AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

PiCO: Contrastive Label Disambiguation for Partial Label Learning

Haobo Wang, Ruixuan Xiao, Yixuan Li, Lei Feng, Gang Niu, Gang Chen, Junbo Zhao

ICLR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Partial and multi-label learning · 55 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 contrastive representation learning and prototype-based label disambiguationPLL classifier · CIFAR-10, q=0.5accuracy: 93.58 vs 87.78Improved
Add contrastive representation learning and prototype-based label disambiguationPLL classifier · CIFAR-100, q=0.05accuracy: 72.74 vs 60.73Improved
Add clean-sample detection, semi-supervised contrastive learning, and mixupPiCO · CIFAR-10, q=0.5, η=0.2accuracy: 80.07 -> 92.59Improved
Add clean-sample detection, semi-supervised contrastive learning, and mixupPiCO · CIFAR-100, q=0.05, η=0.2accuracy: 59.81 -> 72.98Improved
Add prototype-based disambiguation instead of uniform pseudo targets (ablation)PiCO · CIFAR-10, q=0.5accuracy: 84.50 -> 93.58Improved
Add contrastive loss alongside uniform pseudo targets (ablation)PiCO · CIFAR-10, q=0.5accuracy: 76.46 -> 84.50Improved
Use moving-average prototype targets rather than direct soft prototype probabilities (ablation)PiCO · CIFAR-10, q=0.5accuracy: 91.60 -> 93.58Improved
Use moving-average targets rather than direct one-hot nearest-prototype targets (ablation)PiCO · CIFAR-100, q=0.05accuracy: 70.10 -> 72.74Improved
Add neighbor-augmented contrastive loss (ablation)PiCO+ · CIFAR-10, q=0.5, η=0.2accuracy: 87.43 -> 92.59Improved
Add mixup training (ablation)PiCO+ · CIFAR-100, q=0.05, η=0.2accuracy: 68.51 -> 72.98Improved
Add prototype-based label guessing loss for noisy samples (ablation)PiCO+ · CIFAR-100, q=0.05, η=0.2accuracy: 71.98 -> 72.98Improved
Use distance-based clean-sample selection instead of small-loss selection (ablation)PiCO+ · CIFAR-10, q=0.5, η=0.2accuracy: 91.22 -> 92.59Improved

Later papers that compared with this method

22 later accepted papers report PiCO+ as a baseline in their results tables; 21 of them report a better number on most of the results they share with it.

14 more not listed.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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