Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-ViM: Bidirectional State Space Model for Mitigating Label Imbalance in Semi-Supervised Learning
Hongyang He, Hongyang Xie, Haochen You, Victor Sanchez
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 |
|---|---|---|---|
| add SSMixup on top of LyapEMA with ARM pre-training (ablation) | Semi-ViM-Base · ImageNet-1K (10% labels) | top-1 accuracy (%): 80.52 -> 85.40 (vs ProbMixup) | Improved |
| replace EMA with LyapEMA under ARM pre-training (ablation) | Semi-ViM-Base · ImageNet-1K (1% labels) | top-1 accuracy (%): 72.53 -> 76.02 | Improved |
| replace EMA with LyapEMA under ARM pre-training (ablation) | Semi-ViM-Base · ImageNet-1K (10% labels) | top-1 accuracy (%): 79.90 -> 78.03 | Hurt |
| use ARM pre-training instead of MAE (ablation) | Semi-ViM-Base with LyapEMA+SSMixup · ImageNet-1K (10% labels) | top-1 accuracy (%): 76.15 -> 85.40 | Improved |
| use ARM pre-training instead of DINO-V2 (ablation) | Semi-ViM-Base with LyapEMA+SSMixup · ImageNet-1K (10% labels) | top-1 accuracy (%): 76.07 -> 85.40 | Improved |
| add self-supervised pre-training (ARM) before SSL (ablation) | Semi-ViM-Base with LyapEMA+SSMixup · ImageNet-1K (10% labels) | top-1 accuracy (%): 54.73 -> 85.40 | Improved |
| use LyapEMA instead of EMA-Teacher (ablation) | Semi-ViT-S · ImageNet-1K (1% labels) | top-1 accuracy (%): training failed -> 61.30 | Improved |
| use LyapEMA instead of EMA-Teacher (ablation) | Semi-ViM-S · ImageNet-1K (1% labels) | top-1 accuracy (%): training failed -> 80.30 | Improved |
| use LyapEMA instead of EMA-Teacher (ablation) | Semi-ViT-B · ImageNet-1K (10% labels) | top-1 accuracy (%): 78.15 -> 80.43 | Improved |
| use LyapEMA instead of EMA-Teacher (ablation) | FreeMatch (ResNet-50) · ImageNet-1K (10% labels) | top-1 accuracy (%): 74.50 -> 75.07 | Improved |
| use Semi-ViM-Base instead of Semi-ViT-Huge | SSL on ImageNet-LT · ImageNet-LT (10% labels) | top-1 accuracy (%): 73.50 -> 77.40 | Improved |
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.
- REACT — ReAct: Out-of-distribution Detection With Rectified Activations (NeurIPS 2021) · better on 1 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
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · similarity 0.88)
- BaCon: Boosting Imbalanced Semi-supervised Learning via Balanced Feature-Level Contrastive Learning (AAAI 2024 · similarity 0.87)
- Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model (ICML 2024 · similarity 0.86)
- Spectral State Space Model for Rotation-Invariant Visual Representation Learning (CVPR 2025 · similarity 0.85)
- LCGC: Learning from Consistency Gradient Conflicting for Class-Imbalanced Semi-Supervised Debiasing (AAAI 2025 · similarity 0.85)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · 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.
