Receiver-Domain Behavioral Probing for Backdoor-Resilient Federated GPS Spoofing Detection in UAV Networks
Poisoned UAV clients inflate accuracy to backdoor federated GPS spoofing detectors; behavioral probing largely removes the lift.
Federated UAV GPS-spoofing detectors that weight clients by self-reported validation accuracy can be gamed. Two of ten compromised clients that poison data, scale updates, and inflate reported accuracy raise backdoor lift to +0.3036, above the same attack without lying. Receiver-domain behavioral probing evaluates models on counterfactual spoofed samples and reduces lift to -0.0265 under IID clients, near an honest fleet, while flagging compromised aircraft. Unlike Multi-Krum it needs only an honest majority; it degrades under strong client heterogeneity on a simulated partition of one public dataset.
- Two of ten clients inflate accuracy and raise backdoor lift to +0.3036
- Probing scores models on counterfactual GPS features instead of self-reports
- Under IID clients, attacker-induced lift falls to -0.0265
- Multi-Krum rises to +0.2837 when the attacker count is misspecified
- The defense weakens under strong client heterogeneity
From the original article
1 min read at arxiv.orgFederated learning lets a UAV fleet train a shared GPS spoofing detector without raw receiver data leaving any aircraft, and several recent UAV-FL designs weight each client by the validation accuracy it reports about itself.
By Will Jedrzejczak · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.