An artificial neural net employing probability data as weights and parameters

Based on the concept of virtual lateral inhibition, a two-layered connectionist model called RX is developed. The flow of activation is described by 3N differential equations, where N is the number of upper level nodes. The model uses the probabilities of the upper, given the lower level nodes, and the lower, given the upper level nodes, as weights. Thus, no learning is involved in determining the weights. The equations contain the prior probabilities of all the nodes. These equations have been programmed using an RK4 single-step method of integration, and the model has been extensively tested with character-word data. The utility of such a probability oriented model is discussed to explain reasonable qualitative conjectures concerning the evolution of intelligence. >