• 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.100.2019.tde-17012019-092638
Document
Author
Full name
Elthon Manhas de Freitas
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2019
Supervisor
Committee
Delgado, Karina Valdivia (President)
Ferreira, Fernando Fagundes
Oliveira, Patrícia Rufino
Ribeiro, Carlos Henrique Costa
Title in Portuguese
Planejamento probabilístico sensível a risco com ILAO* e função utilidade exponencial
Keywords in Portuguese
Aversão a risco
Planejamento probabilístico
Processos de decisão de Markov
Sensibilidade a risco
Utilidade exponencial
Abstract in Portuguese
Os processos de decisão de Markov (Markov Decision Process - MDP) têm sido usados para resolução de problemas de tomada de decisão sequencial. Existem problemas em que lidar com os riscos do ambiente para obter um resultado confiável é mais importante do que maximizar o retorno médio esperado. MDPs que lidam com esse tipo de problemas são chamados de processos de decisão de Markov sensíveis a risco (Risk-Sensitive Markov Decision Process - RSMDP). Dentre as diversas variações de RSMDP, estão os trabalhos baseados em utilidade exponencial que utilizam um fator de risco, o qual modela a atitude a risco do agente e que pode ser propensa ou aversa. Os algoritmos existentes na literatura para resolver esse tipo de RSMDPs são ineficientes se comparados a outros algoritmos de MDP. Neste projeto, é apresentada uma solução que pode ser usada em problemas maiores, tanto por executar cálculos apenas em estados relevantes para atingir um conjunto de estados meta partindo de um estado inicial, quanto por permitir processamento de números com expoentes muito elevados para os ambientes computacionais atuais. Os experimentos realizados evidenciam que (i) o algoritmo proposto é mais eficiente, se comparado aos algoritmos estado-da-arte para RSMDPs; e (ii) o uso da técnica LogSumExp permite resolver o problema de trabalhar com expoentes muito elevados em RSMDPs.
Title in English
Probabilistic risk-sensitive planning with ILAO* and exponential utility function
Keywords in English
Exponential utility
Markov decision process
Probabilistic planning
Risk averse
Risk sensitive
Abstract in English
Markov Decision Process (MDP) has been used very efficiently to solve sequential decision-making problems. There are problems where dealing with environmental risks to get a reliable result is more important than maximizing the expected average return. MDPs that deal with this type of problem are called risk-sensitive Markov decision processes (RSMDP). Among the several variations of RSMDP are the works based on exponential utility that use a risk factor, which models the agent's risk attitude that can be prone or averse. The algorithms in the literature to solve this type of RSMDPs are inefficient when compared to other MDP algorithms. In this project, a solution is presented that can be used in larger problems, either by performing calculations only in relevant states to reach a set of meta states starting from an initial state, or by allowing the processing of numbers with very high exponents for the current computational environments. The experiments show that (i) the proposed algorithm is more efficient when compared to state-of-the-art algorithms for RSMDPs; and (ii) the LogSumExp technique solves the problem of working with very large exponents in RSMDPs
 
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
2019-01-22
 
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.