Federated Meta-Learning in the Time-Series Domain: a Scoping Review
Abstract
Abstract Wearables, industrial sensors, and other connected systems can create time-series data that is used in making decisions, predictions, and monitoring. Nevertheless, such information can be highly sensitive, and it involves the behaviour of the users, their habits, and health status, which pose a serious privacy risk. Federated learning (FL) has become a powerful tool to improve this issue, enabling models to be trained cooperatively and retaining raw data in local devices. Simultaneously, meta-learning methods have also been developed to provide quick adaptation of models to new users or tasks, and this is especially useful in low-data settings. Despite the extensive research on both FL and meta-learning, their combination as federated meta-learning (FedMeta) to time-series applications has received comparatively limited literature coverage and is scattered across the research. This paper provides a review of the literature in which FedMeta is used in terms of time series, where the interest is specifically on privacy preservation. We do not just look at the performance of the algorithm, but we look at how these methods cope with the major real-world problems, such as client-level personalisation, data distribution heterogeneity, and performance limitations imposed by limited computation or communication resources. We further examine the privacy mechanisms that have been used in previous studies and identify instances where privacy assumptions or threat models are…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.