A Neural Architecture for the Identification of Number Sequences

This paper describes an architecture based on spatiotemporal networks that identifies sequences of numbers. This architecture incorporates an input layer that transforms (by means of a mathematical function) the system's input into a normalized vector that will be applied in a second step to a spatio-temporal network. Finally, the architecture is completed by an output layer using Grossberg's outstar units [1]. We have appreciated our system's complexity to be lower than any other existent method developed to solve problems of this type. By means of this architecture we have implemented a system that reminds a user of the telephone number of a given list, even if the user only remembers part of it, or if the given number contains a series of exchanged digits. The system processes the input and returns the selected telephone number among all the learned ones.

• Type: Conference
• Year: 1998
• Where: SBRN
• Authors: Juan Moreno García, Gabriel Sebastián, Miguel Angel Fernández, Antonio Fernández-Caballero
• Keywords: Architecture, Grossberg's Outstar Units, Input Layer, Neural Networks, SBRN 1998