Protecting Model Updates in Privacy-Preserving Federated Learning
Summary
This NIST Cybersecurity Insights article discusses input privacy in privacy-preserving federated learning (PPFL) systems, particularly focusing on the horizontally-partitioned setting. It builds upon previous posts that described attacks on models, privacy concepts, and data partitioning methods in PPFL.
IFF Assessment
The article explores ways to enhance privacy within federated learning systems, benefiting defenders by reducing the risk of data exposure.
Severity
Defender Context
Federated learning introduces new attack surfaces related to model poisoning and data leakage. Defenders should monitor participant contributions for anomalies and implement robust aggregation mechanisms. Staying informed about privacy-enhancing technologies like differential privacy and secure aggregation is crucial for mitigating these risks.