Author :
Lighttowler, N. ; Allen, A.R. ; Grant, H. ; Hendry, D.C. ; Spracklen, C.T.
Author_Institution :
Dept. of Eng., Aberdeen Univ., UK
Abstract :
The modular map is a fully digital implementation of an artificial neural network (ANN) inspired by the self-organising map (SOM). By utilising a combination of full custom and semi-custom design techniques a device has been developed that implements 256 neurons. While an individual device is a self-contained neural network suitable for real-time applications, these devices have been designed as building blocks for modular neural network systems to provide a scaleable implementation. These devices can be combined to form massively parallel implementations of the SOM in a way that minimises increases in training time for the network
Keywords :
application specific integrated circuits; digital integrated circuits; learning (artificial intelligence); self-organising feature maps; ANN; SOM; artificial neural network; fully digital implementation; massively parallel implementations; modular map; self-organising map; semi-custom design techniques; Artificial neural networks; Data communication; Drilling; Euclidean distance; Face recognition; Neural networks; Neurons; Real time systems; Space technology; Telemetry;
Conference_Titel :
Neural Networks, 1999. IJCNN '99. International Joint Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-5529-6
DOI :
10.1109/IJCNN.1999.831063
