Enhanced label noise robustness through early adaptive filtering for the self-supervised speaker verification task

Abderrahim Fathan, Xiaolin Zhu, Jahangir Alam
Proceedings of The 4th NeurIPS Efficient Natural Language and Speech Processing Workshop, PMLR 262:564-575, 2024.

Abstract

Using clustering-driven annotations to train a neural network can be a tricky task because of label noise. In this paper, we propose a dynamic and adaptive label noise filtering method, called AdaptiveDrop which combines both label noise cleansing and correction simultaneously in cascade to combine their advantages. Contrary to other label noise filtering approaches, our method filters noisy samples on the fly from an early stage of training. We also provide a variant that incorporates sub-centers per each class for enhanced robustness to label noise by continuously tracking the dominant sub-centers via a dictionary table. AdaptiveDrop is a simple general-purpose method, performed end-to-end in only one stage of training, can be integrated with any loss function, and does not require training from scratch on the cleansed dataset. We show through extensive ablation studies for the self-supervised speaker verification task that our method is effective, benefits from long epochs of iterative filtering and provides consistent performance gains across various loss functions and real-world pseudo-labels.

Cite this Paper

BibTeX
@InProceedings{pmlr-v262-fathan24a, title = {Enhanced label noise robustness through early adaptive filtering for the self-supervised speaker verification task}, author = {Fathan, Abderrahim and Zhu, Xiaolin and Alam, Jahangir}, booktitle = {Proceedings of The 4th NeurIPS Efficient Natural Language and Speech Processing Workshop}, pages = {564--575}, year = {2024}, editor = {Rezagholizadeh, Mehdi and Passban, Peyman and Samiee, Soheila and Partovi Nia, Vahid and Cheng, Yu and Deng, Yue and Liu, Qun and Chen, Boxing}, volume = {262}, series = {Proceedings of Machine Learning Research}, month = {14 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v262/main/assets/fathan24a/fathan24a.pdf}, url = {https://proceedings.mlr.press/v262/fathan24a.html}, abstract = {Using clustering-driven annotations to train a neural network can be a tricky task because of label noise. In this paper, we propose a dynamic and adaptive label noise filtering method, called AdaptiveDrop which combines both label noise cleansing and correction simultaneously in cascade to combine their advantages. Contrary to other label noise filtering approaches, our method filters noisy samples on the fly from an early stage of training. We also provide a variant that incorporates sub-centers per each class for enhanced robustness to label noise by continuously tracking the dominant sub-centers via a dictionary table. AdaptiveDrop is a simple general-purpose method, performed end-to-end in only one stage of training, can be integrated with any loss function, and does not require training from scratch on the cleansed dataset. We show through extensive ablation studies for the self-supervised speaker verification task that our method is effective, benefits from long epochs of iterative filtering and provides consistent performance gains across various loss functions and real-world pseudo-labels.} }
Endnote
%0 Conference Paper %T Enhanced label noise robustness through early adaptive filtering for the self-supervised speaker verification task %A Abderrahim Fathan %A Xiaolin Zhu %A Jahangir Alam %B Proceedings of The 4th NeurIPS Efficient Natural Language and Speech Processing Workshop %C Proceedings of Machine Learning Research %D 2024 %E Mehdi Rezagholizadeh %E Peyman Passban %E Soheila Samiee %E Vahid Partovi Nia %E Yu Cheng %E Yue Deng %E Qun Liu %E Boxing Chen %F pmlr-v262-fathan24a %I PMLR %P 564--575 %U https://proceedings.mlr.press/v262/fathan24a.html %V 262 %X Using clustering-driven annotations to train a neural network can be a tricky task because of label noise. In this paper, we propose a dynamic and adaptive label noise filtering method, called AdaptiveDrop which combines both label noise cleansing and correction simultaneously in cascade to combine their advantages. Contrary to other label noise filtering approaches, our method filters noisy samples on the fly from an early stage of training. We also provide a variant that incorporates sub-centers per each class for enhanced robustness to label noise by continuously tracking the dominant sub-centers via a dictionary table. AdaptiveDrop is a simple general-purpose method, performed end-to-end in only one stage of training, can be integrated with any loss function, and does not require training from scratch on the cleansed dataset. We show through extensive ablation studies for the self-supervised speaker verification task that our method is effective, benefits from long epochs of iterative filtering and provides consistent performance gains across various loss functions and real-world pseudo-labels.
APA
Fathan, A., Zhu, X. & Alam, J.. (2024). Enhanced label noise robustness through early adaptive filtering for the self-supervised speaker verification task. Proceedings of The 4th NeurIPS Efficient Natural Language and Speech Processing Workshop, in Proceedings of Machine Learning Research 262:564-575 Available from https://proceedings.mlr.press/v262/fathan24a.html.

Related Material