Redes neurais convolucionais semi-supervisionadas aplicadas a mudança de domínio.

• Ano de defesa: 2022
• Autor(a) principal: Bassani, Cristiano Neia de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Redes neurais (Computação); Programas de aprendizado; Inteligência computacional; Neural networks (Computer science); Apprenticeship programs; Computational intelligence; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/30599

Among the many difficulties observed within related areas of computational intelligence, some stand out due to their similarity of occurrence in the vast majority of cases. Within this area, and specifically within the training process of machine learning algorithms, a recurrent obstacle is obtaining adequate ones to carry out the training of algorithms, as well as the time and costly expense of manual rotation. Thus, in order to mitigate this problem, it is proposed the use of semi-supervised machine learning techniques applied to the context of changing domains in images linked to convolutional neural networks. For that, we used two semisupervised strategies (self-training and co-training), as well as the use of three public image datasets. In short, the contributions of the present work are presented as the development of different architectures of Semi-Supervised machine learning, the application of Fine-Tuning and classification of bases on different domains, the computational gains from the use of SemiSupervised, the applicability of the Transfer Learning process within Machine Learning and finally, the results obtained from the evaluation metrics applied to the different datasets used.