Brief

Researchers have developed a novel defense mechanism called Pseudo-Feature Padding to protect Deep Neural Networks (DNNs) used in Cyber-Physical Systems (CPS) from False Data Injection Attacks (FDIA). This lightweight, model-agnostic method introduces an additional input layer that randomly pads input samples with pseudofeature values derived from the input's statistical distribution. This randomization makes adversarial attacks computationally infeasible and enhances model robustness without altering the core DNN architecture. The framework was successfully tested on power grid applications, demonstrating significant improvement in model resilience against attacks with minimal performance impact. AI