Research map / Continual learning / Class-incremental
Mamba-FSCIL: Dynamic Adaptation With Selective State Space Model for Few-Shot Class Incremental Learning
Xiaojie Li, Yibo Yang, Jianlong Wu, Yue Yu, Ming-Hsuan Yang, Liqiang Nie, Min Zhang
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 |
|---|---|---|---|
| replace the 3-layer MLP projector with a single selective SSM projector (ablation) | Mamba-FSCIL baseline · CIFAR-100 | AVG accuracy: 63.51 -> 66.90 | Improved |
| replace the 3-layer MLP projector with a single selective SSM projector (ablation) | Mamba-FSCIL baseline · CIFAR-100 | performance drop: 33.55 -> 27.57 | Improved |
| replace the 3-layer MLP projector with a single selective SSM projector (ablation) | Mamba-FSCIL baseline · miniImageNet | AVG accuracy: 63.09 -> 66.79 | Improved |
| split into dual (frozen base + incremental) selective SSM branches (ablation) | single selective SSM projector · miniImageNet | AVG accuracy: 66.79 -> 69.20 | Improved |
| split into dual (frozen base + incremental) selective SSM branches (ablation) | single selective SSM projector · miniImageNet | performance drop: 29.11 -> 26.01 | Improved |
| split into dual (frozen base + incremental) selective SSM branches (ablation) | single selective SSM projector · CIFAR-100 | AVG accuracy: 66.90 -> 67.52 | Improved |
| add class-sensitive selective scan (suppression + separation losses) (ablation) | dual selective SSM projector · miniImageNet | session 8 accuracy: 58.92 -> 59.36 | Improved |
| add class-sensitive selective scan (suppression + separation losses) (ablation) | dual selective SSM projector · CIFAR-100 | session 8 accuracy: 56.71 -> 57.51 | Improved |
| add class-sensitive selective scan (suppression + separation losses) (ablation) | dual selective SSM projector · CIFAR-100 | performance drop: 26.09 -> 25.29 | Improved |
| freeze identity and base SSM branches in incremental sessions (ablation) | Mamba-FSCIL (ResNet-12) · miniImageNet | accuracy | Improved |
| increase scan paths K from 1 to 4 (ablation) | Mamba-FSCIL · CIFAR-100 | session 8 accuracy: 55.94 -> 57.51 | Improved |
| increase scan paths K from 1 to 4 (ablation) | Mamba-FSCIL · CIFAR-100 | AVG accuracy: 67.23 -> 68.14 | Improved |
Later papers that compared with this method
1 later accepted paper reports Mamba-FSCIL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Consistency-Driven Calibration and Matching for Few-Shot Class Incremental Learning (ICLR 2026 · better on 4 of 4 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
- Few-Shot Incremental Learning With Continually Evolved Classifiers (CVPR 2021 · similarity 0.89)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.89)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.88)
- Pseudo Informative Episode Construction for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.87)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · similarity 0.86)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · 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.
