Performance assessment of capsule network on different application scenarios
View/ Open
Identifiers
URI: http://hdl.handle.net/10317/8105Share
Metrics
Statistics
View Usage StatisticsMetadata
Show full item recordAuthor
Oliva Aparicio, AntonioDirector/a
Larrey Ruiz, JorgeCenter
Escuela Técnica Superior de Ingeniería de TelecomunicaciónUniversity
Universidad Politécnica de CartagenaDepartment
Tecnologías de la Información y las ComunicacionesKnowledge Area
Teoría de la Señal y las ComunicacionesPublication date
2019-09Keywords
GeometríaGeometry
Tecnología médica
Medical technology
Abstract
[ENG]Capsule networks are newborn deep neural networks that substitute traditional artificial neurons by vectors of them called ‘capsules’. This new entities are thought to allow systems to understand the instantiation parameters of objects and induce the construction of an abstracted geometry from them. Thus, capsules enable to achieve a deeper knowledge of the object whose extrapolation procures a further generalization. These new architectures have been compared to the traditional Convolutional Neural Networks (CNNs) on four different scenarios to study their performance. Three of this scenarios are the commonly used datasets: MNIST, CIFAR-10 and SmallNORB; while the other is a set of retinographies containing non-pathological cases and others presenting glaucoma. The goal is to determine whether any of these networks is viable for an early detection of glaucoma. In addition, there is a discussion about the capacity of the structures based on capsules considered to understand and ...
Collections
The following license files are associated with this item:
Social media