Application of ML tools for extraction of BPM-Q and transverse beam matching

Data inaccuracies make training an accurate Beam Quadrupole Moment at BPM (BPMQ) model challenging. Similarly, reconstructing Courant-Snyder (CS) parameters from BPMQ predictions is difficult due to BPMQ model limitations. Increasing the number of BPMQ predictions helps mitigate overfitting in CS inference caused by inaccurate BPMQ predictions. We present Bayesian Active Learning (BAL) to acquire measurements strategically, improving CS parameter inference despite model limitations.