Dynamics of an artificial recurrent neural network for the problem of modeling a cognitive function

The purpose of this work is to build an artificial recurrent neural network whose activity models a cognitive function relating to the comparison of two vibrotactile stimuli coming with a delay and to analyze dynamic mechanisms underlying its work. Methods of the work are machine learning, analysis of spatiotemporal dynamics and phase space. Results. Activity of the trained recurrent neural network models a cognitive function of the comparison of two stimuli with a delay. Model neurons exhibit mixed selectivity during the course of the task. In the multidimensional activity, the components are found each of which depends on a certain task parameter. Conclusion. The training of the artificial neural network to perform the funciton analogous to the experimentally observed process is accompanied by the emergence of dynamic properties of model neurons which are similar to those found in the experiment.