The Multi-Lane Capsule Network

Autor(es)

do Rosario, Vanderson Martins ^[1] ; Borin, Edson ^[1] ; Breternitz, Jr., Mauricio ^{[2, 3, 4]}

Afiliação do(s) autor(es)

^[1] Univ Estadual Campinas, Inst Comp, BR-13083970 Campinas, SP - Brazil

^[2] Lisbon Univ, ISTAR IUL Lab, Inst ISCTE IUL, P-1649026 Lisbon - Portugal

^[3] IST, P-1649026 Lisbon - Portugal

^[4] INESC ID, P-1649026 Lisbon - Portugal

• Tipo de documento: Artigo Científico
• Fonte: IEEE SIGNAL PROCESSING LETTERS; v. 26, n. 7, p. 1006-1010, JUL 2019.
• Citações Web of Science: 0

Resumo

We introduce multi-lane capsule networks (MLCN), which are a separable and resource efficient organization of capsule networks (CapsNet) that allows parallel processing while achieving high accuracy at reduced cost. A MLCN is composed of a number of (distinct) parallel lanes, each contributing to a dimension of the result, trained using the routing-by-agreement organization of CapsNet. Our results indicate similar accuracy with a much-reduced cost in number of parameters for the Fashion-MNIST and Cifar10 datasets. They also indicate that the MLCN outperforms the original CapsNet when using a proposed novel configuration for the lanes. MLCN also has faster training and inference times, being more than two-fold faster than the original CapsNet in a same accelerator. (AU)

• Processo FAPESP: 13/08293-7 - CECC - Centro de Engenharia e Ciências Computacionais
• Beneficiário: Munir Salomao Skaf
• Modalidade de apoio: Auxílio à Pesquisa - Centros de Pesquisa, Inovação e Difusão - CEPIDs