Scholarlink

<p>Noisy labels pose a major challenge in supervised learning, often undermining the reliability and generalization of deep neural networks. Addressing this issue requires mitigating the adverse impact of mislabeled samples and avoiding overly complex architectures or extended training procedures. To this end, this paper proposes the <i>NLC</i> block (<i>Noisy Label Correction</i>), a lightweight, plug-and-play module inspired by the <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10489_2025_6594_Article_IEq1.gif" Format="GIF" Height="12" Rendition="HTML" Resolution="72" Type="Linedraw" Width="13" /> </InlineMediaObject> <EquationSource Format="TEX">\(\gamma \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>γ</mi> </math></EquationSource> </InlineEquation>-divergence weighting principle. Unlike traditional parameter-dependent methods, the <i>NLC</i> block integrates a feed-forward layer with a closed-form formula computation layer to dynamically reweight samples without introducing additional learnable parameters. This paper provides a theoretical analysis demonstrating its robustness and shows, through extensive experiments on real-world datasets, that the <i>NLC</i> block significantly improves model accuracy and stability under label noise. The implementation is publicly available at <a href="https://github.com/DebtVC2022/NLC-block">https://github.com/DebtVC2022/NLC-block</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

NLC-block: Enhancing neural network training robustness with noisy label reweighting

  • 期刊论文

  • 发表日期：

    2025年5月7日

查看全文
  • Ben Liu,
  • Qiao Hu

摘要

Noisy labels pose a major challenge in supervised learning, often undermining the reliability and generalization of deep neural networks. Addressing this issue requires mitigating the adverse impact of mislabeled samples and avoiding overly complex architectures or extended training procedures. To this end, this paper proposes the NLC block (Noisy Label Correction), a lightweight, plug-and-play module inspired by the \(\gamma \) γ -divergence weighting principle. Unlike traditional parameter-dependent methods, the NLC block integrates a feed-forward layer with a closed-form formula computation layer to dynamically reweight samples without introducing additional learnable parameters. This paper provides a theoretical analysis demonstrating its robustness and shows, through extensive experiments on real-world datasets, that the NLC block significantly improves model accuracy and stability under label noise. The implementation is publicly available at https://github.com/DebtVC2022/NLC-block.