Spectral Deconvolution through Bayesian LARS-OLS

Spectral deconvolution is a method of fitting spectral data to the sum of unimodal basis functions. It is often used to estimate spectral information such as peak position or intensity. Recently, a spectral deconvolution method based on Bayesian inference was proposed that achieved the deconvolution of complex spectral data. However, when the number of spectral model parameter increases, the computational cost of the Bayesian spectral deconvolution increases significantly. To overcome this difficulty, a faster spectral deconvolution method using the L1 regularized vector machine (L1VM) was developed, but it could not estimate peak parameters as accurately as the Bayesian spectral deconvolution. We propose two spectral deconvolution methods, namely Bayesian LARS-OLS spectral deconvolution and LARS-OLS spectral deconvolution, both of which divide the procedure of spectral deconvolution into two steps: basis search and regression. Using artificial spectral data, we demonstrate that the estimation accuracies ...