In this paper, we propose a semi-supervised triplet loss function that realizes semi-supervised representation learning in a novel manner. We extend conventional triplet loss, which uses labeled data to achieve representation learning, so that it can deal with unlabeled data. We estimate, in advance, the degree to which each label applies to each unlabeled data point, and optimize the loss function with unlabeled features according to the resulting ratios. Since the proposed loss function has the effect of adjusting the distribution of all unlabeled data, it complements methods based on consistency regularization, which has been extensively studied in recent years. Combined with a consistency regularization-based method, our method achieves more accurate semi-supervised learning. Experiments show that the proposed loss function achieves a higher accuracy than the conventional fine-tuning method.
Kazuhiko MURASAKI
NTT
Shingo ANDO
NTT
Jun SHIMAMURA
NTT
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Kazuhiko MURASAKI, Shingo ANDO, Jun SHIMAMURA, "Semi-Supervised Representation Learning via Triplet Loss Based on Explicit Class Ratio of Unlabeled Data" in IEICE TRANSACTIONS on Information,
vol. E105-D, no. 4, pp. 778-784, April 2022, doi: 10.1587/transinf.2021EDP7073.
Abstract: In this paper, we propose a semi-supervised triplet loss function that realizes semi-supervised representation learning in a novel manner. We extend conventional triplet loss, which uses labeled data to achieve representation learning, so that it can deal with unlabeled data. We estimate, in advance, the degree to which each label applies to each unlabeled data point, and optimize the loss function with unlabeled features according to the resulting ratios. Since the proposed loss function has the effect of adjusting the distribution of all unlabeled data, it complements methods based on consistency regularization, which has been extensively studied in recent years. Combined with a consistency regularization-based method, our method achieves more accurate semi-supervised learning. Experiments show that the proposed loss function achieves a higher accuracy than the conventional fine-tuning method.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDP7073/_p
Copy
@ARTICLE{e105-d_4_778,
author={Kazuhiko MURASAKI, Shingo ANDO, Jun SHIMAMURA, },
journal={IEICE TRANSACTIONS on Information},
title={Semi-Supervised Representation Learning via Triplet Loss Based on Explicit Class Ratio of Unlabeled Data},
year={2022},
volume={E105-D},
number={4},
pages={778-784},
abstract={In this paper, we propose a semi-supervised triplet loss function that realizes semi-supervised representation learning in a novel manner. We extend conventional triplet loss, which uses labeled data to achieve representation learning, so that it can deal with unlabeled data. We estimate, in advance, the degree to which each label applies to each unlabeled data point, and optimize the loss function with unlabeled features according to the resulting ratios. Since the proposed loss function has the effect of adjusting the distribution of all unlabeled data, it complements methods based on consistency regularization, which has been extensively studied in recent years. Combined with a consistency regularization-based method, our method achieves more accurate semi-supervised learning. Experiments show that the proposed loss function achieves a higher accuracy than the conventional fine-tuning method.},
keywords={},
doi={10.1587/transinf.2021EDP7073},
ISSN={1745-1361},
month={April},}
Copy
TY - JOUR
TI - Semi-Supervised Representation Learning via Triplet Loss Based on Explicit Class Ratio of Unlabeled Data
T2 - IEICE TRANSACTIONS on Information
SP - 778
EP - 784
AU - Kazuhiko MURASAKI
AU - Shingo ANDO
AU - Jun SHIMAMURA
PY - 2022
DO - 10.1587/transinf.2021EDP7073
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E105-D
IS - 4
JA - IEICE TRANSACTIONS on Information
Y1 - April 2022
AB - In this paper, we propose a semi-supervised triplet loss function that realizes semi-supervised representation learning in a novel manner. We extend conventional triplet loss, which uses labeled data to achieve representation learning, so that it can deal with unlabeled data. We estimate, in advance, the degree to which each label applies to each unlabeled data point, and optimize the loss function with unlabeled features according to the resulting ratios. Since the proposed loss function has the effect of adjusting the distribution of all unlabeled data, it complements methods based on consistency regularization, which has been extensively studied in recent years. Combined with a consistency regularization-based method, our method achieves more accurate semi-supervised learning. Experiments show that the proposed loss function achieves a higher accuracy than the conventional fine-tuning method.
ER -