Incremental learning

In computer science, '''incremental learning''' is a method of machine learning in which input data is continuously used to extend the existing model's knowledge i.e. to further train the model. It represents a dynamic technique of supervised learning and unsupervised learning that can be applied when training data becomes available gradually over time or its size is out of system memory limits. Algorithms that can facilitate incremental learning are known as incremental machine learning algorithms. In contemporary machine learning literature, settings that require learning from a sequence of tasks while limiting catastrophic forgetting are often discussed under the closely related term '''continual learning'''.<ref>{{cite web|title=Continual Learning for Large Language Models: A Survey|url=https://arxiv.org/abs/2402.01364|website=arXiv|date=2024-02-02|access-date=2026-05-05}}</ref><ref>{{cite web|title=Continual Learning of Large Language Models: A Comprehensive Survey|url=https://arxiv.org/abs/2404.16789|website=arXiv|date=2024-04-25|access-date=2026-05-05}}</ref>

Many traditional machine learning algorithms inherently support incremental learning. Other algorithms can be adapted to facilitate incremental learning. Examples of incremental algorithms include decision trees (IDE4,<ref>Schlimmer, J. C., & Fisher, D. [https://www.researchgate.net/profile/Doug_Fisher2/publication/221603307_A_Case_Study_of_Incremental_Concept_Induction/links/555b1fc608ae980ca6122e64.pdf A case study of incremental concept induction]. Fifth National Conference on Artificial Intelligence, 496-501. Philadelphia, 1986</ref> ID5R<ref>Utgoff, P. E., [http://people.cs.umass.edu/~utgoff/papers/mlj-id5r.pdf Incremental induction of decision trees]. Machine Learning, 4(2): 161-186, 1989</ref> and [https://github.com/greenfish77/gaenari gaenari]), decision rules,<ref>Ferrer-Troyano, Francisco, Jesus S. Aguilar-Ruiz, and Jose C. Riquelme. [https://idus.us.es/xmlui/bitstream/handle/11441/39713/Incremental%20rule.pdf?sequence=4&isAllowed=y Incremental rule learning based on example nearness from numerical data streams]. Proceedings of the 2005 ACM symposium on Applied computing. ACM, 2005</ref> artificial neural networks (RBF networks,<ref>Bruzzone, Lorenzo, and D. Fernàndez Prieto. [https://rslab.disi.unitn.it/papers/R12-PRL-1999-11-13.pdf An incremental-learning neural network for the classification of remote-sensing images]. Pattern Recognition Letters: 1241-1248, 1999</ref> Learn++,<ref>R. Polikar, L. Udpa, S. Udpa, V. Honavar. [https://www.researchgate.net/profile/Vasant_Honavar/publication/2489080_Learn_An_Incremental_Learning_Algorithm_for_Supervised_Neural_Networks/links/0912f50d151e7d22df000000.pdf Learn++: An incremental learning algorithm for supervised neural networks]. IEEE Transactions on Systems, Man, and Cybernetics. Rowan University USA, 2001.</ref> Fuzzy ARTMAP,<ref>G. Carpenter, S. Grossberg, N. Markuzon, J. Reynolds, D. Rosen. [http://open.bu.edu/bitstream/handle/2144/2071/91.016.pdf?sequence=1 Fuzzy ARTMAP: a neural network architecture for incremental supervised learning of analog multidimensional maps]. IEEE transactions on neural networks, 1992</ref> TopoART,<ref name="TopoART">Marko Tscherepanow, Marco Kortkamp, and Marc Kammer. [http://aiweb.techfak.uni-bielefeld.de/files/tscherepanow.marko2011ahierarchical-nn-r1.pdf A Hierarchical ART Network for the Stable Incremental Learning of Topological Structures and Associations from Noisy Data] {{Webarchive|url=https://web.archive.org/web/20170810022345/http://aiweb.techfak.uni-bielefeld.de/files/tscherepanow.marko2011ahierarchical-nn-r1.pdf |date=2017-08-10 }}. Neural Networks, 24(8): 906-916, 2011</ref> and IGNG<ref>Jean-Charles Lamirel, Zied Boulila, Maha Ghribi, and Pascal Cuxac. [https://www.researchgate.net/profile/Pascal_Cuxac/publication/47760684_A_new_incremental_neural_clustering_approach_for_performing_reliable_large_scope_scientometrics_analysis/links/58dbbb2c458515152b23f075/A-new-incremental-neural-clustering-approach-for-performing-reliable-large-scope-scientometrics-analysis.pdf A New Incremental Growing Neural Gas Algorithm Based on Clusters Labeling Maximization: Application to Clustering of Heterogeneous Textual Data]. IEA/AIE 2010: Trends in Applied Intelligent Systems, 139-148, 2010</ref>) or the incremental SVM.<ref>Diehl, Christopher P., and Gert Cauwenberghs. [http://www.isn.ucsd.edu/pubs/ijcnn03_inc.pdf SVM incremental learning, adaptation and optimization] {{Webarchive|url=https://web.archive.org/web/20171215192844/http://www.isn.ucsd.edu/pubs/ijcnn03_inc.pdf |date=2017-12-15 }}. Neural Networks, 2003. Proceedings of the International Joint Conference on. Vol. 4. IEEE, 2003.</ref>

