On the architecture and implementation of parallel ordinal machines

A new type of parallel artificial intelligence machine is proposed. The machine learns classification rules from past example decisions of multiattribute ordinal decision-making problems, such as credit rating, employee selection, and editorial preference. These classification problems frequently occur in business, management, and social disciplines. The classification rules which are generated by the machine are consistent with each other even when the data is noisy. The resulting rules are also irredundant with respect to each other. The computation is based upon comparison operations, and no scale conversion is needed. Each processing element of the machine is very simple, and the architecture is modular. The machine carries out a learning task in time which is linear with the number of the examples in the training set. Classification is done in m gate delays, where m is the number of the classification rules. Simulation results of the algorithms on a single processor machine are presented, and suggestions regarding efficient utilization of the proposed parallel architecture are discussed