大阪大学　数理・データ科学教育研究センター
Center for Mathematical Modeling and Data Science,Osaka University

Approximate Bayesian computation: convergence, limit, and misspecification

C. P. Roberts (Paris, Dauphine, France, and Warwick, UK)

1. ホーム
2. セミナー・集中講義
3. セミナー
4. Approximate Bayesian computation: convergence, limit, and misspecification

大阪大学 数理・データ科学セミナー データ科学セミナーシリーズ 第47回

Approximate Bayesian computation: convergence, limit, and misspecification

C. P. Roberts (Paris, Dauphine, France, and Warwick, UK)

ABC algorithms have become ubiquitous when handling highly complex models that prohibit exact Monte Carlo approaches. They however produce an approximation to the true posterior distribution and add to the Monte Carlo error found in regular simulation techniques a convolution error and an information error. It is thus of considerable interest to question the pseudo-posterior distributions associated with these algorithms and to characterize the existence and the meaning of the limiting distributions of ABC samples. In particular, we consider the impact of mixing ABC principles with a Gibbs sampler and with a misspecified model.
(Joint papers with G. Clarté, D. Frazier, G. Martin, J. Rousseau, R. Ryder, & J. Stoehr)

本発表はJST CREST JPMJCR14D7によってサポートされています．