Bayesian nonparametrics (Q1874271)

The present book has 11 chapters. Their contents is: Preliminaries and the Finite-Dimensional Case; \(M(\chi)\) and Priors on \(M(\chi)\); Dirichlet and Polya tree processes; Consistency Theorems; Density Estimation; Inference for Location Parameters; Regression Problems; Uniform Distributions on Infinite-Dimensional Spaces; Survival Analysis -- Dirichlet Priors; Neutral to the Right Priors; Exercises. Since Abraham Wald, Bayesian procedures are of high interest in statistical inference and decision theory. The present work gives an overview on the consistency in Bayesian inference also in the case of nonparametric models with infinite dimension. Bayes estimates rely on the posterior distribution of the parameter given the data. The main part of the book is devoted to posterior consistency of different kinds of models: density estimation, location parameters, regression problems and survival analysis. Chapter 1 deals with maximum likelihood estimators and Bayes estimators also for non-regular examples. The reader will find a good overview about consistency and asymptotic normality of these estimators at finite dimensions. For instance, the Bernstein von Mises theorem about the asymptotic normality of the posterior is proved in section 1.4. Chapter 2 introduces priors on the space of probability measures which is needed for nonparametric models. The most important priors are Dirichlet distributions (and processes). They are presented in Chapter 3. Chapter 4 introduces general consistency results for priors. The work is related to the work of Lorraine Schwartz and Le Cam. In the next chapters 5-9 the work is applied to different semiparametric problems. It turns out that the present Bayes procedures are typically consistent and that the principles of the general theory can be applied. Again posterior consistency can typically be proved. In chapter 10 it is pointed out how neutral to the right priors are linked to independent increments processes. They are important for survival models. The present monograph gives a nice overview on the state of the art in Bayesian nonparametrics. Also the classical part is worthwhile where a lot of basic facts are summarized. On the other hand, modern Bayesian statistical inference like the density estimation from the Bayesian point of view is presented. The core of the work seems to be the part about Dirichlet processes. The reader will find a huge amount of references. In conclusion, the present book can be recommended for research and advanced lectures and seminars.