Reproduce “Gradient-Based Defensive Poisoning Against System Identification Attacks on Cyber-Physical Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Gonçalo Fernandes, José Cecílio, Victor Mattos, Daniel S. Menasché, Amit Bhaya, Alan Oliveira de Sá

The authors propose poisoning plant-output measurements in decoy control systems to make an attacker’s learned model less accurate without altering the process. They shape perturbations using gradients from surrogate neural networks, focusing them on setpoint transients and retaining samples. In a motor benchmark, designed perturbations produced higher prediction error than RMS-matched Gaussian and PRBS controls; impact and detectability remain untested.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.