Resilient Control of Switched Vehicle Platoons under False Data Injection Attacks

What the math gives to ML

The paper offers a transferable observer-based mechanism for rejecting false-data injections when the attack amplitude may be unbounded but its rate is bounded. Its key machinery is augmentation of the plant with an attack-generating auxiliary state, link-wise observers, and a uniform switching-observability Gramian that permits reconstruction despite mode changes. In neural networks, this can become a resilient message-passing or distributed-training layer: estimate an additive corruption separately on every communication edge, subtract the estimate, and certify bounded residual propagation under changing layer, routing, or expert modes. The strongest testable prediction is that residual error depends on attack rate and observer bandwidth, not attack amplitude, provided a finite-window observability condition holds.