# Optimal Byzantine-resilient Federated Learning with User-level Differential Privacy

DevFeed: [Optimal Byzantine-resilient Federated Learning with User-level Differential Privacy](<https://devfeed.tech/articles/optimal-byzantine-resilient-federated-learning-with-user-level-differential-privacy-125279.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10898>)

Author: Ming Xiang, Stratis Ioannidis, Edmund Yeh, Carlee Joe-Wong, Lili Su

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Federated Learning](<https://devfeed.tech/topics/federated-learning.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [central](<https://devfeed.tech/tags/central.md>), [client](<https://devfeed.tech/tags/client.md>), [clients](<https://devfeed.tech/tags/clients.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [convex](<https://devfeed.tech/tags/convex.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [federated-learning](<https://devfeed.tech/tags/federated-learning.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [guards](<https://devfeed.tech/tags/guards.md>), [information](<https://devfeed.tech/tags/information.md>), [lower-bound](<https://devfeed.tech/tags/lower-bound.md>), [math-oc](<https://devfeed.tech/tags/math-oc.md>), [privacy](<https://devfeed.tech/tags/privacy.md>)

## AI overview

The paper establishes an information-theoretic lower bound for Byzantine-resilient empirical risk minimization under user-level differential privacy with heterogeneous clients, then proposes ByzULDP, a central-DP method using momentum on both clients and server. Client momentum reduces stochastic gradient variance during robust aggregation. The authors say ordinary nonlinear Byzantine-resilient aggregation combined with central-DP noise may fail to ensure convergence near a stationary point; server momentum uses the way client updates determine user-level privacy sensitivity to support convergence. ByzULDP achieves an order-optimal bound for strongly convex objectives, with error terms decaying as O(1/T) for strongly convex and O(1/√T) for non-convex objectives. Numerical experiments on real-world datasets across privacy budgets support the analysis.

## Source excerpt

arXiv:2611.10898v1 Announce Type: new Abstract: Real-world deployment of federated learning (FL) requires resilience against adversarial clients and data privacy. Byzantine fault captures the worst-case behavior of adversarial clients, while differential privacy (DP) provides statistical guarantees against privacy loss. In particular, user-level DP guards a client's entire contribution and provides more realistic protection against information leakage. Despite intensive efforts, the minimax-optimal convergence bound of Byzantine-resilient FL with user-level DP remains open. We establish an information-theoretic lower bound for any Byzantine-resilient empirical risk minimization problem under user-level DP and heterogeneous clients. We propose ByzULDP, a central DP algorithm featuring a two-level momentum across the client and server. Client-side momentum mitigates stochastic gradient variance during robust aggregation. Unlike the fault-free case, naively combining non-linear Byzantine-resilient aggregation with central DP noise can be insufficient to guarantee convergence to a bounded neighborhood of a stationary point. Our key contribution is a server-side momentum that leverages a property specific to user-level DP, where privacy sensitivity depends on how client updates are constructed, to empower convergence. ByzULDP achieves the order-optimal convergence bound for strongly-convex objectives, with vanishing error terms decaying at the optimal $O(1/T)$ and $O(1/\sqrt{T})$ rates for strongly-convex and non-convex objectives, respectively. We corroborate our analysis with numerical experiments on real-world datasets across diverse privacy budgets.