Research map / Label-efficient and robust learning / Semi-supervised learning
FlatMatch: Bridging Labeled Data and Unlabeled Data with Cross-Sharpness for Semi-Supervised Learning
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 |
|---|---|---|---|
| compute cross-sharpness with labeled-data worst-case model and unlabeled-data prediction difference (vs sharpness on labeled data only) (ablation) | FlatMatch · CIFAR100 2500 labels | error rate (%): 26.85 -> 25.38 | Helped |
| compute sharpness entirely on unlabeled data (vs cross-sharpness) (ablation) | FlatMatch · CIFAR100 2500 labels | error rate (%): 25.38 -> 36.30 | Hurt |
| compute sharpness on full dataset D^l U D^u (ablation) | FlatMatch · CIFAR100 10000 labels | error rate (%): 19.01 -> 23.42 | Hurt |
| add EMA smoothing of the stored labeled gradient (ablation) | FlatMatch-e · CIFAR100 2500 labels | error rate (%): 29.44 -> 25.62 | Helped |
| add EMA smoothing of the stored labeled gradient (ablation) | FlatMatch-e · CIFAR100 10000 labels | error rate (%): 23.23 -> 19.78 | Helped |
| use efficient reuse of previous-batch gradient (FlatMatch-e vs FlatMatch) (ablation) | FlatMatch · CIFAR100 10000 labels | error rate (%): 19.01 -> 19.78 | No clear effect |
| add pseudo-labeled fixed labels for computing sharpness (ablation) | FlatMatch · CIFAR10 10 labels | error rate (%): 15.23 -> 7.36 | Helped |
| add pseudo-labeled fixed labels for computing sharpness (ablation) | FlatMatch · CIFAR100 400 labels | error rate (%): 38.76 -> 36.97 | Helped |
| replace instance-wise consistency of FreeMatch with cross-sharpness | FreeMatch · CIFAR100 2500 labels | error rate (%): 26.47 -> 25.38 | Helped |
| replace instance-wise consistency of FreeMatch with cross-sharpness under very few labels | FreeMatch · CIFAR10 10 labels | error rate (%): 8.07 -> 15.23 | Hurt |
| replace FreeMatch with FlatMatch | FreeMatch · SVHN 250 labels | error rate (%): 1.97 -> 1.43 | Helped |
| increase perturbation magnitude rho above 0.25 (ablation) | FlatMatch · CIFAR100 | error rate (%) | Hurt |
Closest papers by meaning
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.90)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.89)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.88)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.88)
- Unlocking the Potential of Unlabeled Data in Semi-Supervised Domain Generalization (CVPR 2025 · similarity 0.88)
- RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning (AAAI 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.