Research map / Continual learning / Class-incremental
DiAPR: Dimensionally-Allocated Prototype Refinement for Non-Exemplar Class Incremental Learning
Ruixuan Gao, Qijun Zhao, Keren Fu
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 DAP and GSA dimension-wise refinement instead of Gaussian noise augmentation (ablation) | ViT baseline · CIFAR-100 (5 tasks) | LAST accuracy: 60.78 -> 63.47 | Improved |
| add DAP and GSA instead of new feature injection (ablation) | ViT baseline · CIFAR-100 (10 tasks) | LAST accuracy: 60.78 -> 61.67 | Improved |
| add CDT on top of DAP&GSA (ablation) | ViT baseline · CIFAR-100 (5 tasks) | LAST accuracy: 63.47 -> 64.58 | Improved |
| add CDT on top of Gaussian augmentation (ablation) | ViT baseline · CIFAR-100 (5 tasks) | LAST accuracy: 60.78 -> 62.82 | Improved |
| add CDT on top of DAP&GSA (ablation) | ViT baseline · CIFAR-100 (10 tasks) | LAST accuracy: 61.67 -> 62.17 | Improved |
| use closed-form ridge regression instead of learnable layers for task transition (ablation) | CDT · CIFAR-100 | accuracy: +2.17% | Improved |
| use closed-form ridge regression instead of learnable layers for task transition (ablation) | CDT · TinyImageNet | accuracy: +0.38% | Improved |
| DiAPR vs best NECIL baseline | DiAPR (ViT) · CIFAR-100 (5 tasks) | AVG accuracy: 70.44 -> 72.79 | Improved |
| DiAPR vs best NECIL baseline | DiAPR (ViT) · TinyImageNet (20 tasks) | AVG accuracy: 52.50 -> 52.90 | Improved |
| lower average forgetting than baselines | DiAPR (ViT) · CIFAR-100 (5 tasks) | AF: 12.20 -> 10.53 | Improved |
Closest papers by meaning
- Self-Promoted Prototype Refinement for Few-Shot Class-Incremental Learning (CVPR 2021 · similarity 0.90)
- Prospective Representation Learning for Non-Exemplar Class-Incremental Learning (NeurIPS 2024 · similarity 0.89)
- Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · similarity 0.89)
- FCS: Feature Calibration and Separation for Non-Exemplar Class Incremental Learning (CVPR 2024 · similarity 0.87)
- Dual-Consistency Model Inversion for Non-Exemplar Class Incremental Learning (CVPR 2024 · similarity 0.87)
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · 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.
