The paper presents a developed model of a memory element based on a neuron with a log sigmoid transfer function. The element steadily memorizes one bit in the course of the iterations and works on the principle of the electronic SR-latch. It is proved by means of a graphical presentation that the SR-latch functionality can be realized by a single neuron only. The function Iterative gradient is defined, describing the internal iterative dynamics of the model under static external influences. The theoretic estimation of the weight coefficients of the model is confirmed by the experimentally obtained ones.
A memory model through a neural network with sr-latch functionality