Research map / Label-efficient and robust learning / Semi-supervised learning
Sinkhorn Label Allocation: Semi-Supervised Classification via Annealed Self-Training
Kai Sheng Tai, Peter D Bailis, Gregory Valiant
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 SLA label allocation instead of FixMatch pseudo-label thresholding | self-training · CIFAR-10, 40 labels | test error (%): 9.90±3.00 -> 5.17±0.32 | Improved |
| Use SLA label allocation instead of FixMatch pseudo-label thresholding | self-training · CIFAR-100, 2500 labels | test error (%): 27.99±0.42 -> 28.73±0.44 | Hurt |
| Use SLA label allocation instead of FixMatch pseudo-label thresholding | self-training · SVHN, 20 labels | test error (%): 14.92±7.82 -> 22.85±9.84 | Hurt |
| Use SLA label allocation instead of FixMatch pseudo-label thresholding | self-training · SVHN, 40 labels | test error (%): 4.74±3.28 -> 3.63±2.91 | Improved |
| Add class proportion upper bounds (ablation) | SLA · CIFAR-10, 40 labels, Uniform | test error (%): 9.71±5.95 -> 5.17±0.32 | Improved |
| Add class proportion upper bounds estimated from labeled proportions (ablation) | SLA · CIFAR-10, 40 labels, Multinomial | test error (%): 13.40±6.41 -> 14.95±7.12 | Hurt |
| Add a linear label-allocation ramp rather than allocating all examples from the start (ablation) | SLA · CIFAR-10, 40 labels | test error (%): 13.67 ± 1.83 -> 5.17±0.32 | Improved |
| Increase Sinkhorn regularization from γ = 1 to γ = 10 (ablation) | SLA · CIFAR-10, 40 labels | test error (%): 42.48 -> 5.78 | Improved |
| Increase Sinkhorn regularization from γ = 10 to γ = 100 (ablation) | SLA · CIFAR-10, 40 labels | test error (%): 5.78 -> 4.94 | Improved |
| Increase Sinkhorn regularization from γ = 100 to γ = 1000 (ablation) | SLA · CIFAR-10, 40 labels | test error (%): 4.94 -> 5.10 | Hurt |
| Use a truncated allocation schedule that stops before full allocation | SLA · CIFAR-100 | test error (%) | No clear effect |
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.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 9 of 12 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
- LaSSL: Label-Guided Self-Training for Semi-supervised Learning (AAAI 2022 · similarity 0.87)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.87)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.87)
- Partial Label Learning with Self-Guided Retraining (AAAI 2019 · similarity 0.87)
- Exploiting Unlabeled Data via Partial Label Assignment for Multi-Class Semi-Supervised Learning (AAAI 2021 · similarity 0.87)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (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.
