Research map / Continual learning / Class-incremental
Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation
Yixiong Zou, Shanghang Zhang, Yuhua Li, Ruixuan Li
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 extra layer g(.) building PM feature from NM feature (no margins) (ablation) | CLOM baseline · CIFAR100 overall | accuracy: 47.02 -> 48.37 | Improved |
| add extra layer g(.) without margins (ablation) | CLOM baseline · CUB200 overall | accuracy: 57.78 -> 57.21 | Hurt |
| add extra layer g(.) without margins (ablation) | CLOM baseline · miniImageNet overall | accuracy: 46.58 -> 46.79 | No clear effect |
| add negative/positive margins to the g(.) architecture (ablation) | CLOM · CIFAR100 overall | accuracy: 48.37 -> 49.21 | Improved |
| add negative/positive margins to the g(.) architecture (ablation) | CLOM · CUB200 overall | accuracy: 57.21 -> 58.73 | Improved |
| add negative/positive margins to the g(.) architecture (ablation) | CLOM · miniImageNet overall | accuracy: 46.79 -> 47.30 | Improved |
| add class-relation-based margin mapping (ablation) | CLOM with margins · CIFAR100 overall | accuracy: 49.21 -> 50.25 | Improved |
| add class-relation-based margin mapping (ablation) | CLOM with margins · CUB200 overall | accuracy: 58.73 -> 59.58 | Improved |
| add class-relation-based margin mapping (ablation) | CLOM with margins · miniImageNet overall | accuracy: 47.30 -> 48.00 | Improved |
| apply margin directly to baseline with face-recognition adaptive margin AdaCos | baseline · CIFAR100 overall | accuracy: 47.02 -> 44.48 | Hurt |
| replace relation mapping with TRAML relational margin on NM/PM (ablation) | NM/PM architecture · CIFAR100 overall | accuracy: 50.25 -> 48.18 | Hurt |
| swap margin signs: positive margin on lower layer f, negative on higher layer F (ablation) | CLOM margins · CIFAR100 overall | accuracy: 49.60 -> 45.37 | Hurt |
Closest papers by meaning
- Few-Shot Incremental Learning With Continually Evolved Classifiers (CVPR 2021 · similarity 0.92)
- On the Approximation Risk of Few-Shot Class-Incremental Learning (ECCV 2024 · similarity 0.91)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.91)
- Few-Shot Class-Incremental Learning via Training-Free Prototype Calibration (NeurIPS 2023 · similarity 0.91)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.88)
- Pseudo Informative Episode Construction for Few-Shot 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.
