Tesis Doctorado
Theory and applicatións of dependent nonparametric bayesian models for bounded and unbounded responses
Autor
Barrientos, Andrés Felipe
Institución
Resumen
The definition and study of theoretical properties of probability models defined on infinite - dimensional spaces have received increasing attention in the statisticalliterature because these models are the basis for the Bayesian nonparametric (BNP) generalization of finite-dimensional statistical models (see, e.g., Ghosh & Ramamoorthi, 2003; Müller & Quintana, 2004; Hjort et al., 2010). These generalizations allow the user to gain model flexibility and robustness against mis-specification of a parametric statistical model. BNP models are specified by defining a stochastic process whose trajectories belong to a functional space, g, su eh as the space of all probability measures defined on a given measurable space. The law governing such a process is then used as a prior distribution for a functional parameter in the Bayesian framework. The increase in applications of BNP methods in the statisticalliterature has been motivated largely by the availability of simple and efficient methods for posterior computation in Dirichlet process mixture (DPM) models (Ferguson, 1983; Lo, 1984). The DPM models incorporate
Dirichlet process (DP) priors (Ferguson, 1973, 1974) for components in Bayesian hierarchical
models, resulting in an extremely flexible class of models. Due to the flexibility and ease in
implementation, DPM models are now routinely implemented in a wide variety of applications,
ranging from machine leaming to genomics (see, e.g. Hjort et al., 201 0). Furthermore, a lich
theoreticalliterature about support, posterior consistency and rates of convergence (Lo, 1984;
Ghosal et al., 1999; Lijoi et al., 2005; Ghosal & Van der Vaart, 2007) justify the use of DPM
models for inference in single density estimation problems. PFCHA-Becas Doctor en Estadísticas 192p. PFCHA-Becas TERMINADA