arXiv subjects
Publications and source records attributed to Igor Mackarov.
Forecasts of various processes have always been a sophisticated problem for statistics and data science. Over the past decades the solution procedures were updated by deep learning and kernel methods. According to many specialists, these approaches are much more precise, stable, and suitable compared to the classical statistical linear time series methods. Here we investigate how true this point of view is.
Convolutional neural networks (CNN) are known to be an effective means to detect and analyze images. Their power is essentially based on the ability to extract out images common features. There exist, however, images involving unique, irregular features or details. Such is a collection of unusual children drawings reflecting the kids imagination and individuality. These drawings were analyzed by means of a CNN constructed by means of Keras-TensorFlow. The same problem - on a significantly higher level - was solved with newly developed family of networks called IgNet that is described in this paper. It proved able to learn by 100 % all the categorical characteristics of the drawings. In the case of a regression task (learning the young artists ages) IgNet performed with an error of no more than 0.4 %. The principles are discussed of IgNet design that made it possible to reach such substantial results with rather simple network topology.
Oftentimes observed divergence of numerical solutions to benchmark flows of the UCM viscoelastic fluid is a known and widely discussed issue. Some authors consider such singularities 'invincible'. Following the previous research, the article gives more arguments against this position, for which it considers two typical flows with a stagnation point, often a place of the flows' singularity. For the flow spread over a wall, as previously for the counterflows, numerical and asymptotic analytical solutions are presented. Both kinds of flows turn out regular in the stagnation points, in particular, for high Weissenberg numbers. A good accordance is demonstrated between the analytical and numerical results.
Frequently observed divergence of numerical solutions to benchmark flows of the UCM viscoelastic fluid is a known and widely discussed issue. Some authors consider such singularities "invincible". The article argues this position, to which end it considers two typical flows with a stagnation point, often a place of the flows' singularity: counterflows and a flow spread over a wall. For the counterflows numerical and asymptotic analytical solutions are presented. Both kinds of flows turn out regular in the stagnation points, in particular, for high Weissenberg numbers. A good accordance is demonstrated between the analytical and numerical results.
The Correlates of War project scrupulously collects information about disputes between the countries over a long historical period together with other data relevant to the character and reasons of international conflicts. Using methods of modern Data Science implemented in the R statistical software, we investigate the datasets from the project. We study political, economic, and religious factors with respect to the emergence of conflicts and wars between the countries. The results obtained lead to certain conclusions about variances and causalities between the factors considered. Some unpredictable features are presented.
Behavior of the delta algorithm of autonomic switch between two component implementations is considered on several examples of a client-server systems involving, in particular, periodic change of intensities of requests for the component. It is shown that in the cases of some specific combinations of elementary requests costs, the number of clients in the system, the number of requests per unit of time, and the cost of switch between the implementations, the algorithm may reveal behavior that is rather far from the desired. A sufficient criterion of a success of the algorithm is proposed based on the analysis of the accumulated implementations costs difference as a function of time. Suggestions are pointed out of practical evaluation of the algorithm functioning regarding the observations made in this paper.