Randomized Machine Learning of Nonlinear Models with Application to Forecasting the Development of an Epidemic Process

We develop a discrete approach in the theory of randomized machine learning that is aimed at application to nonlinear models. We formulate the problem of entropy estimation of probability distributions and measurement noise for discrete nonlinear models. Issues related to the application of such models to forecasting problems, in particular, the problem of generating entropy-optimal distributions, are considered. The proposed methods are demonstrated on the solution of the problem of forecasting the total number of persons infected with novel coronavirus SARS-CoV-2 in Germany in 2020.