用稳定不变的差异性提升噪声标签下的模型鲁棒性
Leveraging Dissimilarity Invariance as a Robust Anchor for Learning with Noisy Labels
- 利用无关样本间差异性稳定的特点，转向关注鲁棒的差异信号
- 在CIFAR等数据集上显著优于现有方法，噪声下准确率提升5%以上
- 适合处理标签噪声场景，尤其适用于真实世界数据
深度学习在视觉识别中表现优异，但在标签存在噪声时性能急剧下降。现有方法难以学习到准确的相似性，导致训练过程被误导。本文发现一种新现象——语义差异性在标签噪声下依然保持稳定，即不相关样本间的差异具有不变性。基于此，提出可即插即用的NegScale框架，将注意力从易受干扰的相似性转移到更稳健的差异性。该框架包含：(1) 结构化负正交性惩罚（SNOP），强制无关样本在子空间中正交；(2) 差异性校准的相似性调节（DCSA），利用差异性锚点抑制虚假相似性。理论分析证明了差异性不变性及方法有效性。实验表明，NegScale在合成噪声和真实世界数据集上均持续优于现有最佳方法，刷新了基准性能。
原文摘要 · Abstract (English)
Deep learning models excel in visual recognition but suffer severe performance drops when training labels are corrupted by noise. Under label noise prior work cannot learn accurate similarities and thus misguide the learning process. In this paper, we uncover a complementary and novel phenomenon, Dissimilarity Invariance, whereby semantic dissimilarity between unrelated samples remains stable despite label noise. Leveraging this insight, we propose NegScale, a plug-and-play framework that shifts focus from fragile similarity to robust dissimilarity. NegScale integrates: (1) Structured Negative Orthogonality Penalty (SNOP), enforcing subspace orthogonality for unrelated samples; and (2) Dissimilarity-Calibrated Similarity Adjustment (DCSA), suppressing spurious similarity using dissimilarity anchors. We also give theoretical analysis that proves Dissimilarity Invariance and the effectiveness of NegScale. Empirical results demonstrate that NegScale consistently outperforms state-of-the-art baselines, establishing new benchmarks on CIFAR with synthetic noise and real-world datasets.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。