1988 AFIT neural network research

The authors provide a summary of recent research at the Air Force Institute of Technology (AFIT) in the area of neural networks. Specifically, AFIT research in the areas of error drive learning algorithm acceleration, speech recognition, target classification, time series prediction, and optical and VLSI implementations is presented. AFIT has developed and tested an algorithm which fits a curve to the error surface of a backpropagation surface to find more quickly a good set of weights. AFIT has investigated the use of momentum and second-order error-driven learning algorithms. A speech recognition system was developed using a combination of Kohonen nets and dynamic time warping. In the area of target classification, the accelerated backpropagation nets have been successfully applied. Research into predicting chaotic time series was performed. Three optical neural network architectures were designed and tested, and a VLSI implementation was investigated