AI Paper Atlas

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

Understanding and Mitigating Human-Labelling Errors in Supervised Contrastive Learning

Zijun Long, Lipeng Zhuang, George W Killick, Richard Mccreadie, Gerardo Aragon-Camarasa, Paul Henderson

ECCV 2024 · Label-efficient and robust learning › Noisy labels · approach: Noisy cross-modal hashing and multi-label · 4 citations (OpenAlex)

Read the paper

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
replace SupCon loss with SCL-RHE (down-weight easy positives, correct for mislabels)BEiT-3 from scratch · ImageNet-1Ktop-1 accuracy: 82.57 -> 84.21Improved
replace SupCon with SCL-RHEBEiT-3 from scratch · CIFAR-100top-1 accuracy: 60.77 -> 64.47Improved
use synthetic-noise methods Sel-CL / TCL instead of SupCon on low-noise human-labelled dataBEiT-3 from scratch · ImageNet-1Ktop-1 accuracy: 82.57 -> 81.87 (Sel-CL), 81.74 (TCL)Hurt
replace cross-entropy with supervised contrastive objectives on small dataBEiT-3 from scratch · CIFAR-10top-1 accuracy: 71.70 -> 90.16Improved
evaluate on corrected test labels instead of originalSCL-RHE · ImageNet-1Ktop-1 accuracy: 84.21 -> 86.02 (+1.81)Improved
Sel-CL and TCL evaluated on corrected test labelsSel-CL/TCL · ImageNet-1Ktop-1 accuracy: 81.87 -> 81.35 (Sel-CL), 81.74 -> 81.28 (TCL)Hurt
fine-tune pre-trained BEiT-3 with SCL-RHE instead of SupConBEiT-3 · iNat2017top-1 accuracy: 72.85 -> 76.25Improved
fine-tune BEiT-3 with Sel-CL or TCL instead of cross-entropyBEiT-3 · CUB-200top-1 accuracy: 98.00 -> 94.52 (Sel-CL), 93.89 (TCL)Hurt
use SCL-RHE instead of Sel-CL for training timeResNet-18 contrastive training · CIFAR (time per epoch)min/epoch: 29.1 -> 20.4Improved
apply SCL-RHE at 20% synthetic label noiseResNet-18 · CIFAR-10 (20% noise)top-1 accuracy: 94.45 (Sel-CL) -> 94.71Improved
apply SCL-RHE at extreme 40% synthetic noise instead of Sel-CL/TCLResNet-18 · CIFAR-100 (40% noise)top-1 accuracy: 74.62 (TCL) -> 74.08Hurt
also handle false negatives in negative set (mitigation for negatives)SCL-RHE · label-error analysisshare of incorrect learning signals: 99.04% from false positives on CIFAR-100; 99.91% on ImageNet-1KNo clear effect

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