Cognitive EW: Assuring In-Mission Learning for EW
This presentation will discuss the challenges for assuring the performance of a system that can learn from novel experiences in the field. EW systems operate at a timescale that means they cannot afford to learn post-mission, or with human supervision. EW systems must learn from a single observation, using self-supervised reinforcement feedback. The validation infrastructure must therefore support automated closed-loop, multi-resolution testing, and ways to test the effectiveness of actions. We must validate the learning process, rather than validating the learned model.