Bayesian-Inference-based Inverse Estimation of Small Angle Scattering.

As an application of machine-learning algorithms, we improved SAS (Small Angle Scattering), which is common experiment in material science, by developing a Bayesian inference and deriving the confidence-level contour. In the SAS experiment, the grain-size of the sample material has to be estimated from the distribution of the scattered beam. A stochastic model and maximum-likelihood inference with EM-algorithm are often used, but the result is noisy due to data noise. With the proposed method, the grain-size distribution can be estimated similarly to the maximum-likelihood inference method and the confidence levels can be visualized. Thus, researchers can determine estimation reliability and decide whether there are sufficient data. Simulation-generated datasets were processed with the proposed method to evaluate its effectiveness, and it was confirmed that it is useful for automatic SAS data analysis.