Research map / Label-efficient and robust learning / Recognition and long-tail
Self-PU: Self Boosted and Calibrated Positive-Unlabeled Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 self-paced training, calibrated loss, and self-distillation | nnPU · MNIST | accuracy: 93.41 (0.20) -> 94.21 (0.54) | Helped |
| add self-paced training, calibrated loss, and self-distillation | nnPU · CIFAR-10 | accuracy: 88.60 (0.40) -> 89.68 (0.22) | Helped |
| add self-paced training, calibrated loss, and self-distillation | nnPU · ADNI | accuracy: 75.96 (1.42) -> 79.50 (1.80) | Helped |
| use 3,000 instead of 1,000 positive examples | Self-PU · MNIST | accuracy: 94.21 (0.54) -> 96.00 (0.29) | Helped |
| use 3,000 instead of 1,000 positive examples | Self-PU · CIFAR-10 | accuracy: 89.68 (0.22) -> 90.77 (0.21) | Helped |
| replace fixed-size or permanent trusted-set sampling with dynamic-rate in-and-out sampling (ablation) | Self-PU self-paced training · CIFAR-10 | accuracy: 88.05 (0.59) -> 88.66 (0.40); 88.27 (0.43) -> 88.66 (0.40) | Helped |
| add soft labels (ablation) | Self-PU self-paced training · CIFAR-10 | accuracy: 88.66 (0.40) -> 88.75 (0.27) | Helped |
| add self-calibrated loss reweighting (ablation) | Self-PU self-paced training · CIFAR-10 | accuracy: 88.66 (0.40) -> 89.25 (0.42); 88.75 (0.27) -> 89.39 (0.36) | Helped |
| add student distillation (ablation) | Self-PU self-paced training · CIFAR-10 | accuracy: 88.60% to 88.84% | Helped |
| add teacher-network distillation (ablation) | Self-PU with student distillation · CIFAR-10 | accuracy: 88.84% -> 89.43% | Helped |
| use moderate student pace diversity (20% / 30%) rather than equal paces (ablation) | Self-PU student distillation · CIFAR-10 | accuracy: 89.64 (0.47) -> 89.65 (0.33) | No clear effect |
| use a large student pace gap (10% / 40%) instead of moderate diversity (ablation) | Self-PU student distillation · CIFAR-10 | accuracy: 89.65 (0.33) -> 89.32 (0.36) | Hurt |
Closest papers by meaning
- PU-BENCH: A UNIFIED BENCHMARK FOR RIGOROUS AND REPRODUCIBLE PU LEARNING (ICLR 2026 · similarity 0.89)
- Accessible, Realistic, and Fair Evaluation of Positive-Unlabeled Learning Algorithms (ICLR 2026 · similarity 0.88)
- Positive-Unlabeled Learning with Non-Negative Risk Estimator (NeurIPS 2017 · similarity 0.88)
- Learning from positive and unlabeled examples -Finite size sample bounds (NeurIPS 2025 · similarity 0.87)
- Positive and Unlabeled Learning with Controlled Probability Boundary Fence (ICML 2024 · similarity 0.85)
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · similarity 0.85)
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.