The author describes the electronic architecture and dynamic signal processing capabilities of artificial dendritic trees which comprise simple microcircuits of artificial synapses closely modeled after their biological namesakes. It is demonstrated that powerful processing of dynamic signals can be accomplished using these structures, which do not use complex computational devices such as multipliers, adders, look-up-tables, or microprocessors. Conventional weights are replaced with connections which, when combined with the nonlinear behavior of nearby synapses and the nearly linear behavior of widely separated synapses, provide an arbitrarily precise signal processing sensitivity. The connection space is extremely large and is a factorial function of the number of synapses and sensory inputs. Because of the electrical properties of the artificial dendritic tree and the very large connection space, the circuits are virtually immune to variations in the fabrication process, which makes them suitable for implementing on a wafer scale.<<ETX>>
[1]
J. Hopfield,et al.
Computing with neural circuits: a model.
,
1986,
Science.
[2]
W. Rall.
Distinguishing theoretical synaptic potentials computed for different soma-dendritic distributions of synaptic input.
,
1967,
Journal of neurophysiology.
[3]
J. G. Elias,et al.
Silicon implementation of an artificial dendritic tree
,
1992,
[Proceedings 1992] IJCNN International Joint Conference on Neural Networks.
[4]
Terrence J. Sejnowski,et al.
Parallel Networks that Learn to Pronounce English Text
,
1987,
Complex Syst..
[5]
M. Berry,et al.
The application of network analysis to the study of branching patterns of large dendritic fields
,
1976,
Brain Research.
[6]
J. G. Elias,et al.
A generic algorithm for training networks with artificial dendritic trees
,
1992,
[Proceedings 1992] IJCNN International Joint Conference on Neural Networks.