Toward Smart Security Enhancement of Federated Learning Networks

As traditional centralized learning networks (CLNs) are facing increasing challenges in terms of privacy preservation, communication overheads, and scalability, federated (FLNs) have been proposed as a promising alternative paradigm to support the training machine (ML) models. In contrast data storage processing CLNs, FLNs exploit number edge devices (EDs) store perform distributively. this way, EDs can keep locally, which preserves reduces overheads. However, since model within relies on contribution all EDs, process be disrupted if some upload incorrect or falsified results, that is, poisoning attacks. article, we review vulnerabilities FLNs, particularly give an overview attacks mainstream countermeasures. Nevertheless, existing countermeasures only provide passive protection fail consider fees paid for contributions resulting unnecessarily high cost. Hence, present smart security enhancement framework FLNs. particular, verify-before-aggregate (VBA) procedure is developed identify remove non-benign results from EDs. Afterward, deep reinforcement (DRL) applied learn behaving patterns actively select benign charge low fees. Simulation reveal protect effectively efficiently.