Skip to Main Content
The process of learning is manifested by the modification of an organism's response to a given set of input stimuli. This altered response to brought about by a gradual change in the neural logic of the animal's nervous system. The authors show that gradual changes in logic can be achieved by the use of digital and analog properties of the natural prototype. A two-input, one-output neural network is described which gives a continuum of logic functions, including the analog equivalent for each of the sixteen binary functions. This multifunction response is accomplished by varying four interconnecting weighting elements which control the excitatory and inhibitory signals to the three neurons of the network. The logic capabilities of the basic network can be increased by replacing some of its fixed weights with variable ones and expanding the network to accommodate additional input signals. A simple procedure has been developed which automatically sets the weighting elements in a reinforcement learning process. Rapid convergence to the desired logic function is achieved. It is shown that human learning and behavior can be approximated by expanding the flexible neural logic technique to functional networks.