Development of model selection method for Bayesian estimation from behavior of Markov chain Monte Carlo method

2017 Fiscal Year Final Research Report

• Project/Area Number: 25330283
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Soft computing
• Research Institution: National Institute of Advanced Industrial Science and Technology (2016-2017); The University of Tokyo (2013-2015)
• Principal Investigator: Nagata Kenji 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (10556062)
• Project Period (FY): 2013-04-01 – 2018-03-31
• Keywords: マルコフ連鎖モンテカルロ法 / ベイズ推定 / モデル選択 / スペクトル分解

Outline of Final Research Achievements

In this research, we develop a new model selection method for Bayesian estimation based on behavior and relaxation process of MCMC method. An expected result is that the design guidelines of the MCMC method are clarified. The MCMC method used in statistical mechanics and others is often discrete, and in this research the behavior of continuous algorithms is more complicated, and the design guidelines of MCMC method have a spillover effect on a wide range of fields. It is also important to develop a model selection method in a hierarchical probabilistic model. The model selection criteria typified by AIC and BIC assumes asymptotic normality, and when used in a hierarchical model, the result of bias is obtained. Establishing a general-purpose model selection method is a very important task.

Free Research Field

データ駆動科学