Three types of incremental learning

By Gido M. van de Ven, Tinne Tuytelaars, Andreas S. Tolias. 2022

787 citations recorded by the source corpus.

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. A challenge for any machine learning system is to continually adapt to new data. While methods to address this issue are developed, their performance is hard to compare. A new framework to facilitate benchmarking divides approaches into three categories, defined by whether models need to adapt to new tasks, domains or classes.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Han-Jia Ye
2. De-chuan Zhan
3. Da-Wei Zhou
4. Qiwen Wang
5. Ziwei Liu
6. A. Mahmood
7. Zhiyuan Qi
8. Zhi-Hong Qi

In Computer Vision and Pattern Recognition

1. Han-Jia Ye
2. De-chuan Zhan
3. Da-Wei Zhou
4. Ziwei Liu
5. Luc Van Gool
6. Dacheng Tao
7. Bernt Schiele
8. Qiwen Wang

Citations in the corpus

• Incorporating neuro-inspired adaptability for continual learning in artificial intelligence (2023, neutral): “Finally, we evaluate CAF under the setting of task-incremental learning 52 and compare it with a range of representative methods 7–9 , 53– 56 .”
• The Ideal Continual Learner: An Agent That Never Forgets (2023, background): “…Ramesh & Chaudhari, 2022; Douillard et al., 2022), or combinations thereof (Yan et al., 2021; Xie et al., 2022; Wang et al., 2022b; Frascaroli et al., 2023); for surveys, see, e.g., Parisi et al. (2019); De Lange et al. (2022); Qu et al. (2021); Shaheen et al. (2022); van de Ven et al. (2022).”
• Loss of plasticity in deep continual learning (2024, positive): “We use it here because it is the standard practice in deep continual learning for this type of problem in which the learning system has to learn a sequence of independent tasks 52 .”
• Hebbian Learning based Orthogonal Projection for Continual Learning of Spiking Neural Networks (2024, background): “Meanwhile, continual learning can be classified into three fundamental types: task-incremental, domain-incremental, and class-incremental (van de Ven et al., 2022).”
• Continual Learning of Large Language Models: A Comprehensive Survey (2024, neutral): “Institutions, regardless of size, often possess significant amounts of unlabeled, domain-specific data.”
• A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts (2023, neutral): “There are three fundamental scenarios of CL [64]: task-incremental, domain-incremental, and class-”
• Deep Class-Incremental Learning: A Survey (2023, background): “Other surveys either focuses on the specific field [25], [26], [27] or lacking the performance evolution among state-of-the-arts [28], [29], [30].”
• Class-Incremental Learning: A Survey (2023, background): “Hence, although these works achieve competitive results in task-incremental learning, many works [8] find that parameter regularization-based methods perform poorly in the class-incremental learning scenario.”
• CaSpeR: Latent Spectral Regularization for Continual Learning (2023, methodology): “These methods are evaluated by dividing a classification dataset into disjoint subsets of classes, called tasks, letting the model fit one task at a time and evaluating it on all previously seen data [45].”
• Steering Prototypes with Prompt-tuning for Rehearsal-free Continual Learning (2023, background): “Prevalent algorithms can typically be categorized into three primary branches [41, 54, 19].”