Research map / Continual learning / Class-incremental
NAPA-VQ: Neighborhood-Aware Prototype Augmentation with Vector Quantization for Continual Learning
Tamasha Malepathirana, Damith Senanayake, Saman Halgamuge
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 |
|---|---|---|---|
| replace categorical cross-entropy with distance-based cross-entropy (DCE) (ablation) | KD baseline · CIFAR-100 (T=10) | Average accuracy: 16.70 -> 32.90 | Improved |
| add neighborhood-adaptation loss L_NA to DCE and KD (ablation) | KD + DCE · CIFAR-100 (T=20) | Average forgetting: 81.71 -> 30.09 | Improved |
| add neighborhood-adaptation loss L_NA to DCE and KD (ablation) | KD + DCE · CIFAR-100 (T=5) | Average accuracy: 49.83 -> 49.31 | No clear effect |
| add Gaussian-noise prototype augmentation to NA-VQ (ablation) | KD + NA-VQ · CIFAR-100 (T=20) | Average accuracy: 42.13 -> 62.39 | Improved |
| replace Gaussian prototype augmentation with neighborhood-aware NA-PA (ablation) | NAPA-VQ · CIFAR-100 (T=20) | Average accuracy: 62.39 -> 67.42 | Improved |
| replace Gaussian prototype augmentation with neighborhood-aware NA-PA (ablation) | NAPA-VQ · CIFAR-100 (T=20) | Average forgetting: 18.30 -> 9.08 | Improved |
| increase connectivity factor K (2 to 50) (ablation) | NA-VQ · CIFAR-100 | accuracy and running time | Mixed |
| NAPA-VQ vs SSRE on ImageNet-Subset (T=10) forgetting | NAPA-VQ · ImageNet-Subset (T=10) | Average forgetting: 8.30 -> 9.67 | Hurt |
| NAPA-VQ vs SOTA NECIL on accuracy | NAPA-VQ · CIFAR-100, TinyImageNet, ImageNet-Subset | Average accuracy: avg +5%, +2%, +4% | Improved |
Later papers that compared with this method
1 later accepted paper reports NAPA-VQ as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Dual-Estimator: Decoupling Global and Local Semantic Shift for Drift Compensation in Class-Incremental Learning (CVPR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Prototype Augmentation and Self-Supervision for Incremental Learning (CVPR 2021 · similarity 0.88)
- Incremental few-shot learning via vector quantization in deep embedded space (ICLR 2021 · similarity 0.88)
- Prototype Reminiscence and Augmented Asymmetric Knowledge Aggregation for Non-Exemplar Class-Incremental Learning (ICCV 2023 · similarity 0.87)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.87)
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · similarity 0.87)
- Prospective Representation Learning for Non-Exemplar Class-Incremental Learning (NeurIPS 2024 · 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.
