Full text
| |
| Author(s): |
Victor Freguglia Souza
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Matemática, Estatística e Computação Científica |
| Defense date: | 2022-05-06 |
| Examining board members: |
Nancy Lopes Garcia;
Helio dos Santos Migon;
Florencia Graciela Leonardi;
Jesus Enrique Garcia;
Guilherme Vieira Nunes Ludwig
|
| Advisor: | Nancy Lopes Garcia |
| Abstract | |
This work addresses the problem of selecting the interaction structure of a class of Markov Random Fields and Variable-Length Markov Chain models from a Bayesian perspective. The interaction neighborhood of the Markov Random Fields and the context tree of the Variable-Length Markov Chain model are treated as unobserved random objects of a Bayesian system, and Monte-Carlo Markov Chain methods are proposed to generate samples of these objects, constructing proposal kernels on these spaces of arbitrary objects that represent models, overcoming computational challenges intrinsic to these models. A complete framework for inference on Markov Random Fields is proposed in the R package mrf2d. The proposed methods are applied to selecting the interaction neighborhood in a texture image analysis problem and for detecting renewal states in a Markov Chain of encoded written texts (AU) | |
| FAPESP's process: | 17/25469-2 - Bayesian inference for signal segmentation in dyed textile images |
| Grantee: | Victor Freguglia Souza |
| Support Opportunities: | Scholarships in Brazil - Doctorate |