Research map / Label-efficient and robust learning / Semi-supervised learning
Dash: Semi-Supervised Learning with Dynamic Thresholding
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 |
|---|---|---|---|
| Use dynamic loss thresholding instead of FixMatch's fixed threshold (CTA) | FixMatch · CIFAR-10, 40 labels | top-1 test error (%): 11.39±3.35 -> 9.16±4.31 | Helped |
| Use dynamic thresholding instead of fixed thresholding (RA) | FixMatch · CIFAR-10, 250 and 4000 labels; CIFAR-100, 2500 and 10000 labels; SVHN, 250 and 1000 labels | top-1 test error (%): 5.07±0.65 -> 4.56±0.13; 4.26±0.05 -> 4.08±0.06; 28.29±0.11 -> 27.18±0.21; 22.60±0.12 -> 21.97±0.14; 2.48±0.38 -> 2.17±0. | Helped |
| Use dynamic thresholding instead of fixed thresholding (CTA) | FixMatch · STL-10, 1000 labels | top-1 test error (%): 5.17±0.63 -> 3.96±0.25 | Helped |
| Use dynamic thresholding instead of fixed thresholding (RA) | FixMatch · CIFAR-100, 400 labels | top-1 test error (%): 48.85±1.75 -> 44.76±0.96 | Hurt |
| Add distribution alignment to Dash | Dash · CIFAR-100, 400 labels | top-1 test error (%): 43.31% | Helped |
| Set γ to 1.01 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 250 labels | top-1 test error (%): 4.76 -> 4.85 | Hurt |
| Set γ to 1.2 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 250 labels | top-1 test error (%): 4.76 -> 4.99 | Hurt |
| Set γ to 1.3 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 250 labels | top-1 test error (%): 4.76 -> 4.82 | Hurt |
| Set γ to 1.01 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 4000 labels | top-1 test error (%): 4.11 -> 4.39 | Hurt |
| Set γ to 1.1 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 4000 labels | top-1 test error (%): 4.11 -> 4.28 | Hurt |
| Set γ to 1.3 rather than the best tested value (ablation) | Dash-CTA · CIFAR-10, 4000 labels | top-1 test error (%): 4.11 -> 4.31 | Hurt |
| Replace FixMatch with Pseudo-Labeling as the pipeline inside Dash (ablation) | Dash · CIFAR-10, 250 labels | top-1 test error (%): 49.78 -> 46.90 | Helped |
Closest papers by meaning
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.89)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.86)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.86)
- Bi-Level Optimization for Semi-Supervised Learning with Pseudo-Labeling (AAAI 2025 · similarity 0.86)
- Enhancing Semi-Supervised Learning via Representative and Diverse Sample Selection (NeurIPS 2024 · similarity 0.86)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 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.