Research map / Continual learning / Online and replay-based
A continual learning survey: Defying forgetting in classification tasks
Matthias Delange, Rahaf Aljundi, Marc Masana, Sarah Parisot, Xu Jia, Aleš Leonardis, Greg Slabaugh, Tinne Tuytelaars
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 |
|---|---|---|---|
| use iCaRL instead of basic rehearsal with full memory | replay with 4.5k buffer · Tiny ImageNet (BASE) | average accuracy: 37.31 -> 47.27 | Improved |
| use GEM instead of basic rehearsal with partial memory | replay with 4.5k buffer · Tiny ImageNet (BASE) | average accuracy: 36.09 -> 45.13 | Improved |
| double replay buffer from 4.5k to 9k (ablation) | iCaRL · Tiny ImageNet (BASE) | average accuracy: 47.27 -> 48.76 | Improved |
| double replay buffer from 4.5k to 9k (ablation) | GEM · Tiny ImageNet (BASE) | average accuracy: 45.13 -> 41.75 | Hurt |
| add feature-autoencoder (EBLL) to LwF | LwF · Tiny ImageNet (BASE) | average accuracy: 41.91 -> 45.34 | Improved |
| use MAS instead of EWC | regularization-based CL · Tiny ImageNet (BASE) | average accuracy: 42.43 -> 46.90 | Improved |
| use DEEP model instead of BASE (ablation) | PackNet · Tiny ImageNet (random order) | average accuracy: 49.13 -> 35.54 | Hurt |
| use WIDE model instead of BASE (ablation) | EWC · Tiny ImageNet (random order) | average accuracy: 42.43 -> 31.10 | Hurt |
| use SMALL model instead of BASE (ablation) | EWC · Tiny ImageNet (random order) | average accuracy: 42.43 -> 45.13 | Improved |
| add dropout (p=0.5) (ablation) | finetuning · Tiny ImageNet (BASE) | average accuracy: 21.30 -> 29.23 | Improved |
| add dropout (p=0.5) (ablation) | PackNet · Tiny ImageNet (BASE) | average accuracy: 47.67 -> 54.28 | Improved |
| add dropout (p=0.5) (ablation) | SI · Tiny ImageNet (BASE) | average accuracy: 33.93 -> 43.15 | Improved |
Closest papers by meaning
- Probing Representation Forgetting in Supervised and Unsupervised Continual Learning (CVPR 2022 · similarity 0.93)
- A Comprehensive Survey of Continual Learning: Theory, Method and Application (TPAMI 2024 · similarity 0.93)
- CRNet: A Fast Continual Learning Framework With Random Theory (TPAMI 2023 · similarity 0.92)
- Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers (ICML 2025 · similarity 0.92)
- Optimal Task Order for Continual Learning of Multiple Tasks (ICML 2025 · similarity 0.90)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.90)
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.
