Research map / Label-efficient and robust learning / Semi-supervised learning
InstanT: Semi-supervised Learning with Instance-dependent Thresholds
Muyang Li, Runze Wu, Haoyu Liu, Jun Yu, Xun Yang, Bo Han, Tongliang Liu
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 instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · CIFAR-10 (10 labels), pre-trained ViT | top-1 accuracy: 85.15{ $\pm$20.4} -> 87.32{ $\pm$10.2} | Improved |
| use instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · CIFAR-10 (40 labels), pre-trained ViT | top-1 accuracy: 97.94{ $\pm$0.1} -> 97.93{ $\pm$0.1} | No clear effect |
| use instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · CIFAR-100 (200 labels), pre-trained ViT | top-1 accuracy: 73.61{ $\pm$0.1} -> 74.17{ $\pm$0.3} | Improved |
| use instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · STL-10 (40 labels), pre-trained ViT | top-1 accuracy: 83.50{ $\pm$4.2} -> 85.09{ $\pm$2.8} | Improved |
| use instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · CIFAR-100 (400 labels), WRN-28-2 | top-1 accuracy: 52.02{ $\pm$1.7} -> 53.94{ $\pm$1.8} | Improved |
| use instance-dependent thresholds instead of AdaMatch's thresholding | InstanT · CIFAR-10 (250 labels), WRN-28-2 | top-1 accuracy: 94.76{ $\pm$0.1} -> 94.72{ $\pm$0.2} | Hurt |
| add instance-dependent thresholding to a FixMatch-style setup without RT or DA (ablation) | InstanT · STL-10 (40 labels) | top-1 accuracy: 77.72{ $\pm$4.4} -> 79.92{ $\pm$5.8} | Improved |
| add RT and DA (ablation) | InstanT-1 · STL-10 (40 labels) | top-1 accuracy: 79.92{ $\pm$5.8} -> 85.09{ $\pm$2.8} | Improved |
| add DA (ablation) | InstanT-2 · STL-10 (40 labels) | top-1 accuracy: 81.49{ $\pm$4.9} -> 85.09{ $\pm$2.8} | Improved |
| add RT (ablation) | InstanT-3 · STL-10 (40 labels) | top-1 accuracy: 82.97{ $\pm$2.7} -> 85.09{ $\pm$2.8} | Improved |
| use an instance-dependent threshold instead of a fixed threshold (ablation) | InstanT-1 · STL-10 (40 labels) | top-1 accuracy: 77.72{ $\pm$4.4} -> 79.92{ $\pm$5.8} | Improved |
| train a transition-matrix estimator alongside the classifier | InstanT · STL-10 (40 labels) | seconds per epoch: 30.1 -> 31.1 | Hurt |
Later papers that compared with this method
1 later accepted paper reports InstanT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- AdaMatch — AdaMatch: A Unified Approach to Semi-Supervised Learning and Domain Adaptation (ICLR 2022) · better on 12 of 16 shared results
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 0 of 1 shared result
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
- Rethinking Confidence Scores and Thresholds in Pseudolabeling-based SSL (ICML 2025 · similarity 0.89)
- Bi-Level Optimization for Semi-Supervised Learning with Pseudo-Labeling (AAAI 2025 · similarity 0.89)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.88)
- Reduction-based Pseudo-label Generation for Instance-dependent Partial Label Learning (NeurIPS 2025 · similarity 0.87)
- Leveraging Local Variance for Pseudo-Label Selection in Semi-supervised Learning (AAAI 2024 · similarity 0.87)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.87)
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.
