Coverage Probability

Coverage probability, a measure of how well a model or system accounts for uncertainty or encompasses a given space, is a central theme in various fields, aiming to improve reliability and robustness. Current research focuses on optimizing coverage in diverse contexts, including reinforcement learning (using multi-objective optimization and convex coverage sets), selective classification (leveraging hierarchical structures and risk-coverage curves), and test generation (employing LLMs and iterative coverage-guided approaches). These advancements enhance the trustworthiness and efficiency of models across applications ranging from robotics and machine learning to software testing and Bayesian inference, ultimately leading to more reliable and dependable systems.