Research map / Continual learning / Class-incremental
Few-Shot Class-Incremental Learning via Class-Aware Bilateral Distillation
Read the paperCheck 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 |
|---|---|---|---|
| Replace previous-novel-only distillation with bilateral guidance from the base and previous novel branches (ablation) | Novel branch · mini-ImageNet, final session (8) | accuracy: 45.33 -> 51.47 | Helped |
| Add previous-novel-branch guidance to base-branch distillation (ablation) | Novel branch · mini-ImageNet, final session (8) | accuracy: 50.83 -> 51.47 | Helped |
| Add attention-based prediction aggregation to the novel branch (ablation) | Two-branch model · mini-ImageNet, final session (8) | accuracy: 51.47 -> 52.22 | Helped |
| Replace attention aggregation with concatenated branch features (ablation) | Two-branch model · mini-ImageNet, final session (8) | accuracy: 52.22 -> 51.58 | Hurt |
| Replace attention aggregation with base-branch-prediction routing (ablation) | Two-branch model · mini-ImageNet, final session (8) | accuracy: 52.22 -> 49.39 | Hurt |
| Replace attention aggregation with novel-branch-prediction routing (ablation) | Two-branch model · mini-ImageNet, final session (8) | accuracy: 52.22 -> 51.81 | Hurt |
| Replace attention aggregation with equal averaging of branch predictions (ablation) | Two-branch model · mini-ImageNet, final session (8) | accuracy: 52.22 -> 51.28 | Hurt |
| Train conv5_x rather than freezing the entire novel branch (ablation) | Novel branch · mini-ImageNet, final session (8) | accuracy: 49.96 -> 52.22 | Helped |
| Train the full backbone rather than only conv5_x (ablation) | Novel branch · mini-ImageNet, final session (8) | accuracy: 52.22 -> 51.34 | Hurt |
| Increase the exemplar memory from 1 to 5 examples per class (ablation) | Proposed method · mini-ImageNet, final session (8) | accuracy: 52.22 -> 53.39 | Helped |
| Remove the exemplar memory (ablation) | Proposed method · mini-ImageNet, final session (8) | accuracy: 52.22 -> 50.49 | Hurt |
Closest papers by meaning
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.93)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.92)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · similarity 0.91)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.91)
- Pseudo Informative Episode Construction for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.89)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.89)
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.