Research map / Continual learning / Class-incremental
On the Stability-Plasticity Dilemma of Class-Incremental Learning
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 |
|---|---|---|---|
| Apply DER only to Layer 4 rather than maintaining full feature extractors | DER on ResNet-18 · ImageNet-1K B500-5step | retrained-classifier accuracy at stage 5: 66.9 -> 68.4 | Helped |
| Apply DER only to Layer 4 rather than maintaining full feature extractors | DER on ResNet-18 · ImageNet-1K B500-5step | GMACs (F5 forward pass): 10.9 -> 3.9 | Helped |
| Freeze the base feature extractor and train separate classifiers for each class increment | ImageNet-1K class-incremental learning · ImageNet-1K B500-5step | average incremental accuracy: 67.2 | Helped |
| Freeze the base feature extractor and train separate classifiers for each class increment | ImageNet-1K class-incremental learning · ImageNet-1K B500-5step | training epochs per incremental stage: 90 to 120 -> around 10 | Helped |
| Use pDER Layer 4 instead of pDER Layer 2 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | retrained-classifier accuracy at stage 5: 67.8 -> 68.4 | Helped |
| Use pDER Layer 4 instead of pDER Layer 2 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | average incremental accuracy: 69.2 -> 69.7 | Helped |
| Use pDER Layer 4 instead of pDER Layer 2 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | final accuracy: 64.1 -> 64.7 | Helped |
| Use pDER Layer 4 instead of pDER Layer 3 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | retrained-classifier accuracy at stage 5: 68.0 -> 68.4 | Helped |
| Use pDER Layer 4 instead of pDER Layer 3 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | average incremental accuracy: 69.4 -> 69.7 | Helped |
| Use pDER Layer 4 instead of pDER Layer 3 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | final accuracy: 64.4 -> 64.7 | Helped |
| Use pDER Layer 4 instead of pDER Layer 2 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | GMACs (F5 forward pass): 8.0 -> 3.9 | Helped |
| Use pDER Layer 4 instead of pDER Layer 3 (ablation) | pDER on ResNet-18 · ImageNet-1K B500-5step | GMACs (F5 forward pass): 5.9 -> 3.9 | Helped |
Closest papers by meaning
- Bring Evanescent Representations to Life in Lifelong Class Incremental Learning (CVPR 2022 · similarity 0.90)
- On the Power of Statistics in Class-Incremental Learning with Pretrained Models (ICML 2026 · similarity 0.89)
- General Incremental Learning With Domain-Aware Categorical Representations (CVPR 2022 · similarity 0.89)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.88)
- Compress to One Point: Neural Collapse for Pre-Trained Model-Based Class-Incremental Learning (AAAI 2025 · similarity 0.87)
- AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection (NeurIPS 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.