AI Paper Atlas

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

Distribution-Aware Robust Learning from Long-Tailed Data with Noisy Labels

Jae Soon Baik, In Young Yoon, Kun Hoon Kim, Jun Won Choi

ECCV 2024 · Label-efficient and robust learning › Noisy labels · approach: Contrastive learning on noisy data · 4 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
estimate class centroids with DaCC (all samples, prediction-weighted) instead of target-class samples (ablation)DaSC · Long-tailed CIFAR-10 (imbalance 0.1, symmetric noise 0.4)accuracy: +1.85%Improved
add temperature scaling to DaCC weights (ablation)DaSC · Long-tailed CIFAR-10 (imbalance 0.1, symmetric noise 0.4)accuracy: +0.4%Improved
add SBCL on high-confidence samples (ablation)DaSC · Long-tailed CIFAR-10 (imbalance 0.1, symmetric noise 0.4)accuracy: +0.76%Improved
add MIDL on low-confidence samples (ablation)DaSC · Long-tailed CIFAR-10 (imbalance 0.1, symmetric noise 0.4)accuracy: SBCL+MIDL +1.51%; total +3.76% over baselineImproved
use mixup samples as negative keys in MIDL instead of low-confidence samples (ablation)DaSC · Long-tailed CIFAR (various noise)accuracyImproved
apply SBCL to high-confidence and MIDL to low-confidence groups rather than other group assignments (ablation)DaSC · Long-tailed CIFAR (various noise)accuracyImproved
estimate centroids from confident subset D^I instead of all training samples D (ablation)DaCC · Long-tailed CIFAR-10 (imbalance 0.1, symmetric noise 0.4)accuracy: 87.89 -> 89.04Improved
use projector features z' rather than backbone features f(x) for centroids (ablation)DaCC · Long-tailed CIFAR-100 (imbalance 0.1, symmetric noise 0.4)accuracy: 61.16 -> 61.85Improved
use conventional classifier predictions rather than balanced classifier predictions to weight centroids (ablation)DaCC · Long-tailed CIFAR-10 (imbalance 0.1, asymmetric noise 0.2)accuracy: 89.57 -> 89.89Improved
DaSC vs strongest prior NL-LT method on CIFAR-100 symmetric noise 0.6DaSC · Long-tailed CIFAR-100 (imbalance 0.1, symmetric noise 0.6)accuracy: 47.71 -> 54.40 (vs SFA)Improved

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