Research map / Continual learning / Class-incremental
MetaFSCIL: A Meta-Learning Approach for Few-Shot Class Incremental Learning
Check 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 |
|---|---|---|---|
| train with MetaFSCIL rather than CEC | ResNet18 · MiniImageNet, session 8 | top-1 accuracy: 47.63 -> 49.19 | Helped |
| train with MetaFSCIL rather than CEC | ResNet20 · CIFAR100, session 8 | top-1 accuracy: 49.14 -> 49.97 | Helped |
| train with MetaFSCIL rather than CEC | ResNet18 · CUB200, session 10 | top-1 accuracy: 52.28 -> 52.64 | Helped |
| add meta-learning to the updateable initialization baseline (ablation) | ResNet20 · CIFAR100, session 8 | top-1 accuracy: 43.80 -> 49.41 | Helped |
| add meta-learning to the fixed representation baseline (ablation) | ResNet20 · CIFAR100, session 8 | top-1 accuracy: 46.47 -> 49.30 | Helped |
| modulate only the last activation instead of leaving activations unmodulated (ablation) | meta-learned ResNet20 · CIFAR100, session 8 | top-1 accuracy: 49.41 -> 48.34 | Hurt |
| place modulation units uniformly throughout the classification network instead of only at the last activation (ablation) | meta-learned ResNet20 · CIFAR100, session 8 | top-1 accuracy: 48.34 -> 49.67 | Helped |
| add bi-directional guidance to uniform modulation (ablation) | ResNet20 · CIFAR100, session 8 | top-1 accuracy: 49.67 -> 49.97 | Helped |
| update meta-parameters after each incremental task rather than only at sequence end (ablation) | MetaFSCIL · CIFAR100, session 8 | top-1 accuracy: 49.02 -> 49.97 | Helped |
| replace the CEC backbone with the MetaFSCIL meta-trained backbone | CEC · MiniImageNet, final session | top-1 accuracy: 47.63 -> 48.95 | Helped |
| replace the CEC backbone with the MetaFSCIL meta-trained backbone | CEC · CIFAR100, final session | top-1 accuracy: 49.14 -> 49.71 | Helped |
| replace the CEC backbone with the MetaFSCIL meta-trained backbone | CEC · CUB200, final session | top-1 accuracy: 52.28 -> 52.64 | Helped |
Closest papers by meaning
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.90)
- Pseudo Informative Episode Construction for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.90)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.89)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · similarity 0.88)
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.88)
- Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental 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.