Efficient Secure Aggregation for Privacy-Preserving Federated Machine Learning - E-seaML Secure Aggregation Protocol

Aug 4, 2026

Upload by Logic

Federated learning enables training machine learning models on distributed data without exposing users’ raw information by sending the model to clients, computing local updates, and aggregating them at a central server. Secure aggregation protects against data leakage from local updates, yet many existing protocols impose heavy communication and computation costs and do not efficiently handle large model update vectors. This paper proposes E-seaML, a one-round secure aggregation protocol that substantially reduces overhead for both users and the server while supporting integrity verification of the final model via a proof of honest aggregation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Introduction

Federated learning overview

Privacy risks and secure aggregation

Existing approaches and their limitations

Efﬁcient Secure Aggregation for Privacy-Preserving Federated Machine Learning Rouzbeh Behnia University of South Florida Mohammadreza Ebrahimi University of South Florida Aramn Riasi Virginia Tech Balaji Padmanabhan University of South Florida Thang Hoang Virginia Tech arXiv :2304 .0384 1v 1 [ cs .CR] 7 Apr 2023 Abstract Federated learning introduces a novel approach to training machine learning (ML) models on distributed data while preserving user's data privacy. This is done by distributing the model to clients to perform training on their local data and computing the ﬁnal model at a central server. To prevent any data leakage from the local model updates, various works with focus on secure aggregation for privacy preserving federated learning have been proposed. Despite their merits, most of the existing protocols still incur high communication and computation overhead on the participating entities and might not be optimized to efﬁciently handle the large update vectors for ML models. In this paper, we present E-seaML, a novel secure aggregation protocol with high communication and computation efﬁciency. E-seaML only requires one round of communication in the aggregation phase and it is up to 318 􀀂 and 1224􀀂 faster for the user and the server (respectively) as compared to its most efﬁcient counterpart. E-seaML also allows for efﬁciently verifying the integrity of the ﬁnal model by allowing the aggregation server to generate a proof of honest aggregation for the participating users. This high efﬁciency and versatility is achieved by extending (and weakening) the assumption of the existing works on the set of honest parties (i.e., users) to a set of assisting nodes. Therefore, we assume a set of assisting nodes which assist the aggregation server in the aggregation process. We also discuss, given the minimal computation and communication overhead on the assisting nodes, how one could assume a set of rotating users to as assisting nodes in each iteration. We provide the open-sourced implementation of E-seaML for public veriﬁability and testing. 1 Introduction Machine Learning (ML) has provided breakthrough results that outperform humans in many critical ﬁelds such as medical screening [1], cyber threat hunting [2], object recognition [3], and sequential decision making [4] . Today, many companies are eager to beneﬁt from ML on their private data but they cannot afford the infrastructure needed to run large ML modelson-premise. This has resulted in a surge in collaborative learning, in which several parties join forces to contribute to a single learning problem of interest. The widespread adoption of cloud-based services and mobile devices has enabled the generation of a huge amount of rich data and an untapped source of computation empowered by commodity hardware [5] . In such a setting, collaborative learning has provided the opportunity to leverage the fragmented data residing locally at other parties to increase the ML model's accuracy. However, such data often include sensitive private information which mandates severe restrictions on data transmission to other locations for training (e.g., due to compliance related restrictions such as HIPPA or GDPR [6]) . Accordingly, collecting the data from different isolated islands across different organizations [5] or large set of distributed users [7, 8] while assuming a central trusted party is not feasible. As a result, companies who govern sensitive private data (e.g., electronic health records or personal identiﬁable information) have been ﬁghting an uphill battle in which they frequently have to choose between gaining collaborative ML capabilities and giving up on their data privacy. To address this challenge, Federated Learning (FL) has emerged as a promising solution to train a centralized ML models by relying on the contribution of multiple parties while maintaining the privacy of the local data [9, 10] . In an FL setting, a central server is tas

FAQs

What problem does federated learning address regarding data privacy?

Federated learning trains models using distributed data by keeping local data confidential while only transmitting model updates for server-side aggregation. This avoids direct data movement to other locations for training.

Why do secure aggregation protocols matter in federated learning?

Local model updates can leak sensitive information about users’ data. Secure aggregation computes the aggregated result without revealing individual updates, reducing the risk of information leakage.

What efficiency improvement does E-seaML provide over prior secure aggregation protocols?

E-seaML performs aggregation with only one communication round and achieves significant reductions in communication and computation overhead for users and the server compared with the most efficient existing counterpart.

Aug 4, 2026

Upload by Logic