FedPS: Federated data Preprocessing via aggregated Statistics
Keywords: Data Preprocessing, Federated Learning, Aggregated Statistics
TL;DR: A comprehensive suite of tools designed for data preprocessing in federated learning.
Abstract: Data preprocessing is a crucial step in machine learning that significantly influences model accuracy and performance. In Federated Learning (FL), where multiple entities collaboratively train a model using decentralized data, the importance of preprocessing is often overlooked. This is particularly true in Non-IID settings, where clients hold heterogeneous datasets, requiring aggregated parameter estimates to perform consistent data preprocessing. In this paper, we introduce FedPS, a comprehensive suite of tools for federated data preprocessing. FedPS leverages aggregated statistics, data sketching, and federated machine learning models to address the challenges posed by distributed and diverse datasets in FL. Additionally, we resolve key numerical issues in power transforms by improving numerical stability through log-space computations and constrained optimization. Our proposed Federated Power Transform algorithm, based on Brent’s method, achieves superlinear convergence. Experimental results demonstrate the impact of effective data preprocessing in federated learning, highlighting FedPS as a versatile and robust solution compared to existing frameworks. The implementation of FedPS is open-sourced.
Supplementary Material: zip
Primary Area: infrastructure, software libraries, hardware, systems, etc.
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: 11217
Loading