Research map / Continual learning / Online and replay-based
A simple but strong baseline for online continual learning: Repeated Augmented Rehearsal
Yaqian Zhang, Bernhard Pfahringer, Eibe Frank, Albert Bifet, Nick Jin Sean Lim, Alvin Jia
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 repeated augmented rehearsal to ER | ER · Seq-CIFAR100, Seq-MiniImageNet, CORE50-NC, and CLRS25-NC (2K and 5K memory) | accuracy: +8.8, +10.0, +10.0, +13.5, +15.3, +17.2, +9.9, +9.7 | Improved |
| Add repeated augmented rehearsal to SCR | SCR · Seq-CIFAR100, Seq-MiniImageNet, CORE50-NC, and CLRS25-NC (2K and 5K memory) | accuracy: +5.1, +8.4, +5.7, +10.6, +8.3, +10.8, +14.9, +17.9 | Improved |
| Add repeated augmented rehearsal to MIR | MIR · Four OCL benchmarks, 2K and 5K memory | accuracy: +9.1, +10.4, +10.1, +12.6, +13.9, +17.7, +13.5, +14.0 | Improved |
| Add repeated augmented rehearsal to ASER | ASER · Four OCL benchmarks, 2K and 5K memory | accuracy: +7.2, +11.5, +11.3, +14.7, +7.8, +13.3, +9.3, +9.8 | Improved |
| Add repeated augmented rehearsal to DER | DER · Four OCL benchmarks, 2K and 5K memory | accuracy: +21.6, +32.8, +14.4, +23.2, +11.9, +18.6, +16.6, +14.8 | Improved |
| Tune RAR hyperparameters with RL rather than HTOCL | RAR · Seq-CIFAR100, Seq-Mini-ImageNet, CORE50-NC, and CLRS25-NC | accuracy: 23.4 -> 29.6; 26.0 -> 32.1; 40.8 -> 44.4; 26.9 -> 35.0 | Improved |
| Use augmentation alone without repeated rehearsal (ablation) | Online rehearsal · Three of four OCL datasets; CORE50-NC is the exception | accuracy | Mixed |
| Use repeated rehearsal alone without augmentation (ablation) | Online rehearsal · Three of four OCL datasets; CLRS is the exception | accuracy | Mixed |
| Increase repeated rehearsal to more than two iterations without augmentation (ablation) | Repeated rehearsal on CIFAR100 and MiniImageNet · Seq-CIFAR100 and Seq-MiniImageNet | accuracy | Hurt |
| Use RAR with up to 20 repeats (ablation) | RAR · OCL benchmarks | accuracy | No clear effect |
| Augment both memory and incoming batches rather than only one (ablation) | RAR · Four OCL benchmarks | accuracy | Improved |
| Augment incoming data only (ablation) | Repeated experience replay · CORE50-NC | accuracy: 7.7% | Improved |
Closest papers by meaning
- Rehearsal Revealed: The Limits and Merits of Revisiting Samples in Continual Learning (ICCV 2021 · similarity 0.92)
- Unlocking the Power of Rehearsal in Continual Learning: A Theoretical Perspective (ICML 2025 · similarity 0.92)
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.89)
- Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization (CVPR 2025 · similarity 0.88)
- Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation (CVPR 2024 · similarity 0.87)
- Online Task-Free Continual Learning via Dynamic Expansionable Memory Distribution (CVPR 2025 · 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.
