Beyond Corner Patches: Semantics-Aware Backdoor Attack in Federated Learning

ArXi:2603.29328v1 Announce Type: cross Backdoor attacks on federated learning (FL) are most often evaluated with synthetic corner patches or out-of-distribution (OOD) patterns that are unlikely to arise in practice. In this paper, we revisit the backdoor threat to standard FL (a single global model) under a realistic setting where triggers must be semantically meaningful, in-distribution, and visually plausible.