• 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.3.2020.tde-31032021-163248
Document
Author
Full name
Lucas Francisco Amaral Orosco Pellicer
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2020
Supervisor
Committee
Pait, Felipe Miguel (President)
Fernandes Neto, Fernando
Verri, Filipe Alves Neto
Title in Portuguese
Otimização de hiperparâmetros de modelos machine learning com BarySearch.
Keywords in Portuguese
Aprendizado computacional
Computação evolutiva
Hibridização
Abstract in Portuguese
Obter bons desempenhos de modelos de aprendizado de máquina geralmente requer que os hiperparâmetros sejam ajustados. Entretanto, é complicado encontrar uma função matemática bem definida entre os valores do hiperparâmetro e o desempenho do modelo. A coleta de valores de desempenho do modelo é cara e ainda o comportamento das funções de desempenho tende a ser imprevisível, com muitas regiões de oscilação ou mesmo regiões descontínuas. Além disso os hiperparâmetros podem ser contínuos, discretos, categóricos ou condicionais, o que torna o problema de ajuste de hiperparâmetros complexo. Muitas técnicas foram desenvolvidas para solucionar esse problema. Neste trabalho, é apresentada a técnica BarySearch: um método livre de derivação com baixo custo computacional. Essa técnica apresenta um bom compromisso entre resultado e esforço de execução. O BarySearch utiliza do método do baricentro já utilizado em sintonização de controladores. O método apresenta características de convergências interessantes e pode comportar similar a um gradiente descendente ou métodos evolucionários sob suposições razoáveis.
Title in English
Hyperparameter optimization of macnhine learning models with BarySearch.
Keywords in English
Hybridization
Hyperparameters
Machine learning
Optimization
Searching methods
Abstract in English
Obtaining good models in machine learning applications often requires hyperparameters to be tuned. However, it is rarely possible to find a well-defined mathematical relationship between the values of the hyperparameters and the performance of the model. Performance evaluation is computationally expensive, the behavior of the objective function tends to be unpredictable and oscillatory, and hyperparameters may not be continuous, which makes the tuning process a challenging optimization problem. Many optimization techniques have been developed to solve this complicated problem. In this work, we present the BarySearch technique: a derivative-free optimization technique with a low computational cost. This technique exhibits a good compromise between performance and execution effort. BarySearch uses the barycenter method, already used in some controller optimization techniques. The method has interesting properties in terms of convergence and may behave in a manner similar to the descending gradient or evolutionary methods under reasonable assumptions.
 
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
2021-04-01
 
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.