Research map / Continual learning / Class-incremental
Striking a Balance Between Stability and Plasticity for Class-Incremental Learning
Guile Wu, Shaogang Gong, Pan 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 reciprocal adaptive weights on Lce and Lem (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 36.4 -> 60.9 | Improved |
| use reciprocal adaptive weights instead of LUCIR adaptive weight (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 57.7 -> 60.9 | Improved |
| add class-independent learner with rotation prediction (SPB-I) (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 60.9 -> 62.6 | Improved |
| use contrastive learning as class-independent learner (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 60.9 -> 61.4 | Improved |
| add multi-perspective classifiers (SPB-M) (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 60.9 -> 65.5 | Improved |
| use multiple perspective classifiers instead of a single classifier (ablation) | SPB-M · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 58.2 -> 65.5 | Improved |
| compute Lce on additionally rotated samples (ablation) | SPB-I (Rotation) · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 62.6 -> 59.6 | Hurt |
| compute Lce on additionally transformed samples (ablation) | SPB-I (Contrastive) · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 61.4 -> 60.2 | Hurt |
| enlarge training set with rotation augmentation (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 60.9 -> 58.2 | Hurt |
| enlarge training set with strong augmentation (ablation) | SPB · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 60.9 -> 60.7 | No clear effect |
| use self-supervised pre-training as a separate stage instead of joint optimization (ablation) | SPB-I (rotation) · CIFAR-100 (K=6) | average incremental accuracy (top-1, %): 62.6 -> 61.1 | Hurt |
| increase number of rotation transformations (beta, gamma) (ablation) | SPB-I and SPB-M · CIFAR-100 (K=6) | average incremental accuracy (top-1, %) | Improved |
Closest papers by meaning
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · similarity 0.90)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.89)
- Harnessing Neural Unit Dynamics for Effective and Scalable Class-Incremental Learning (ICML 2024 · similarity 0.89)
- Knowledge Memorization and Rumination for Pre-trained Model-based Class-Incremental Learning (CVPR 2025 · similarity 0.89)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.89)
- Defying Imbalanced Forgetting in Class Incremental Learning (AAAI 2024 · 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.
