Deep CollabNet: Rede Deep Learning Colaborativa

• Ano de defesa: 2018
• Autor(a) principal: LIMA JUNIOR, Moisés Laurence de Freitas
• Orientador(a): ALMEIDA NETO, Areolino de
• Banca de defesa: BRAZ JUNIOR, Geraldo, SANTOS, Sérgio Ronaldo Barros dos, ALMEIDA, Will Ribamar Mendes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM CIÊNCIA DA COMPUTAÇÃO/CCET
• Departamento: DEPARTAMENTO DE INFORMÁTICA/CCET
• País: Brasil
• Palavras-chave em Português: Aprendizado profundo
• Palavras-chave em Inglês: Deep Feedforward; Deep Stacked Autoencoder; Deep Learning
• Área do conhecimento CNPq: Ciência da Computação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/2318

In order to improve the learning of deep neural networks, this work presents the CollabNet network, a new method of insertion of new layers into a Deep FeedForward neural networks, changing the traditional stacked autoencoders method. This new way of insertion is considered collaborative and seeks to improve training against approaches based on stacked autoencoders. In this new approach, the insertion of a new layer is performed in a coordinated and gradual manner, keeping under designer’s control the influence of the new layer on the training and no longer as random and stochastic as in traditional stacking. The collaboration proposed in this work consists of making the learning of the new inserted layer continues the learning obtained by the previous layers, without prejudice to the global learning of the network. In this way, the new inserted layer collaborates with the previous layers and the set of layers works in a way more aligned to the learning. CollabNet was tested in the Wisconsin Breast Cancer Dataset, obtaining satisfactory and promising results.