The aim of incremental learning is for the learning model to adapt to new data without forgetting its existing knowledge. Some incremental learners have built-in some parameter or assumption that controls the relevancy of old data, while others, called stable incremental machine learning algorithms, learn representations of the training data that are not even partially forgotten over time. Fuzzy ART<ref name="Fuzzy ART">Carpenter, G.A., Grossberg, S., & Rosen, D.B., [http://dcommon.bu.edu/bitstream/handle/2144/2070/91.015.pdf?sequence=1&isAllowed=y Fuzzy ART: Fast stable learning and categorization of analog patterns by an adaptive resonance system], Neural Networks, 4(6): 759-771, 1991</ref> and TopoART<ref name="TopoART" /> are two examples for this second approach.

Incremental algorithms are frequently applied to data streams or big data, addressing issues in data availability and resource scarcity respectively. Stock trend prediction and user profiling are some examples of data streams where new data becomes continuously available. Applying incremental learning to big data aims to produce faster classification or forecasting times.

== Generative AI == Recent work extends incremental or continual learning to generative AI systems, including large language models, multimodal models, and diffusion models, with the goal of updating capabilities over time without requiring full retraining for every change.<ref>{{cite web|title=Continual Learning for Generative AI: From LLMs to Vision-Language-Action Models|url=https://arxiv.org/abs/2506.13045|website=arXiv|date=2025-06-19|access-date=2026-05-05}}</ref><ref>{{cite web|title=Continual Learning of Large Language Models: A Comprehensive Survey|url=https://arxiv.org/abs/2404.16789|website=arXiv|date=2024-04-25|access-date=2026-05-05}}</ref> Andreessen Horowitz has argued that continual learning could reduce the need for repeated large retraining runs by helping systems incorporate new information over time, although this remains an active research challenge rather than a solved production capability.<ref>{{cite web|title=Why We Need Continual Learning|url=https://a16z.com/why-we-need-continual-learning/|website=a16z|date=2026-04-22|access-date=2026-05-05}}</ref>

==See also== *Transduction (machine learning)

==References== {{Reflist}}

== External links == * {{cite web|title=Brief Introduction to Streaming data and Incremental Algorithms|url=https://blog.bigml.com/2013/03/12/machine-learning-from-streaming-data-two-problems-two-solutions-two-concerns-and-two-lessons/|author=charleslparker|date=March 12, 2013|website=BigML Blog}} * {{cite conference|title=Incremental learning algorithms and applications|url=https://www.esann.org/sites/default/files/proceedings/legacy/es2016-19.pdf|first1=Alexander|last1=Gepperth|first2=Barbara|last2=Hammer|year=2016|conference=ESANN|pages=357–368}} * [https://www.LibTopoART.eu LibTopoART: A software library for incremental learning tasks] * {{cite web|url=https://creme-ml.github.io |title=Creme: Library for incremental learning |archive-url=https://web.archive.org/web/20190803170741/https://creme-ml.github.io/ |archive-date=2019-08-03 }} * gaenari: [https://github.com/greenfish77/gaenari C++ incremental decision tree algorithm] * YouTube search results [https://www.youtube.com/results?search_query=incremental+learning Incremental Learning]

Category:Machine learning algorithms