• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.18.2023.tde-10042023-154512
Document
Author
Full name
Vitor Bruno de Oliveira Barth
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2023
Supervisor
Committee
Maciel, Carlos Dias (President)
Nascimento, Luiz Fernando Costa
Vencio, Ricardo Zorzetto Nicoliello
Title in Portuguese
Um método Bayesiano orientado a dados para o aprendizado estrutural de Redes Bayesianas
Keywords in Portuguese
Aprendizado Estrutural
Monte Carlo via Cadeias de Markov
Redes Bayesianas
Abstract in Portuguese
A transparência em algoritmos de Inteligência Artificial é o foco de um novo ramo de pesquisa, chamado Inteligência Artificial Explicável, o qual sugere que especialistas devam ser capazes de replicar os resultados encontrados pelo modelo, bem como atualizá-los quando os resultados não estão de acordo com seus conhecimentos prévios a respeito do domínio analisado. Redes Bayesianas são um tipo de modelo explicável construídos sobre o formalismo de teoria da probabilidade, capazes de executar com transparência tarefas de previsão, classificação e descobertas de sistemas baseadas em dados. No entanto, o aprendizado de Redes Bayesianas baseado em dados não é observável, dificultando a seleção de modelos quando não se há conhecimentos prévios do sistema. Este trabalho apresenta uma nova metodologia de aprendizado de Redes Bayesianas baseado em Monte Carlo via Cadeias de Markov, que fornece informações da credibilidade do modelo ao final do processo de aprendizado. A avaliação desta nova metodologia mostrou ser equivalentes a algoritmos tradicionais de aprendizagem de Redes Bayesianas, mas com maior observabilidade do modelo obtido e com benefícios no aprendizado com poucos dados.
Title in English
A data-driven Bayesian methodology for Bayesian Network structure learning
Keywords in English
Bayesian Networks
Markov chain Monte Carlo
Structure learning
Abstract in English
Transparency in Artificial Intelligence algorithms is the focus of a new branch of research, called Explainable Artificial Intelligence, which suggests that specialists must be capable of replicating the results found by models, and also updating it when the results are not in agreement with their previous knowledge of the analyzed domain. Bayesian Networks are a type of explainable models built upon the formalism of probability theory, capable of executing in a transparent manner tasks of data-driven forecasting, classification and systems identification. However, data-driven Bayesian Networks learning is not observable, which makes it difficult to select models in cases when there is not previous knowledge available of the system. This work presents a new Bayesian Networks learning methodology based in Markov chain Monte Carlo, which provides model credibility at the end of the learning process. The evaluation of this new methodology showed it is equivalent to traditional Bayesian Networks learning algorithms, but with greater observability of the obtained model and with improvement in the learning with less data.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2023-04-11
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.