Fine-tuning restricted Boltzmann machines using quaternion-based flower pollination algorithm

Recent works spotlighted machine learning as the technology of the new century. One can see its fantastic capability in solving problems throughout several domains, ranging from computer vision to natural language processing tasks. Nevertheless, there is still a burden when designing a new architect...

ver descrição completa

Detalhes bibliográficos
Autores: Passos, Leandro Aparecido [UNESP], de Rosa, Gustavo Henrique [UNESP], Rodrigues, Douglas [UNESP], Papa, João Paulo [UNESP]
Tipo de documento: capítulo de livro
Estado:Versão publicada
Data de publicação:2020
País:Brasil
Recursos:Universidade Estadual Paulista (UNESP)
Repositório:Repositório Institucional da UNESP
Idioma:inglês
OAI Identifier:oai:repositorio.unesp.br:11449/309923
Acesso em linha:http://dx.doi.org/10.1016/B978-0-12-819714-1.00019-1
https://hdl.handle.net/11449/309923
Access Level:Acceso aberto
Palavra-chave:flower pollination algorithm
metaheuristic optimization
quaternionic hypercomplex representation
restricted Boltzmann machine
Descrição
Resumo:Recent works spotlighted machine learning as the technology of the new century. One can see its fantastic capability in solving problems throughout several domains, ranging from computer vision to natural language processing tasks. Nevertheless, there is still a burden when designing a new architecture, mainly due to its parameter setting up phase, which is often performed empirically. In the pursuit of mitigating this issue, optimization techniques have risen and taken the lead, being capable of providing feasible solutions to the problem. In this work, we introduce a quaternion-based flower pollination algorithm to fine-tune restricted Boltzmann machines. The effectiveness of the proposed approach is compared with several metaheuristic techniques, from naïve versions to quaternion-based ones. The experiments were carried out in three public datasets available in the literature and demonstrated promising results when compared with other techniques.