Lambda Science, Inc. (LSI) will develop innovative and operationally effective approaches to exploit weaknesses (i.e., defeat) in an adversarys cognitive sensing systems, and in doing so devise methods and techniques including CONOPS that protect our own cognitive radar and EW sensor systems. LSI will develop methods to construct adversarial examples and demonstrate their effectiveness to defeat ML architectures. The concept of adversarial sample transferability will be employed to quantify misclassification rates against ML methods. The Phase 1 effort will also develop and evaluate adversarial example detectors using a generalized threat model.
Benefit: The current proposal to develop innovative and operationally effective approaches to exploit weaknesses (i.e., defeat) in an adversarys cognitive sensing systems, and in doing so devise methods and techniques including CONOPS that protect our own cognitive radar and EW sensor systems, will produce AI capabilities that both fit synergistically with ongoing cognitive technology transition activities for PMA-262, PMA 266, PMA-290 and PMA-299, as well as inform the broader deployment strategy of cognitive radar and EW sensor systems. Any of these transition successes will allow LSI to pursue similar opportunities for transition on several other Navy, Air Force, Army, Coast Guard and Department of Homeland Security (DHS) Programs.
Keywords: Machine Learning, Machine Learning, Neural networks, Cognitive Sensors, Artificial Intelligence
