Fed-REACT: Federated Representation Learning for Heterogeneous Time Series Data

ICLR 2025 Conference Submission12645 Authors

Keywords: Federated learning, representation learning, time series data

TL;DR: We study an FL framework where clients train on heterogeneous time series data and introduce Fed-REACT, a novel federated learning method leveraging representation learning and evolutionary clustering for time series data.

Abstract:

Motivated by high resource costs and privacy concerns that characterize centralized machine learning, federated learning (FL) emerged as an efficient alternative that allows the participating clients to collaboratively train global model while keeping their data local.
In practice, distributions of clients' data vary over time and from one client to another, creating heterogeneous conditions that deteriorate performance of conventional FL algorithms. In this work, we study an FL framework where clients train on heterogeneous time series data and introduce to these settings Fed-REACT, a novel federated learning method leveraging representation learning and evolutionary clustering. The algorithm consists of two stages: (1) in the first stage, the clients learn a model that extracts meaningful features from local time series data; (2) in the second stage, the server adaptively groups clients into clusters and coordinated cluster-wise learning of task (i.e., post-representation) models for local downstream tasks, e.g., classification or regression. We demonstrated high accuracy and robustness of the proposed algorithm in experiments on real-world time series datasets, and provided theoretical analysis of its performance.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12645