Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 56 of the 56 references it cites were found in a registry or an index.

Three types of incremental learning

Open access
van de Ven, Gido M., Tuytelaars, Tinne, Tolias, Andreas S.·2022·Nature Machine Intelligence

Abstract

Incrementally learning new information from a non-stationary stream of data, referred to as ‘continual learning’, is a key feature of natural intelligence, but a challenging problem for deep neural networks. In recent years, numerous deep learning methods for continual learning have been proposed, but comparing their performances is difficult due to the lack of a common framework. To help address this, we describe three fundamental types, or ‘scenarios’, of continual learning: task-incremental, domain-incremental and class-incremental learning. Each of these scenarios has its own set of challenges. To illustrate this, we provide a comprehensive empirical comparison of currently used continual learning strategies, by performing the Split MNIST and Split CIFAR-100 protocols according to each scenario. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of the effectiveness of different strategies. The proposed categorization aims to structure the continual learning field, by forming a key foundation for clearly defining benchmark problems.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s42256-022-00568-3

Cites

56 of 56 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Decision tree for the three continual learning scenarios. The scenarios can be defined based on whether at test time context identity is known and if it is not, whether it must be inferred.
Fig. 1 Decision tree for the three continual learning scenarios. The scenarios can be defined based on whether at test time context identity is known and if it is not, whether it must be inferred.CC BY · PubMed Central
Fig. 2 Split MNIST according to the three scenarios. a, The Split MNIST protocol is obtained by splitting the original MNIST dataset into five contexts, with each context consisting of two digits. b, Overview of what is expected of the algorithm at test time when the Split MNIST protocol is performed according to each continual learning scenario. aWith task-incremental learning, at the computational level, there is no difference between whether the algorithm must return the within-context label or the global label, because the within-context label can be combined with the context label (which is provided as input) to get the global label.
Fig. 2 Split MNIST according to the three scenarios. a, The Split MNIST protocol is obtained by splitting the original MNIST dataset into five contexts, with each context consisting of two digits. b, Overview of what is expected of the algorithm at test time when the Split MNIST protocol is performed according to each continual learning scenario. aWith task-incremental learning, at the computational level, there is no difference between whether the algorithm must return the within-context label or the global label, because the within-context label can be combined with the context label (which is provided as input) to get the global label.CC BY · PubMed Central
Fig. 3 Schematic illustrations of different continual learning strategies. a, Context-specific components. Certain parts of the network are only used for specific contexts. b, Parameter regularization. Parameters important for past contexts are encouraged not to change too much when learning new contexts. c, Functional regularization. The input–output mapping learned previously is encouraged not to change too much at a particular set of inputs (the ‘anchor points’) when training on new contexts. d, Replay. The training data of a new context is complemented with data representative of past context. The replayed data is sampled from M, which can be a memory buffer or a generative model. e, Template-based classification. A ‘template’ is learned for each class (for example, a prototype, an energy value or a generative model), and classification is performed based on which template is most suitable for the sample to be classified. See Supplementary Note 4 for a detailed discussion of these strategies.
Fig. 3 Schematic illustrations of different continual learning strategies. a, Context-specific components. Certain parts of the network are only used for specific contexts. b, Parameter regularization. Parameters important for past contexts are encouraged not to change too much when learning new contexts. c, Functional regularization. The input–output mapping learned previously is encouraged not to change too much at a particular set of inputs (the ‘anchor points’) when training on new contexts. d, Replay. The training data of a new context is complemented with data representative of past context. The replayed data is sampled from M, which can be a memory buffer or a generative model. e, Template-based classification. A ‘template’ is learned for each class (for example, a prototype, an energy value or a generative model), and classification is performed based on which template is most suitable for the sample to be classified. See Supplementary Note 4 for a detailed discussion of these strategies.CC BY · PubMed Central
Extended Data Fig. 1 Comparison of methods using stored data with different buffer sizes. Shown is the average test accuracy (as %, over all contexts) of different methods that use stored data on Split MNIST (a) and on Split CIFAR-100 (b) as a function of the number of examples per class that is allowed to be stored in memory. Due to its high computational costs, we were not able to run FROMP on Split CIFAR-100, or on Split MNIST with a memory budget above 1,000 samples per class. Displayed are the means over 5 repetitions, shaded areas are ± 1 SEM. ER: exact replay, A-GEM: averaged gradient episodic memory, FROMP: functional regularization of the memorable past, iCaRL: incremental classifier and representation learning, None: sequential training in the standard way, Joint: training on all data at the same time.
Extended Data Fig. 1 Comparison of methods using stored data with different buffer sizes. Shown is the average test accuracy (as %, over all contexts) of different methods that use stored data on Split MNIST (a) and on Split CIFAR-100 (b) as a function of the number of examples per class that is allowed to be stored in memory. Due to its high computational costs, we were not able to run FROMP on Split CIFAR-100, or on Split MNIST with a memory budget above 1,000 samples per class. Displayed are the means over 5 repetitions, shaded areas are ± 1 SEM. ER: exact replay, A-GEM: averaged gradient episodic memory, FROMP: functional regularization of the memorable past, iCaRL: incremental classifier and representation learning, None: sequential training in the standard way, Joint: training on all data at the same time.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 43 references

