Filtered Weighted Correction Training Method for Data with Noise Label

Yulong Wang, Yulong Wang, Yulong Wang, Xiaohui Hu, Xiaohui Hu, Zhe Jia

2021

Abstract

To solve the problem of low model accuracy under noisy data sets, a filtered weighted correction training method is proposed. This method uses the idea of model fine-tuning to adjust and correct the trained deep neural network model using filtered data, which has high portability. In the data filtering process, the noise label filtering algorithm, which is based on the random threshold in the double interval, reduces the dependence on artificially set parameters, increases the reliability of the random threshold, and improves the filtering accuracy and the recall rate of clean samples. In the calibration process, to deal with sample imbalance, different types of samples are weighted to improve the effectiveness of the model. Experimental results show that the propose method can improve the F1 value of deep neural network model.

Paper Citation

in Harvard Style

Wang Y., Hu X. and Jia Z. (2021). Filtered Weighted Correction Training Method for Data with Noise Label. In Proceedings of the 2nd International Conference on Deep Learning Theory and Applications - Volume 1: DeLTA, ISBN 978-989-758-526-5, pages 177-184. DOI: 10.5220/0010577901770184

in Bibtex Style

@conference{delta21,
author={Yulong Wang and Xiaohui Hu and Zhe Jia},
title={Filtered Weighted Correction Training Method for Data with Noise Label},
booktitle={Proceedings of the 2nd International Conference on Deep Learning Theory and Applications - Volume 1: DeLTA,},
year={2021},
pages={177-184},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010577901770184},
isbn={978-989-758-526-5},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Deep Learning Theory and Applications - Volume 1: DeLTA,
TI - Filtered Weighted Correction Training Method for Data with Noise Label
SN - 978-989-758-526-5
AU - Wang Y.
AU - Hu X.
AU - Jia Z.
PY - 2021
SP - 177
EP - 184
DO - 10.5220/0010577901770184