Research map / Label-efficient and robust learning / Recognition and long-tail
Learning from Noisy Pairwise Similarity and Unlabeled Data
Songhua Wu, Tongliang Liu, Bo Han, Jun Yu, Gang Niu, Masashi Sugiyama
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 |
|---|---|---|---|
| use noise-aware nSU risk instead of SU risk (linear) | nSU-LC · australian (pi+=0.7, rho_d=0.2) | accuracy (%): 63.7 -> 83.1 | Improved |
| use noise-aware nSU risk instead of SU risk (linear) | nSU-LC · svmguide1 (pi+=0.7, rho_d=0.2) | accuracy (%): 52.8 -> 76.1 | Improved |
| use noise-aware nSU risk instead of SU risk (linear) | nSU-LC · htru 2 (pi+=0.7, rho_d=0.2) | accuracy (%): 96.1 -> 96.4 | No clear effect |
| use nSU risk instead of nSD baseline | nSU-LC · australian (pi+=0.7, rho_d=0.2) | accuracy (%): 60.0 -> 83.1 | Improved |
| use linear classifier instead of deep classifier | nSU · australian (pi+=0.7, rho_d=0.2) | accuracy (%): 68.0 -> 83.1 | Improved |
| use nSU instead of SU under lower noise (rho_d=0.1) | nSU-LC · australian (pi+=0.8, rho_d=0.1) | accuracy (%): 82.6 -> 85.4 | Mixed |
| increase noise rate rho_d from 0.2 to 0.3 on synthetic data | SU classifier · syn1/syn2 | accuracy | Hurt |
| increase noise rate rho_d from 0.2 to 0.3 on synthetic data | nSU classifier · syn1/syn2 | accuracy | No clear effect |
| move class prior pi+ closer to 0.5 (0.7 -> 0.6 -> 0.55) | nSU-LC · australian (rho_d=0.2) | accuracy (%): 83.1 -> 75.1 -> 66.9 | Hurt |
| move class prior pi+ closer to 0.5 (0.7 -> 0.55) | nSU-LC · htru 2 (rho_d=0.2) | accuracy (%): 96.4 -> 86.4 | Hurt |
| add instance-pair weighting for shifted similar-pair distribution (ablation) | nSU-DC · breast-cancer (shifted, pi+=0.7, rho_d=0.2) | accuracy (%): 82.9 -> 93.6 | Improved |
| add instance-pair weighting for shifted similar-pair distribution (ablation) | nSU-DC · htru 2 (shifted, pi+=0.7, rho_d=0.2) | accuracy (%): 86.8 -> 90.8 | Improved |
Closest papers by meaning
- Classification from Pairwise Similarity and Unlabeled Data (ICML 2018 · similarity 0.91)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · similarity 0.83)
- Unsupervised Learning for Class Distribution Mismatch (ICML 2025 · similarity 0.83)
- Leveraging Data Symmetries to Select an Optimal Subset of Training Data under Label Noise (NeurIPS 2026 · similarity 0.82)
- Evaluating multiple models using labeled and unlabeled data (NeurIPS 2025 · similarity 0.82)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.82)
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.
