Abstract
An Artificial Neural Network (ANN) Module has been designed having capabilities of onboard supervised backpropagation learning. Several modules are comected through an error detection scheme. The Artificial Neural Network Module is a three-layered network comprising twenty-four digital nodes that can be fully connected. The module allows full user configurable options of connections between nodes in adjacent layers, ,activation or deactivation of any node, learning high precision floating point user specified input initial weights, and user choice of linear or nonlinear activation hnction for the input, hidden, and output nodes. In addition to the Artificial Neural Network Module implementation, a circuit that utilizes three complete modules for improved convergence rate, convergence to a deep minimum, parallel operation of the moduizs during testing, very high fault tolerance and robustness, has been implemented with klly digital circuitry.
Date of this Version
December 1992