black-box validation for reinforcement learning; PAC-based guarantees for black-box validation

Project Description: Over the last few years, Artificial Intelligence (AI) systems have achieved super-human performance in specific yet complex tasks across diverse environments (e.g., image recognition, language translation, complex games like Go). Given their effectiveness, we aim to employ AI (or Agent) Training Programs (ATPs) to generate scenarios automatically for training (on specific tasks) in safety-critical applications while addressing the issue of trust to improve adoption: