Research map / Continual learning / Class-incremental
Few-Shot Incremental Learning With Continually Evolved Classifiers
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 |
|---|---|---|---|
| Add continual classifier adaptation trained with pseudo incremental learning (ablation) | Decoupled cosine classifier with data initialization · CUB200 | session 10 accuracy: 49.31 -> 52.28 | Helped |
| Train the adaptation module with pseudo incremental learning instead of meta-learning (ablation) | CEC · CUB200 | performance dropping rate: 26.45 -> 23.57 | Helped |
| Add the adaptation module without pseudo incremental learning (ablation) | Decoupled cosine classifier with data initialization · CUB200 | session 10 accuracy: 49.31 -> 49.15 | No clear effect |
| Decouple representation learning from classifier learning (ablation) | Cosine classifier · CUB200 | PD: 72.80 -> 44.00 | Helped |
| Decouple representation learning from classifier learning (ablation) | Linear classifier · CUB200 | PD: 70.49 -> 67.10 | Helped |
| Initialize classifier weights with average training-data embeddings (ablation) | Classifiers · CUB200 | session 10 accuracy | Helped |
| Use a cosine classifier rather than linear, L2, or DeepEMD classifiers (ablation) | Decoupled classifier with data initialization · CUB200 | session 10 accuracy: 32.76, 41.62, 47.60 -> 49.31 | Helped |
| Use a decoupled cosine classifier instead of a non-decoupled cosine classifier (ablation) | Cosine classifier · CUB200 | session 10 accuracy: 0.52 -> 30.36 | Helped |
| Use a decoupled linear classifier instead of a non-decoupled linear classifier (ablation) | Linear classifier · CUB200 | session 10 accuracy: 0.53 -> 3.92 | Helped |
| Use larger way and fewer shots when constructing pseudo incremental episodes (ablation) | CEC pseudo incremental learning · CUB200 | last-session accuracy | Helped |
| Synthesize pseudo classes with large rotations rather than small rotations (ablation) | CEC pseudo incremental learning · CUB200 | average accuracy: 61.33% | Helped |
| Use random 90°, 180°, and 270° rotations to synthesize pseudo classes (ablation) | CEC pseudo incremental learning · CUB200 | PD: 23.57% | Helped |
Closest papers by meaning
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.94)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.93)
- Pseudo Informative Episode Construction for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.91)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · similarity 0.89)
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.89)
- Diffusion-Classifier Synergy: Reward-Aligned Learning via Mutual Boosting Loop for FSCIL (NeurIPS 2025 · similarity 0.88)
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.