Research map / Continual learning / Class-incremental
Class-balanced knowledge retention
Preserves per-class knowledge and decision boundaries when new classes arrive.
73 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Class-incremental, keyword cluster “shot · incremental cil · feature”
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.
19 papers in 2023–24 and 28 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 20.7% to 20.7%.
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
- 2017 · Fast Online Incremental Learning on Mixture Streaming Data (AAAI 2017 · 10 citations)
- 2018 · Low-shot Learning via Covariance-Preserving Adversarial Augmentation Networks (NeurIPS 2018 · 30 citations)
- 2019 · Large Scale Incremental Learning (CVPR 2019 · 1,269 citations)
- 2020 · Maintaining Discrimination and Fairness in Class Incremental Learning (CVPR 2020 · 499 citations)
- 2021 · Prototype Augmentation and Self-Supervision for Incremental Learning (CVPR 2021 · 403 citations)
- 2022 · Not Just Selection, but Exploration: Online Class-Incremental Continual Learning via Dual View Consistency (CVPR 2022 · 73 citations)
- 2023 · PCR: Proxy-Based Contrastive Replay for Online Class-Incremental Continual Learning (CVPR 2023 · 75 citations)
- 2024 · Resurrecting Old Classes with New Data for Exemplar-Free Continual Learning (CVPR 2024 · 27 citations)
- 2025 · Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning (AAAI 2025 · 8 citations)
- 2026 · GOAL: Geometrically Optimal Alignment for Continual Generalized Category Discovery (AAAI 2026 · 1 citation)
Earliest papers here
- Fast Online Incremental Learning on Mixture Streaming Data (AAAI 2017 · 10 citations)
- Low-shot Learning via Covariance-Preserving Adversarial Augmentation Networks (NeurIPS 2018 · 30 citations)
- Large Scale Incremental Learning (CVPR 2019 · 1,269 citations)
What has been tried
50 papers here report 548 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 4%, hurt 19%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Increase perturbation iterations from 1 to 3 | CIFAR-100, T=10 · Alast | 45.96 -> 46.48 | Helped | Resurrecting Old Classes with New Data for Exemplar-Free Continual Learning (CVPR 2024 · 27 citations) |
| Add DKL | CUB · ACC (All) | 61.9 -> 62.5 | Helped | Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · 20 citations) |
| Add patch-level knowledge selection (PKS) | CIFAR100, 5/10/20 tasks · top-1 accuracy | 56.44 -> 58.27; 51.90 -> 56.82; 51.20 -> 52.87 | Helped | Fine-Grained Knowledge Selection and Restoration for Non-exemplar Class Incremental Learning (AAAI 2024 · 17 citations) |
| Use inter-class loss only rather than intra-class loss only | CIFAR-100 Split 5 Tasks · FAA | 17.03 -> 47.51 | Helped | Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning (AAAI 2025 · 8 citations) |
| Add soft entropy regularization and clustering-guided initialization | CUB, 5-stage average · New accuracy | 4.70 -> 32.91 | Helped | Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 2024 · 7 citations) |
| Use only one VGG11 layer for context detection instead of all layers | CIFAR10 and CIFAR100 · context detection accuracy | Hurt | NICE: Neurogenesis Inspired Contextual Encoding for Replay-free Class Incremental Learning (CVPR 2024 · 5 citations) | |
| Add prototypical replay (PR) | I2B2 FG-12-PG-1 · Macro-F1 (AT) | +17.77% | Helped | Prototypical Replay with Old-class Focusing Knowledge Distillation for Incremental Named Entity Recognition (AAAI 2025 · 2 citations) |
| add Latent Context Sharing (LCS) instead of optimizing latent contexts independently | CIFAR-10 · H | 72.37 (± 0.13) -> 95.07 (± 0.01) | Helped | Black-Box Forgetting (NeurIPS 2024 · 1 citation) |
| add distribution-based subset selection alongside error-based selection | Yelp, Reddit, and Amazon · AP | 0.0438 -> 0.0654; 0.0579 -> 0.0866; 0.1063 -> 0.1004 | Helped | A Selective Learning Method for Temporal Graph Continual Learning (ICML 2025 · 0 citations) |
| add stochastic variational inference instead of point-estimating class distributions | C100, TinyImageNet, IN100, and CUB, session 5 · All accuracy | 68.90 -> 81.23; 61.14 -> 74.52; 79.90 -> 87.32; 27.06 -> 66.44 | Helped | Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 2025 · 0 citations) |
| add graph-augmented inference using graph relations and generated descriptions | ImageNet-R B0-10 tasks · Accuracy | 82.77% -> 84.29% | Helped | Knowledge Graph Enhanced Generative Multi-modal Models for Class-Incremental Learning (NeurIPS 2025 · 0 citations) |
| add Feature Fusion | CIFAR-100 · A_avg (%) | 90.6 -> 91.1 | Helped | F-OAL: Forward-only Online Analytic Learning with Fast Training and Low Memory Footprint in Class Incremental Learning (NeurIPS 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 12 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 · 5 of 12 papers
- TinyImageNet · 5 of 12 papers
- ImageNet-Subset · 2 of 12 papers
- CUB200 · 2 of 12 papers
- ImageNet-R · 2 of 12 papers
Baselines
- LwF · 4 of 12 papers
- iCaRL · 4 of 12 papers
- PODNet · 2 of 12 papers
- EWC · 2 of 12 papers
- CODA-Prompt · 2 of 12 papers
- PASS · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 47 papers; from OpenAlex and the papers' affiliations).
- Peking University · 3 papers
- Zhejiang University · 2 papers
- Peng Cheng Laboratory · 2 papers
- Nankai University · 2 papers
- Chinese Academy of Sciences · 2 papers
- The Hong Kong Polytechnic University · 2 papers
Use it
Download BibTeX of 39 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
- Exemplar-efficient class-incremental learning (111 papers)
- Few-shot class-incremental prototypes (47 papers)
- Incremental object detection with replay (32 papers)
Elsewhere in the topic
- Unknown-class mining for incremental segmentation (23 papers)
- Distillation-based continual segmentation (17 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).