Adversarial AI for RF
Abstract
The Adversarial AI for RF program aims to develop artificial intelligence (AI) capabilities with applications for national security, particularly in areas such as electronic warfare. Given that U.S. and potential adversaries are developing AI technology with potentially revolutionary capabilities, DoD must prepare for conflicts that include offensive and defensive AI actors. Adversarial AI will develop methodologies for protecting AI-enabled DoD systems from adversary attempts to elicit an erroneous response (spoofing) and for significantly increasing AI system reliability and safety. The resulting AI algorithms should also ensure that AI-enabled DoD devices offer human-understandable explanations for their suggested course of action, to the maximum extent possible. To enable this future, the Adversarial AI program will leverage and advance newly formed links between machine learning and security and look to extend these emerging techniques to military specific domains that are emerging such as cognitive Electronic Warfare (EW) systems. Finally, the program may inform the correct mechanism to train AI systems for use in adversarial situations.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2019
- Source ID
- cf5375b18cf9a617c2433e7a12cb9096