@article{oai:ipsj.ixsq.nii.ac.jp:00232856,
 author = {Tomohiro, Nabika and Kenji, Nagata and Shun, Katakami and Masaichiro, Mizumaki and Masato, Okada and Tomohiro, Nabika and Kenji, Nagata and Shun, Katakami and Masaichiro, Mizumaki and Masato, Okada},
 issue = {1},
 journal = {情報処理学会論文誌数理モデル化と応用（TOM）},
 month = {Feb},
 note = {In some spectral measurements, such as absorption spectra, data are obtained as observation rates. When analyzing such data, Gaussian noise is typically assumed. However, the process of data generation can be modeled with binomial distribution noise. Conversely, in Bayesian analysis for spectral measurements, selecting an appropriate noise model is important. Therefore, we developed Bayesian spectral deconvolution based on a binomial distribution and compared it with Bayesian spectral deconvolution based on a Gaussian distribution. Using artificial data, we show that different noise models change the posterior distribution of peak numbers and their parameters, thereby affecting the analysis results. Moreover, we found that Bayesian spectral deconvolution based on a binomial distribution can analyze data with flattened peak structures, which was previously impossible to analyze. Using real data from X-ray emission spectroscopy, we confirmed that binomial distribution noise is more appropriate than Gaussian noise by Bayesian inference., In some spectral measurements, such as absorption spectra, data are obtained as observation rates. When analyzing such data, Gaussian noise is typically assumed. However, the process of data generation can be modeled with binomial distribution noise. Conversely, in Bayesian analysis for spectral measurements, selecting an appropriate noise model is important. Therefore, we developed Bayesian spectral deconvolution based on a binomial distribution and compared it with Bayesian spectral deconvolution based on a Gaussian distribution. Using artificial data, we show that different noise models change the posterior distribution of peak numbers and their parameters, thereby affecting the analysis results. Moreover, we found that Bayesian spectral deconvolution based on a binomial distribution can analyze data with flattened peak structures, which was previously impossible to analyze. Using real data from X-ray emission spectroscopy, we confirmed that binomial distribution noise is more appropriate than Gaussian noise by Bayesian inference.},
 pages = {47--56},
 title = {Bayesian Spectral Analysis with Binomial Distribution Noise},
 volume = {17},
 year = {2024}
}