Title :
Circuit implementation of a nonmonotone activation function
Author_Institution :
Polytech. Inst. Iasi, Romania
Abstract :
Associative memories with nonmonotone dynamics exhibiting improved performances have recently been reported. They show greater memory capacity, better recollection ability and fewer spurious states. This paper presents two circuits implementing this activation function: one is based on the use of the analytical expression of the function, and the other is a more intuitive realization. The circuits presented may offer a solution for the implementation of nonmonotone activation functions in analogue artificial neural networks.
Keywords :
analogue processing circuits; content-addressable storage; network synthesis; neural nets; analogue neural networks; associative memories; nonmonotone activation function; transfer function; Associative memory; Circuit simulation; Data preprocessing; Joining processes; Rectifiers; Stress; Transfer functions; Voltage;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.714057