AI Paper Atlas

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

When Noisy Labels Meet Long Tail Dilemmas: A Representation Calibration Method

Manyi Zhang, Xuyang Zhao, Jun Yao, Chun Yuan, Weiran Huang

ICCV 2023 · Label-efficient and robust learning › Noisy labels · approach: Noise-aware loss and pairwise methods · 29 citations (OpenAlex)

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 mixup to RCAL (ablation)RCAL · CIFAR-10 (imbalance 100, noise 0.4)test accuracy (%): 64.83 -> 69.78Improved
add representation distance regularization (REG) toward contrastive representations (ablation)RCAL · CIFAR-10 (imbalance 100, noise 0.4)test accuracy (%): 58.27 -> 64.83Improved
add distributional calibration (DC) with sampled Gaussian points (ablation)RCAL · CIFAR-10 (imbalance 100, noise 0.4)test accuracy (%): 54.61 -> 58.27Improved
add MoCo contrastive pretraining (CL) (ablation)RCAL · CIFAR-10 (imbalance 100, noise 0.4)test accuracy (%): 48.11 -> 54.61Improved
add contrastive learning pretraining (ablation)RCAL · CIFAR-10 (imbalance 10, noise 0.4)test accuracy (%): 70.13 -> 74.37Improved
change batch size from 64 to 256 (ablation)RCAL · CIFAR-10/100 (noise 0.5)test accuracy (%): CIFAR-10 imb 10: 79.53 -> 80.92; imb 100: 65.69 -> 63.27No clear effect
RCAL versus best baseline at high noise and imbalanceRCAL · CIFAR-10 (imbalance 100, noise 0.5)test accuracy (%): 56.21 -> 65.05Improved
Co-teaching small-loss trick under long-tailed dataCo-teaching · CIFAR-10 (imbalance 10, noise 0.2)test accuracy (%): 75.61 -> 78.54 vs ERM at imb 10; 62.17 -> 50.29 at imb 100Hurt
add semi-supervised learning to RCAL (RCAL+)RCAL · WebVision-50 (WebVision val)top-1 accuracy (%): 76.24 -> 79.56Improved
RCAL versus MiSLAS on many/medium/few classesRCAL · CIFAR-10 (imbalance 10, noise 0.5)overall accuracy (%): 72.46 -> 80.80 overall; few classes 81.66 -> 73.98Mixed

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