Real-time neural networks: conjunctoid parallel implementation

Conjunctoids are model-based neural networks for categorical data, having features that include: generality, with special cases ranging from simple perceptron-like linear versions to full-blown versions that account for all possible associations among external variables; continuous learning and performance, with provisions for optimal updating as each new datum is received, based on Bayes decision theory; and separable learning as well as performance formulas, with provisions for breaking down necessary global computations into parallel components. In the paper, a simple PC implementation is described for a full-blown conjunctoid model on a small-scale setting. A design and implementation of the model on an NCUBE parallel platform and on a special purpose parallel platform are also described