Speaker
Description
Advanced sensors and readout electronics used in high-energy physics experiments increasingly operate in large, strongly coupled parameter spaces, making characterization, calibration, and performance optimization a growing bottleneck for instrumentation development. Bayesian optimization applied to charge-coupled devices (CCDs) has already shown promise for addressing this challenge by using closed-loop measurements and probabilistic surrogate models to efficiently identify high-performance operating regions without exhaustively scanning the full parameter space. Building on this foundation, this work explores agentic approaches in which separate agents coordinate instrument control, image acquisition and analysis, optimization, and persistent experimental knowledge; this allows the optimization loop to incorporate not only a scalar objective, but also image statistics, operator notes, and lessons accumulated across campaigns. We also investigate surrogate models that learn the multidimensional detector response itself. These surrogate models can support rapid virtual exploration of operating conditions and, with uncertainty quantification, guide the selection of new measurements where the model is least certain. Together, these approaches are intended to provide a generalizable framework for autonomous characterization and optimization across a broad range of detector and readout-electronics platforms.