PI: Rick Blum
University: Lehigh University
This research project will develop detection and mitigation techniques for cyber-attacks on systems trying to classify objects based on sensor observations. The Pennsylvania company partner I2R Electronics Inc., part of Nanowave Technologies Inc., is making such products, and the goal of this project is to develop technology that will ultimately protect their products from cyber-attacks. The project will build on research results from previous work at Lehigh University for systems trying to estimate parameters, like object position. The team intends to devise provably effective algorithms for identifying attacked data as well as employing the unattacked and attacked data in provably near optimum, low complexity classification approaches. Similar to the previous Lehigh research, the team will develop analytical formulations of the best possible classification approaches given either attacks or no attacks. These results will be used to describe the best possible processing under attack, but these approaches will require high complexity. Performance bounds can be generated by assuming the attacked system knows exactly which data is attacked. The team will search for low complexity algorithms that can provide performance close to these bounds. Low complexity unsupervised machine learning approaches will be studied that mimic the developed high complexity analytical formulations of optimum classification under attack based on the previous Lehigh research.