TY - GEN
T1 - Dual-Classifier Collaborative Method Based on Semi-Supervised Active Learning
AU - Wei, Baoguo
AU - Wang, Xinyu
AU - Zhang, Yue
AU - Cai, Mingzhi
AU - Li, Xu
AU - Li, Lixin
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Active Learning aims to reduce the annotation cost by selecting the most informative and representative samples to label. Typical active learning methods tend to overlook information from unlabeled samples. In contrast, semi-supervised learning utilizes unlabeled data that the model considers to be correctly classified temporarily. By integrating the advantages of both methods, we propose dual-classifier collaborative method based on semi-supervised active learning. Specifically, samples with the largest prediction difference between dual classifiers are queried for labeling by degree of prediction difference, while those with high confidence receive pseudo labels informed by the two classifiers. Under the increasing of labeled data, experiments on CIFAR and SVNH datasets show our method outperforms existing methods by 0.8% 5.84% in terms of classification accuracy.
AB - Active Learning aims to reduce the annotation cost by selecting the most informative and representative samples to label. Typical active learning methods tend to overlook information from unlabeled samples. In contrast, semi-supervised learning utilizes unlabeled data that the model considers to be correctly classified temporarily. By integrating the advantages of both methods, we propose dual-classifier collaborative method based on semi-supervised active learning. Specifically, samples with the largest prediction difference between dual classifiers are queried for labeling by degree of prediction difference, while those with high confidence receive pseudo labels informed by the two classifiers. Under the increasing of labeled data, experiments on CIFAR and SVNH datasets show our method outperforms existing methods by 0.8% 5.84% in terms of classification accuracy.
KW - active learning
KW - dual classifiers
KW - semi-supervised learning
UR - https://www.scopus.com/pages/publications/85184852135
U2 - 10.1109/ICSPCC59353.2023.10400229
DO - 10.1109/ICSPCC59353.2023.10400229
M3 - 会议稿件
AN - SCOPUS:85184852135
T3 - Proceedings of 2023 IEEE International Conference on Signal Processing, Communications and Computing, ICSPCC 2023
BT - Proceedings of 2023 IEEE International Conference on Signal Processing, Communications and Computing, ICSPCC 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 IEEE International Conference on Signal Processing, Communications and Computing, ICSPCC 2023
Y2 - 14 November 2023 through 17 November 2023
ER -