Neural network technology appears promising for many applications, including signal processing, image analysis, speech recognition, pattern recognition, and knowledge processing. Although a variety of paradigms have been studied, the neural network paradigms in use currently are limited to those with no specific temporal structure in the interconnections between neurons. Superimposing a pulse transmission structure in the interconnections between neurons provides a temporal structure in those interconnections. In this study, pulse superposition on neural net paradigms is being investigated. Special emphasis is being given to evaluate the limitation of current neural net paradigms in solving problems that involve temporally varying inputs and outputs. The issue of whether any increased capabilities result from pulse superposition is being examined. Any implementation advantages in building massively parallel systems using pulse transmissions is being examined. Model structures are being identified that may be useful for applications, and their feasibility is being determined. Emphasis is being given to potential applications in the classification and synthesis of temporally varying patterns, such as speech signals and moving images. An extensive simulation study on pulse superposition neural networks is also being performed. When successful, anticipated benefits include artificial neural network simulators; systems that involve analysis of time_varying data; speech analysis systems; communications systems; and, systems that include motion detection, tracking, and analysis.
