• 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.76.2022.tde-17082022-100916
Document
Author
Full name
Ana Carolina Ferreira Luchesi
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2022
Supervisor
Committee
Bruno, Odemir Martinez (President)
Mesquita, Marcos Eduardo Ribeiro do Valle
Pedrini, Hélio
Title in Portuguese
Utilizando o aprendizado de máquina para análise de órbitas caóticas
Keywords in Portuguese
Aprendizado de máquina
Mapa logístico
Órbitas caóticas
Redes neurais
Teoria do caos
Abstract in Portuguese
O comportamento caótico pode ser observado nos mais diversos sistemas, incluindo órbitas planetárias, clima, mercado de ações, entre outros. Por conseguinte, investigar meios de prever o futuro de sistemas caóticos podem nos ajudar a compreender muitos dos sistemas que nos cercam. O mapa logístico, originalmente proposto como um modelo para descrever um crescimento populacional, apresenta sensibilidade a condições iniciais, o que resulta em uma imprevisibilidade no futuro desse sistema. Ele é um dos exemplos mais famosos de comportamento caótico emergindo de um sistema dinâmico simples. A escolha por estudar as órbitas caóticas do mapa logístico se deu pela sua simplicidade e grande grau de complexidade. As redes neurais são a categoria mais popular de aprendizado de máquina e alcançaram resultados estado da arte para diversas tarefas distintas. As redes neurais recorrentes são capazes de recordar entradas anteriores, sendo, portanto, as mais adequadas para lidar com dados sequenciais. Nesse trabalho, dois tipos de redes recorrentes foram utilizadas para investigar como preveem o futuro de órbitas caóticas do mapa logístico: Long Short-Term Memory (LSTM) e Echo State Network (ESN). Os resultados obtidos mostram que as ESNs são capazes de prever essas órbitas com maior acurácia que LSTMs e confirmam que são uma ferramenta promissora para desafiar a imprevisibilidade do caos.
Title in English
Using machine learning to analyze chaotic orbits
Keywords in English
Chaos theory
Chaotic orbits
Logistic map
Machine learning
Neural networks
Abstract in English
Chaotic behaviour can be observed in the most diverse systems, including planetary orbits, weather, stock market and so on. Therefore, investigating means to predict the future of chaotic systems can help us comprehend many systems that surround us. The Logistic Map, originally proposed as a model to describe a population growth, presents sensitivity to initial conditions, which leads to the unpredictability of the future of this system. It is one of the most famous examples of chaotic behaviour emerging from a simple dynamical system. In this study, the chaotic orbits of the logistic map were chosen because of both their simplicity and wide range of complexity. Neural networks are the current most popular approach to machine learning and have achieved state-of-the art results in many different tasks. Recurrent Neural Networks are able to recollect previous inputs, they are therefore more suitable to handle sequential data. In this dissertation, two types of recurrent neural networks were used to investigate how they predict future terms of chaotic orbits from the Logistic Map: Long Short-Term Memory (LSTM) and Echo State Networks (ESN). The results obtained show that ESNs are able to predict these orbits with higher accuracy than LSTMs and confirm them as promising tool to challenge the unpredictability of chaos.
 
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
2022-09-13
 
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.