Research works printed but not linked: 19

  1. Chen, Z. & Liu, B. Lifelong machine learning. Synth. Lect. Artif. Intell. Mach. Learn. 12, 1–207 (2018).
  2. Mundt, M., Lang, S., Delfosse, Q. & Kersting, K. CLEVA-compass: A continual learning evaluation assessment compass to promote research transparency and comparability. In International Conference on Learning Representations (2022).
  3. Ruvolo, P. & Eaton, E. ELLA: An efficient lifelong learning algorithm. In International Conference on Machine Learning 507–515 (PMLR, 2013).
  4. Ramesh, R. & Chaudhari, P. Model Zoo: A growing brain that learns continually. In International Conference on Learning Representations (2022).
  5. Lopez-Paz, D. & Ranzato, M. Gradient episodic memory for continual learning. In Advances in Neural Information Processing Systems Vol. 30, 6470–6479 (2017).
  6. Mirza, M. J., Masana, M., Possegger, H. & Bischof, H. An efficient domain-incremental learning approach to drive in all weather conditions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops 3001–3011 (2022).
  7. von Oswald, J., Henning, C., Sacramento, J. & Grewe, B. F. Continual learning with hypernetworks. In International Conference on Learning Representations (2020).
  8. Lomonaco, V. & Maltoni, D. Core50: a new dataset and benchmark for continuous object recognition. In Conference on Robot Learning 17–26 (PMLR, 2017).
  9. Shin, H., Lee, J. K., Kim, J. & Kim, J. Continual learning with deep generative replay. In Advances in Neural Information Processing Systems Vol. 30, 2994–3003 (2017).
  10. Zeno, C., Golan, I., Hoffer, E. & Soudry, D. Task agnostic continual learning using online variational bayes. Preprint at https://arxiv.org/abs/1803.10123v3 (2019). · arXiv
  11. Caccia, L., Belilovsky, E., Caccia, M. & Pineau, J. Online learned continual compression with adaptive quantization modules. In International Conference on Machine Learning 1240–1250 (PMLR, 2020).
  12. Lee, S., Ha, J., Zhang, D. & Kim, G.A neural dirichlet process mixture model for task-free continual learning. In International Conference on Learning Representations (2020).
  13. Jin, X., Sadhu, A., Du, J. & Ren, X. Gradient-based editing of memory examples for online task-free continual learning. In Advances in Neural Information Processing Systems Vol. 34, 29193–29205 (2021).
  14. Rolnick, D., Ahuja, A., Schwarz, J., Lillicrap, T. & Wayne, G. Experience replay for continual learning. In Advances in Neural Information Processing Systems Vol. 32 (2019).
  15. Chaudhry, A., Ranzato, M., Rohrbach, M. & Elhoseiny, M. Efficient lifelong learning with a-gem. In International Conference on Learning Representations (2019).
  16. Aljundi, R. et al. Online continual learning with maximally interfered retrieval. In Advances in Neural Information Processing Systems Vol. 32 (2019). · DOI · Paper page
  17. Hess, T., Mundt, M., Pliushch, I. & Ramesh, V. A procedural world generation framework for systematic evaluation of continual learning. In Thirty-fifth Conference on Neural Information Processing Systems, Datasets and Benchmarks Track (2021).
  18. Ioffe, S. & Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning 448–456 (PMLR, 2015).
  19. Rezende, D. & Mohamed, S. Variational inference with normalizing flows. In International Conference on Machine Learning 1530–1538 (PMLR, 2015).

Cited by

See all 769

Metadata

Authors
van de Ven, G. M., Tuytelaars, T., & Tolias, A. S.
Affiliations
  • Baylor College of Medicine (US)
  • KU Leuven (BE)
Corresponding author
van de Ven, Gido M.
Venue
Nature Machine Intelligence
Published
2022-12-05
Volume
4
Issue
12
Pages
1185-1197
ISSN
2522-5839 (ISSN-L)
OpenAlex
W4312210066
Cited by
665
References
43
Citations by year
2026: 173 · 2025: 258 · 2024: 172 · 2023: 60 · 2022: 2
FWCI
40.48
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Machine Intelligence (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Lirias (KU Leuven) (submittedVersion, cc-by)
  • Rice Research Repository (Rice University) (submittedVersion, cc-by)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Apollo
Funders
  • International Brain Research Organization
  • National Institutes of Health
  • Defense Advanced Research Projects Agency (HR0011, HR0011-18-2-0025)
  • Advanced Research Projects Agency
  • Intelligence Advanced Research Projects Activity (D16PC00003)
  • Interior Business Center
  • National Institute of Mental Health (R01MH109556)
  • National Eye Institute (P30EY002520)
  • H2020 European Research Council (101021347)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
  • Physical Sciences › Computer Science › Artificial Intelligence › Data Stream Mining Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
continual learning, incremental learning, task-incremental learning, domain-incremental learning, class-incremental learning, catastrophic forgetting, non-stationary data streams, deep neural network

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Three types of incremental learning. https://app.cochat.ai/papers/10.1038/s42256-022-00568-3
The paper
van de Ven, G. M., Tuytelaars, T., & Tolias, A. S. (2022). Three types of incremental learning. Nature Machine Intelligence. https://doi.org/10.1038/s42256-022-00568-3

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.