Radiofrequency Ensemble Classifier

To address the Navy’s need for trustworthy AI systems that are resilient to adversarial attacks, Intellisense Systems, Inc. (ISI) proposes to develop a new Radiofrequency Ensemble Classifier (RADEC) system based on an ensemble of complimentary classifiers trained on randomly transformed/augmented data. Specifically, the innovation in using a novel meta-learning genetic algorithm to optimally select the most effective set of complimentary classifiers will enable RADEC to successfully identify adversarial examples and thus determine whether standard AI systems may be used to interpret incoming RF data correctly. As a result, RADEC will provide the Navy the capabilities to develop resilient AI systems by (1) understanding failure modes of standard AI systems and (2) mitigating potential deceit of those AI systems. In Phase I, ISI will develop and demonstrate feasibility of RADEC’s ability to detect and correctly classify adversarial examples for both RF device fingerprinting and modulation classification. The feasibility demonstration will include simulated adversarial transmissions and actual over-the-air (OTA) transmissions using a software-defined radio (SDR). During Phase II, ISI will complete the algorithmic development of RADEC based on lessons learned from Phase I, test RADEC in high-fidelity operationally representative scenarios dealing with electronic warfare, and deliver a prototype for testing and evaluation.