Research map / Continual learning / Domain and prompt-based
Prompt-based continual learning
Learns small prompt sets over frozen pretrained models without rehearsal buffers.
26 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2025) · an approach in Domain and prompt-based, keyword cluster “prompts · person · prompting”
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.
14 papers in 2023–24 and 8 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 31.8% to 16.3%.
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
- 2021 · Towards a robust experimental framework and benchmark for lifelong language learning (NeurIPS 2021 · 1 citation)
- 2022 · Learning To Prompt for Continual Learning (CVPR 2022 · 762 citations)
- 2023 · CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning (CVPR 2023 · 320 citations)
- 2024 · Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · 39 citations)
- 2025 · Adaptive Prompting for Continual Relation Extraction: A Within-Task Variance Perspective (AAAI 2025 · 3 citations)
Earliest papers here
- Towards a robust experimental framework and benchmark for lifelong language learning (NeurIPS 2021 · 1 citation)
- Learning To Prompt for Continual Learning (CVPR 2022 · 762 citations)
- S-Prompts Learning with Pre-trained Transformers: An Occam’s Razor for Domain Incremental Learning (NeurIPS 2022 · 58 citations)
What has been tried
23 papers here report 268 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 4%, hurt 19%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add CPM to the model without CPM or SCL | MVTec AD · image AUROC | 0.693 -> 0.894 | Helped | Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · 39 citations) |
| Add CCL | Split StanfordCars 10-task · Last-acc | 61.96 -> 62.52 | Helped | Consistent Prompting for Rehearsal-Free Continual Learning (CVPR 2024 · 38 citations) |
| add FFN prompt memory (MLP bottleneck prompts) | Split ImageNet-R 10 steps · Avg Acc | +18.15 over FT-seq (50.28 to 68.43) | Helped | Evolving Parameterized Prompt Memory for Continual Learning (AAAI 2024 · 18 citations) |
| use 100 prompts instead of 150 | ADE20K panoptic 100-10 · PQ all | 33.9 -> 33.7 | No clear effect | ECLIPSE: Efficient Continual Learning in Panoptic Segmentation with Visual Prompt Tuning (CVPR 2024 · 17 citations) |
| add task IDs for threshold selection | Split CIFAR100 (10 tasks) · AUC_N | 91.01 -> 92.69 | Helped | Learning to Prompt Knowledge Transfer for Open-World Continual Learning (AAAI 2024 · 16 citations) |
| add affinity projection B₁ | four class-incremental benchmarks · accuracy | 3.31%~9.03% improvement | Helped | Visual Prompt Tuning in Null Space for Continual Learning (NeurIPS 2024 · 6 citations) |
| add a task-specific prompt pool rather than use a single task-specific prompt | TACRED task-incremental learning, T10 · average accuracy (%) | 83.4 -> 85.2 | Helped | Adaptive Prompting for Continual Relation Extraction: A Within-Task Variance Perspective (AAAI 2025 · 3 citations) |
| use text-only rather than flow+text prompt query | Lighting L3 (warm→cold→dark) · success rate | 0.61 -> 0.80 | Helped | Think Small, Act Big: Primitive Prompt Learning for Lifelong Robot Manipulation (CVPR 2025 · 3 citations) |
| Add object-injected state prompting | Split-Clothing (5 tasks) · Avg Acc | 88.45±0.10 -> 89.21±0.24 | Helped | Not Just Object, But State: Compositional Incremental Learning without Forgetting (NeurIPS 2024 · 2 citations) |
| Add TRP-AVA | AVE protocol-3 · Average Acc. | 38.63 -> 55.74 | Helped | Few-Shot Audio-Visual Class-Incremental Learning with Temporal Prompting and Regularization (AAAI 2025 · 1 citation) |
| Add ALD | Split ImageNet-R · iteration time | 270 -> 240 | Helped | REP: Resource-Efficient Prompting for Rehearsal-Free Continual Learning (NeurIPS 2025 · 1 citation) |
| Add LCDC (global class distribution estimation and local class distribution compensation) to the baseline | ImageNet-R · Avg | 1.88% | Helped | C$^2$Prompt: Class-aware Client Knowledge Interaction for Federated Continual Learning (NeurIPS 2025 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ImageNet-R · 4 of 8 papers
- CIFAR-100 · 3 of 8 papers
- CUB200 · 2 of 8 papers
- DomainNet · 2 of 8 papers
Baselines
- L2P · 5 of 8 papers
- HiDe-Prompt · 3 of 8 papers
- DualPrompt · 3 of 8 papers
- CODA-Prompt · 3 of 8 papers
- ConvPrompt · 2 of 8 papers
- CPrompt · 2 of 8 papers
- InfLoRA · 2 of 8 papers
- LWF · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 22 papers; from OpenAlex and the papers' affiliations).
- Chinese Academy of Sciences · 3 papers
- Peking University · 3 papers
- Shanghai Jiao Tong University · 2 papers
- Hong Kong University of Science and Technology · 2 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- Northwestern Polytechnical University · 2 papers
- Amazon · 2 papers
Use it
Download BibTeX of 26 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
- Lifelong person re-identification (14 papers)
Elsewhere in the topic
- Domain-incremental learning with memory (49 papers)
- Continual test-time adaptation (25 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).