AI Paper Atlas

Research map / Label-efficient and robust learning / Noisy labels

Do We Need to Penalize Variance of Losses for Learning with Label Noise?

Yexiong Lin, Yu Yao, Yuxuan Du, Jun Yu, Bo Han, Mingming Gong, Tongliang Liu

JMLR 2026 · Label-efficient and robust learning › Noisy labels · approach: Noise-aware loss and pairwise methods

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add VRNL (encourage large loss variance)Reweight · Fashion-MNIST Sym-90%accuracy: 45.63 -> 77.74Improved
add VRNLForward · CIFAR-100 Sym-50%accuracy: 43.45 -> 56.49Improved
add VRNLReweight · CIFAR-10 Sym-80%accuracy: 51.70 -> 63.24Improved
add VRNLBLTM · CIFAR-100 Sym-50%accuracy: 30.56 -> 31.32No clear effect
add VRNLVolMinNet · CIFAR-10 IDN-20%accuracy: 90.29 -> 90.27No clear effect
add VRNLVolMinNet · WebVision val (top-1)accuracy: 72.96 -> 74.16Improved
add VRNLForward · Clothing1Maccuracy: 71.27 -> 72.43Improved
add VRNLReweight · CIFAR-100 Sym-10%accuracy: 65.72 -> 67.57Improved
add VRNLForward · CIFAR-10 clean labels (Sym-00%)accuracy: 92.78 -> 92.71No clear effect
add linear warmup of alpha (ablation)VolMinNet-VRNL · CIFAR-10 Sym-80%accuracy: 59.88 -> 61.23Improved
add linear warmup of alpha (ablation)VolMinNet-VRNL · CIFAR-10 Sym-90%accuracy: 34.79 -> 37.41Improved
add linear warmup of alpha (ablation)VolMinNet-VRNL · CIFAR-10 IDN-50%accuracy: 61.48 -> 64.94Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy