Deep active sampling with self-supervised learning

doi:10.1007/s11704-022-2189-z

Front. Comput. Sci.

2023, Vol. 17

Issue (4) : 174323 https://doi.org/10.1007/s11704-022-2189-z

LETTER

Deep active sampling with self-supervised learning

Haochen SHI^1,²(

), Hui ZHOU²

¹. College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
². College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China

Download: PDF(2186 KB) HTML
Export: BibTeX | EndNote | Reference Manager | ProCite | RefWorks

Corresponding Author(s): Haochen SHI

Just Accepted Date: 23 August 2022 Issue Date: 03 November 2022

Cite this article:

Haochen SHI,Hui ZHOU. Deep active sampling with self-supervised learning[J]. Front. Comput. Sci., 2023, 17(4): 174323.

URL:

https://academic.hep.com.cn/fcs/EN/10.1007/s11704-022-2189-z
https://academic.hep.com.cn/fcs/EN/Y2023/V17/I4/174323

Fig.1 Deep active sampling framework with self-supervised learning

Fig.2 Performance comparison of various methods when the self-supervised model is MoCo. (a) Fashion-MNIST; (b) CIFAR10

Fig.3 Performance comparison of various methods when the self-supervised method is to predict the image rotations. (a) Fashion-MNIST; (b) CIFAR10

Sampling	Query sample quantity
Sampling	1000	2000	3000	4000	5000
$I = u$	0.8022	0.8209	0.8315	0.8457	0.8570
$I = ν$	0.8021	0.8163	0.8247	0.8265	0.8293
$I = u + λ ν$	0.8053	0.8273	0.8418	0.8530	0.8597

Tab.1 Experimental results of various sampling metric under different query numbers on fashion-MNIST

Sampling	Query sample quantity
Sampling	1000	2000	3000	4000	5000
$I = u$	0.6722	0.7120	0.7343	0.7478	0.7576
$I = ν$	0.6749	0.6972	0.7133	0.7354	0.7480
$I = u + λ ν$	0.6875	0.7280	0.7528	0.7625	0.7757

Tab.2 Experimental results of various sampling metric under different query numbers on CIFAR10

1	J Z, Bengar de Weijer J, van B, Twardowski B Raducanu . Reducing label effort: self-supervised meets active learning. In: Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. 2021, 1631–1639
2	K, He H, Fan Y, Wu S, Xie R Girshick . Momentum contrast for unsupervised visual representation learning. In: Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020, 9726–9735
3	H, Xiao K, Rasul R Vollgraf . Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms. 2017, arXiv preprint arXiv: 1708.07747
4	A, Krizhevsky G Hinton . Learning multiple layers of features from tiny images. See Google website, 2009

[1]

FCS-22189-OF-HS_suppl_1

Download

Viewed

Full text

Abstract

Cited

Shared

Discussed