Guaranteeing AI Robustness against Deception (GARD)

The Defense Advanced Research Projects Agency is soliciting innovative research proposals through its Guaranteeing AI Robustness against Deception Broad Agency Announcement to develop theoretical foundations, principled algorithms, and evaluation frameworks to enable machine learning systems to be robust against adversary deception. Proposals are sought across two technical areas - defense theories and algorithms, as well as an evaluation framework - and must address at least two sensor modalities from still images, audio, and video. Proposals will be evaluated biannually with the deadline for proposals on April 11, 2019.

There is no set-aside designation for this opportunity, which involves research and development services under NAICS code 541715 and PSC code R408. Evaluation scenarios will become more complex over three phases incorporating factors such as multiple sensors, modalities, and physical world attacks. Deliverables include theoretical findings, defense implementations, test results, and publications, with the overall budget to be determined based on the quality and number of proposals received. Awards will be made through contracts, grants, cooperative agreements or other transactions for a period of 48 months beginning in December 2019.