Presentation Information
[2A02【基調講演】]Deepening and extension of Bayesian measurement
*Masato Okada1 (1. Graduate School of Frontier Sciences, The University of Tokyo)
I introduce Bayesian inference on spectral decomposition that decomposes a multimodal spectrum into a linear sum of unimodal basis functions such as Gaussian functions. How to determine the number K of basis functions is an important issue in spectral decomposition. Selecting this optimal K is called model selection in statistics. Spectral decomposition is not just function fitting or data analysis. The center position and width of the basis function correspond to the energy level and relaxation time. If the number of basis functions K is not appropriate, the position of the basis functions will shift and physical quantities cannot be extracted correctly. Model selection is a very important issue for physics. Based on Bayesian inference, we proposed a theoretical framework for estimating the number of basis functions K from data.