Fig. 6: Adversarial Bayesian Optimization (ABO) approach workflow.

Starting with an arbitrary number of random initial energy points and their corresponding measured spectral values, two BO blocks are trained and updated competitively. The Fitting BO block fits the parameters of the Quanty model to identify the XAS curve that best matches the current measurements, while the Sampling BO block identifies the most likely point that could cause the fitted physical model to fail.