Title :
A study of extended learning strategies for bidirectional associative memories
Author :
Neubauer, Andreas
Author_Institution :
Dept. of Commun. Eng., Duisburg Gerhard-Mercator-Univ., Duisburg
Abstract :
This paper presents a study of extended learning mechanisms for Kosko´s bidirectional associative memory-a classical artificial neural network with applications to, for example, pattern recognition and image processing. The incorporation of sigma-pi (ΣΠ)-units instead of the conventional Σ-units in the BAM is considered leading to the ΣΠ-BAM. A Hebbian learning algorithm for ΣΠ-units is proposed and simulation results are given, indicating the increased performance of the ΣΠ-BAM as a pattern association device
Keywords :
Hebbian learning; content-addressable storage; neural nets; BAM; Hebbian learning algorithm; artificial neural network; bidirectional associative memories; extended learning strategies; image processing; pattern association device; pattern recognition; Artificial neural networks; Computed tomography; Digital signal processing; Hebbian theory; Image processing; Learning systems; Magnesium compounds; Pattern recognition; Signal processing algorithms; Transfer functions;
Conference_Titel :
Neural Networks, 1995. Proceedings., IEEE International Conference on
Conference_Location :
Perth, WA
Print_ISBN :
0-7803-2768-3
DOI :
10.1109/ICNN.1995.488889
