage

U2 - 10.1109/ICPR48806.2021.9413158

DO - 10.1109/ICPR48806.2021.9413158

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 978-1-7281-8809-6

T3 - Proceedings - International Conference on Pattern Recognition

SP - 407

EP - 414

BT - Proceedings of ICPR 2020

PB - IEEE

Y2 - 10 January 2021 through 15 January 2021

ER -

MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization

Abstract

Publication series

Conference

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this