Research map / Label-efficient and robust learning / Noisy labels
Noisy-Pair Robust Representation Alignment for Positive-Unlabeled Learning
Hengwei Zhao, Zhengzhong Tu, Zhuo Zheng, Wei Wang, Junjue Wang, Rusty Feagin, Wenzhe Jiao
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add NoiSNCL non-contrastive representation loss to label disambiguation (ablation) | NcPU · CIFAR-100 | overall accuracy: 61.54 -> 88.28 | Improved |
| add label disambiguation (PhantomGate pseudo targets) to NoiSNCL (ablation) | NcPU · CIFAR-100 | overall accuracy: 50.27 -> 88.28 | Improved |
| use noisy-pair robust NoiSNCL instead of supervised non-contrastive L_r (ablation) | NcPU · CIFAR-100 | overall accuracy: 84.58 -> 88.28 | Improved |
| use supervised pseudo-label NoiSNCL instead of self-supervised L_self-r (ablation) | NcPU · CIFAR-100 | overall accuracy: 73.22 -> 88.28 | Improved |
| use PhantomGate (regret-based update with SAT) instead of plain prototype target s' (ablation) | NcPU · CIFAR-100 | overall accuracy: 75.14 -> 88.28 | Improved |
| add SAT negative supervision without regret-based update to prototype disambiguation (ablation) | NcPU · CIFAR-100 | overall accuracy: 75.14 -> 50.25 | Hurt |
| add NoiSNCL to uPU (ablation) | uPU · CIFAR-100 (real class prior) | overall accuracy: 61.68 -> 83.71 | Improved |
| add NoiSNCL to nnPU (ablation) | nnPU · CIFAR-100 (real class prior) | overall accuracy: 71.22 -> 87.81 | Improved |
| add NoiSNCL to uPU (ablation) | uPU · CIFAR-10 (real class prior) | overall accuracy: 69.43 -> 97.35 | Improved |
| use NoiSNCL instead of L_r in fully supervised learning (ablation) | Supervised · CIFAR-100 | overall accuracy: 94.45 -> 94.56 | No clear effect |
| use NcPU instead of best baseline without auxiliary negatives | NcPU · CIFAR-100 | overall accuracy: 12.89% OA improvement | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PiCO — PiCO: Contrastive Label Disambiguation for Partial Label Learning (ICLR 2022) · better on 3 of 3 shared results
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
- A Closer Look to Positive-Unlabeled Learning from Fine-grained Perspectives: An Empirical Study (NeurIPS 2025 · similarity 0.89)
- Accessible, Realistic, and Fair Evaluation of Positive-Unlabeled Learning Algorithms (ICLR 2026 · similarity 0.89)
- Balancing Positive and Negative Classification Error Rates in Positive-Unlabeled Learning (NeurIPS 2025 · similarity 0.89)
- Learning from positive and unlabeled examples -Finite size sample bounds (NeurIPS 2025 · similarity 0.87)
- Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation (CVPR 2024 · similarity 0.86)
- Positive and Unlabeled Learning with Controlled Probability Boundary Fence (ICML 2024 · similarity 0.86)
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.
