Neurocomputer interfaces and performance measures

A brief discussion of methods for interfacing neurocomputers with host computers is presented, followed by a discussion of neurocomputer performance measures. Both of these issues are central to the development of neurocomputers to support neural network research as well as the development and deployment of practical applications of neurocomputing. In using the author´s neurocomputer performance measures, a methodology that starts with operational requirements assessment is often a useful approach. The first step is to define the class of applications that are to be solved by the neurocomputer. From these, a list of neural networks of specified architectural types and sizes that will have to run (along with duty-cycle information such as the expected mean and standard deviation of the time durations of the runs for each network, as well as the expected frequency of those runs as a function of the time of day). The end result of this analysis is a table expressing the neural network implementation requirements that must be met by the neurocomputer