Keywords: Federated Learning, Differential Privacy
Abstract: Federated learning (FL) is a distributed learning paradigm that allows multiple decentralized edge devices to collaboratively learn toward a common objective without sharing local data. Although local data is not exposed directly, privacy concerns nonetheless exist as sensitive information can be inferred from intermediate computations. As the same data is repeatedly used over an iterative process, information leakage accumulates substantially over time, making it difficult to balance the trade-off between privacy and accuracy. In this paper we introduce Upcycled-FL, a novel federated learning framework, where first-order approximation is applied at every even iteration. Under such a scheme, half of the steps incur no privacy loss and require much less computation. Theoretically, we establish the convergence rate performance of Upcycled-FL and provide privacy analysis based on objective and output perturbations. Experiments on real-world data show that Upcycled-FL consistently outperforms existing methods over heterogeneous data, and significantly improves privacy-accuracy trade-off, while reducing 48% of the training time on average.
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.
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: Yes
Please Choose The Closest Area That Your Submission Falls Into: Social Aspects of Machine Learning (eg, AI safety, fairness, privacy, interpretability, human-AI interaction, ethics)
TL;DR: We propose a federated learning framework that improves accuracy-privacy tradeoff with less computation.
Supplementary Material: zip
4 Replies
Loading