Extraction of Latent Dynamical Structure from Time-Series Spectral Data

The estimation of latent dynamics from time-series data is an important problem in a broad range of fields. In this research, we focused on time-series spectral data, which are obtained in planetary science, condensed matter science, and many other fields, and their latent dynamics. Time-series spectral data have a multiple-peak structure and the center, width, and amplitude of each peak reflect the nature of the subject. Here, we propose a method to estimate the parameters of peaks, the parameters of their latent dynamics, the number of peaks, and the order of the model by using Bayesian inference.