"Securing What Privacy Hides: Verifiable Client Execution for Privacy-Preserving Edge Federated Learning"
Abstract
"Secure aggregation protects individual client updates in edge federated learning, but it also creates an integrity blind spot. Once individual edge client updates are protected, the server cannot determine whether a submitted payload was produced by the prescribed local training procedure or replaced by compromised host software. This limitation is especially relevant when distributed clients operate outside the server's administrative domain. Existing defenses mainly inspect geometric properties such as update norms or coordinate ranges, but these output-level checks cannot establish execution provenance and may still admit adaptively crafted parameter-tampering poisoning updates.We present SDV-Aggregation, a hardware-rooted client dual-validation framework. It binds the declared input state, approved local execution, and exact protected payload into a unified verification chain while preserving the confidentiality of local data and individual updates. The server can therefore reject payloads generated outside the attested workflow using compact evidence, without requiring a separate interactive proof protocol.We implement SDV-Aggregation on a hardware-backed Trusted Execution Environment platform and evaluate it under representative parameter-tampering poisoning attacks and heterogeneous client data distributions. The results show that the framework maintains global convergence and benign model utility while introducing compact authentication evidence and server-side…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.