Keywords
real-time neural network, neural network self-adaptation, dynamic neural network (DNN), topologies representing network (TRN), elliptical Gaussian domain of neurons, radial basis function network (RBFN), probability density function (PDF)
Real Time Dynamic Neural Network
Abstract
This paper describes a new neural network able to adapt itself, both its parameters and its structure, to a data set in real-time conditions. The adaptation is based on a non-supervised learning procedure. The new neural network can automatically create interconnections between neurons using any generic activation function. Still another important feature of this new neural network is the use of few neurons to make a good prediction using a reduced number of examples. This is relevant in order to make fast calculations using few resources in real-time applications. Some examples using this neural network are included in order to demonstrate its good performance. These examples use elliptical Gausian functions as domains for the neurons.
This paper describes a new neural network able to adapt itself, both its parameters and its structure, to a data set in real-time conditions. The adaptation is based on a non-supervised learning procedure. The new neural network can automatically create interconnections between neurons using any generic activation function. Still another important feature of this new neural network is the use of few neurons to make a good prediction using a reduced number of examples. This is relevant in order to make fast calculations using few resources in real-time applications. Some examples using this neural network are included in order to demonstrate its good performance. These examples use elliptical Gausian functions as domains for the neurons.
This paper describes a new neural network able to adapt itself, both its parameters and its structure, to a data set in real-time conditions. The adaptation is based on a non-supervised learning procedure. The new neural network can automatically create interconnections between neurons using any generic activation function. Still another important feature of this new neural network is the use of few neurons to make a good prediction using a reduced number of examples. This is relevant in order to make fast calculations using few resources in real-time applications. Some examples using this neural network are included in order to demonstrate its good performance. These examples use elliptical Gausian functions as domains for the neurons.
Información detallada
Materias, derechos, colecciones e identificadores