Research map / Continual learning / Online and replay-based
Plug-and-play online training enhancements
Adds distillation, collaborative learners or state space modules to improve plasticity of online learners.
39 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in Online and replay-based, keyword cluster “online continual · replay · experience”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
13 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 8.0% to 5.6%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Task-Free Continual Learning (CVPR 2019 · 327 citations)
- 2020 · Dark Experience for General Continual Learning: a Strong, Simple Baseline (NeurIPS 2020 · 90 citations)
- 2021 · Contextual Transformation Networks for Online Continual Learning (ICLR 2021 · 20 citations)
- 2022 · GCR: Gradient Coreset Based Replay Buffer Selection for Continual Learning (CVPR 2022 · 111 citations)
- 2023 · Real-Time Evaluation in Online Continual Learning: A New Hope (CVPR 2023 · 35 citations)
- 2024 · Pre-trained Online Contrastive Learning for Insurance Fraud Detection (AAAI 2024 · 22 citations)
- 2025 · Salient Frequency-aware Exemplar Compression for Resource-constrained Online Continual Learning (AAAI 2025 · 1 citation)
- 2026 · MePo: Meta Post-Refinement for Rehearsal-Free General Continual Learning (ICML 2026 · 0 citations)
Earliest papers here
- Task-Free Continual Learning (CVPR 2019 · 327 citations)
- Online Continual Learning with Maximal Interfered Retrieval (NeurIPS 2019 · 202 citations)
- Scalable Recollections for Continual Lifelong Learning (AAAI 2019 · 53 citations)
What has been tried
30 papers here report 340 measured changes (a component added, removed or swapped, compared with a baseline): helped 79%, no clear effect 4%, hurt 14%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove Temporal MAS and directly update parameters | Medical fraud dataset · Accuracy | Hurt | Pre-trained Online Contrastive Learning for Insurance Fraud Detection (AAAI 2024 · 22 citations) | |
| Add multi-level supervision with CE and SCL | Split CIFAR-100, M=5k · ACC (%) | 45.00 -> 52.29 | Helped | Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation (CVPR 2024 · 18 citations) |
| Add multiview classification training without CCL-DC | CIFAR-100, M=2000 · average accuracy | 31.89±1.45 -> 38.18±1.46 | Helped | Improving Plasticity in Online Continual Learning via Collaborative Learning (CVPR 2024 · 8 citations) |
| add PFI | CIFAR100 (M=2K), disjoint · ACC | 40.85± 1.68 -> 45.27± 1.32 | Helped | Doubly Perturbed Task Free Continual Learning (AAAI 2024 · 3 citations) |
| Use student weights alone for inference rather than averaging student and teacher weights | CIFAR100 clear, M=1000 · final average accuracy (%) | 38.5{ ±0.5} -> 37.7{ ±0.7} | Hurt | Rethinking Momentum Knowledge Distillation in Online Continual Learning (ICML 2024 · 2 citations) |
| use JPEG compression without saliency-aware quantization or compression-aware buffer management | CIFAR-100, M=2000 · final average accuracy | 21.74 ± 0.4 -> 19.23 ± 1.0 | Hurt | Salient Frequency-aware Exemplar Compression for Resource-constrained Online Continual Learning (AAAI 2025 · 1 citation) |
| Add an SSM branch and LDiff | CIFAR-100, M=2000 · Average accuracy (%) | 33.33±0.93 -> 36.61±1.04 | Helped | Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization (CVPR 2025 · 1 citation) |
| Add adaptive layer freezing | CIFAR-10 Gaussian · TFLOPs | 163.74 -> 142.23 | Helped | Budgeted Online Continual Learning by Adaptive Layer Freezing and Frequency-based Sampling (ICLR 2025 · 1 citation) |
| add label-specific cut-and-replay to experience replay | PASCAL VOC 2007 · Avg mAP | 75.05±1.28 -> 77.92±0.78 | Helped | Cut out and Replay: A Simple yet Versatile Strategy for Multi-Label Online Continual Learning (ICML 2025 · 0 citations) |
| Increase model sparsity by using a larger Λ | Continual Bench · performance | Helped | Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · 0 citations) | |
| Add ARC-R | Split CIFAR-100 and Split Imagenet-R · Average Accuracy | 1.8% | Helped | Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · 0 citations) |
| Add Entropy Selection to the memory strategy | C100/SDXL (P=70%) · accuracy | 44.84{ ±0.80} -> 45.92{ ±0.42} | Helped | Dealing with Synthetic Data Contamination in Online Continual Learning (NeurIPS 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-100 · 3 of 9 papers
- CIFAR-10 · 2 of 9 papers
Baselines
- MIR · 2 of 9 papers
- GSS · 2 of 9 papers
- ER · 2 of 9 papers
- iCarl · 2 of 9 papers
- EWC · 2 of 9 papers
- Memo · 2 of 9 papers
- DER · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 24 papers; from OpenAlex and the papers' affiliations).
- Seoul National University · 3 papers
- University of Oxford · 3 papers
- The University of Tokyo · 3 papers
- King Abdullah University of Science and Technology · 2 papers
- Amazon · 2 papers
- Ecole Supérieure d'Ingénieurs en Electronique et Electrotechnique · 2 papers
- Nanjing University of Aeronautics and Astronautics · 2 papers
Use it
Download BibTeX of 35 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Memory-based and Bayesian online learning (56 papers)
- Replay buffer construction and update (33 papers)
Elsewhere in the topic
- Theoretical analysis of forgetting (141 papers)
- Unified and unsupervised continual frameworks (120 papers)
- Gradient projection and plasticity preservation (80 papers)
- Dynamic expansion and isolation (70 papers)
- Plasticity-preserving continual reinforcement learning (43 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).