Research map / Continual learning / Class-incremental
Prototype Reminiscence and Augmented Asymmetric Knowledge Aggregation for Non-Exemplar Class-Incremental Learning
Wuxuan Shi, Mang Ye
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 prototype reminiscence (interpolate/extrapolate stored prototypes with new features) to the prototype-replay baseline (ablation) | NECIL baseline with stored prototypes · CIFAR-100 / TinyImageNet | average incremental accuracy | Improved |
| add self-supervised label augmentation alone (ablation) | NECIL baseline · CIFAR-100 / TinyImageNet | average incremental accuracy | Mixed |
| add asymmetric knowledge aggregation on top of SLA (ablation) | PRAKA · CIFAR-100 (initial vs new tasks) | new-task accuracy | Improved |
| add asymmetric knowledge aggregation (ablation) | PRAKA · CIFAR-100 (5 phases) | average forgetting: 7.18 -> 12.59 | Hurt |
| raise prototype reminiscence threshold eta from low values to about 0.6 (ablation) | PRAKA · CIFAR-100 | average incremental accuracy | Improved |
| raise threshold eta beyond 0.6 (ablation) | PRAKA · CIFAR-100 | average incremental accuracy | Hurt |
| use prototype reminiscence instead of Gaussian noise augmentation of prototypes (ablation) | old-class feature generation · CIFAR-100 (t-SNE) | old class feature separation | Improved |
| use PRAKA instead of SSRE | NECIL on CIFAR-100 · CIFAR-100 (5 phases) | average incremental accuracy: 65.88 -> 70.02 | Improved |
| use PRAKA without exemplars vs exemplar-based DER | PRAKA · ImageNet-Subset (10 phases) | average incremental accuracy: 68.98 vs 75.36 (DER) | Hurt |
Closest papers by meaning
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.90)
- Prospective Representation Learning for Non-Exemplar Class-Incremental Learning (NeurIPS 2024 · similarity 0.88)
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · similarity 0.87)
- Adaptive Discovering and Merging for Incremental Novel Class Discovery (AAAI 2024 · similarity 0.87)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.87)
- Knowledge Memorization and Rumination for Pre-trained Model-based Class-Incremental Learning (CVPR 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.
