Staff View: Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement

Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement

The task of continual learning is to design algorithms that can address the problem of catastrophic forgetting. However, in the real world, there are noisy labels due to inaccurate human annotations and other factors, which seem to exacerbate catastrophic forgetting. To tackle both catastrophic forg...

Full description

Saved in:

Bibliographic Details
Main Authors:	Guo G., Wei Z., Cheng J.
Other Authors:	58805753200
Format:	Conference paper
Published:	Springer Science and Business Media Deutschland GmbH 2025
Subjects:	Catastrophic forgetting Continual learning Feature enhancement Noisy data Noisy labels Real-world Replay Sample features Samples selection Uncertainty Entropy
Tags:	Add Tag No Tags, Be the first to tag this record!

id	my.uniten.dspace-37178
record_format	dspace
spelling	my.uniten.dspace-371782025-03-03T15:48:19Z Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement Guo G. Wei Z. Cheng J. 58805753200 58805777500 22833734200 Catastrophic forgetting Continual learning Feature enhancement Noisy data Noisy labels Real-world Replay Sample features Samples selection Uncertainty Entropy The task of continual learning is to design algorithms that can address the problem of catastrophic forgetting. However, in the real world, there are noisy labels due to inaccurate human annotations and other factors, which seem to exacerbate catastrophic forgetting. To tackle both catastrophic forgetting and noise issues, we propose an innovative framework. Our framework leverages sample uncertainty to purify the data stream and selects representative samples for replay, effectively alleviating catastrophic forgetting. Additionally, we adopt a semi-supervised approach for fine-tuning to ensure the involvement of all available samples. Simultaneously, we incorporate contrastive learning and entropy minimization to mitigate noise memorization in the model. We validate the effectiveness of our proposed method through extensive experiments on two benchmark datasets, CIFAR-10 and CIFAR-100. For CIFAR-10, we achieve a performance gain of 2% under 20% noise conditions. ? The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2024. Final 2025-03-03T07:48:18Z 2025-03-03T07:48:18Z 2024 Conference paper 10.1007/978-981-99-8543-2_40 2-s2.0-85181983496 https://www.scopus.com/inward/record.uri?eid=2-s2.0-85181983496&doi=10.1007%2f978-981-99-8543-2_40&partnerID=40&md5=f79b7a4392845c0d29b3d4190ac18737 https://irepository.uniten.edu.my/handle/123456789/37178 14432 LNCS 498 510 Springer Science and Business Media Deutschland GmbH Scopus
institution	Universiti Tenaga Nasional
building	UNITEN Library
collection	Institutional Repository
continent	Asia
country	Malaysia
content_provider	Universiti Tenaga Nasional
content_source	UNITEN Institutional Repository
url_provider	http://dspace.uniten.edu.my/
topic	Catastrophic forgetting Continual learning Feature enhancement Noisy data Noisy labels Real-world Replay Sample features Samples selection Uncertainty Entropy
spellingShingle	Catastrophic forgetting Continual learning Feature enhancement Noisy data Noisy labels Real-world Replay Sample features Samples selection Uncertainty Entropy Guo G. Wei Z. Cheng J. Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
description	The task of continual learning is to design algorithms that can address the problem of catastrophic forgetting. However, in the real world, there are noisy labels due to inaccurate human annotations and other factors, which seem to exacerbate catastrophic forgetting. To tackle both catastrophic forgetting and noise issues, we propose an innovative framework. Our framework leverages sample uncertainty to purify the data stream and selects representative samples for replay, effectively alleviating catastrophic forgetting. Additionally, we adopt a semi-supervised approach for fine-tuning to ensure the involvement of all available samples. Simultaneously, we incorporate contrastive learning and entropy minimization to mitigate noise memorization in the model. We validate the effectiveness of our proposed method through extensive experiments on two benchmark datasets, CIFAR-10 and CIFAR-100. For CIFAR-10, we achieve a performance gain of 2% under 20% noise conditions. ? The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2024.
author2	58805753200
author_facet	58805753200 Guo G. Wei Z. Cheng J.
format	Conference paper
author	Guo G. Wei Z. Cheng J.
author_sort	Guo G.
title	Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
title_short	Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
title_full	Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
title_fullStr	Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
title_full_unstemmed	Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement
title_sort	enhancing continual noisy label learning with�uncertainty-based sample selection and�feature enhancement
publisher	Springer Science and Business Media Deutschland GmbH
publishDate	2025
_version_	1826077772317982720
score	13.244413

Enhancing Continual Noisy Label Learning with�Uncertainty-Based Sample Selection and�Feature Enhancement

Similar Items