.png){kind=icon}
.png){kind=icon}
Return
Federated continual learning for task-incremental and class-incremental problems: A survey
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129278.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In today’s information-driven world, valuable data for training intelligent systems is often distributed across private devices. Transferring such data to a central server introduces privacy concerns and significant communication overhead. Federated Learning enables devices to share knowledge and train a model collaboratively. Advanced smart devices often need to learn new tasks over time, making Continual Learning techniques crucial to retrain models with new data while retaining previously learned concepts. Consider a scenario in which evolving private data is distributed across devices, and we want to train a model in Continual Learning manner. For example, assume a healthcare application on smartphones monitors factors such as heart rate, body temperature, and sleep patterns over time to detect diseases. With Continual Learning, this application can adapt to detect new disease types as they emerge. Establishing Federated Learning for this process enables collaborative Continual Learning, enhancing privacy and potentially improving generalization. Such problems are common in real-world applications, including smart vehicles, mobile devices, and IoT systems. Federated Continual Learning algorithms are designed to address these problems. This survey presents concepts and provides concise descriptions of approaches proposed for Federated Continual Learning focusing on task-incremental and class-incremental problems.
Keywords:
Federated Learning
Continual Learning
Privacy-preserving
Distributed Learning
Task-incremental Learning
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available