Tuesday, October 8 11:30 – 13:00

Lecture-hall Ауд. 4.18 (5.17)

Chair: Prof. DOLENKO SERGEY

1

2

2. STANKEVICH S. A.

In the given lection development ways of dialog systems are discussed. The special attention is taken away to consideration existing architectures of dialog systems and possibility of building on their basis cognitive dialog systems. One of directions of development of such systems is connected with development cognitive means of dialog management. It is shown that development of such means is topical for modern humanoid robots, which have not only anthropomorphic form, but should carry out conversation with human mentality. Probabilities of natural language processing based on deep neural networks are discussed. Variants of dialog management using planning and incremental learning are considered. It is shown that suitable for humanoid robots behavior can be achieved using associative means. Variant of cognitive dialog system for humanoid robot is considered..

Tuesday, October 8 14:00 – 16:00

Lecture-hall Ауд. 4.18 (5.17)

Chair: Prof. TIUMENTSEV YURY

A survey of open mathematical problems in the theory of deep neural networks (such as expressiveness, robustness, convergence, capacity problems) will be presented. Practical applications of these problems will be illustrated using examples from Huawei industry projects.

4. EZHOV A. A.

The lecture "Efros effect (pattern recognition and cognitive blindness)" has been presented at the conference "Neuroinformatics-2010". The lecture has been placed and discussed on the Internet. Here we review this discussion and also consider such forms of blindness as willful and attentional blindness and also overview new examples of cogniitive blindness. We also consider the connections of these phenomena with other brain cognitive processes including the anosognosia.

5. MAKARENKO N. G.

The presentation of the dynamics of complex systems using graphs and networks has become increasingly popular. Graphs are even used to represent the correlation structures of scalar time series. But a finite sets of vertices and edges connecting them seem too simple to model distributed systems. However, if we supplement the sets with a suitable mathematical interface, we obtain a universal tool for diagnosing the dynamics of various systems. Such an interface can be borrowed from Riemannian geometry. This lecture describes how to obtain different variants of the discrete Ricci curvature, its relation with the discrete Laplacian and the Boltzmann entropy. The theory is illustrated by examples from various areas of knowledge.

Thursday, October 10 11:20 – 12:00

Lecture-hall Ауд. 4.18 (5.17)

Chair: Prof. EZHOV ALEXANDER

The retail along with the other industries is entering the new digital-era economics and control. Some specific features of this process are already observed. The logic of a transition period is associated with growing dimensions of inter-system links, as well as with acceleration of information exchange rates. This in turn requires the new approach to the decision making, based on solid math and machine learning algorithms. This lecture discusses 7 examples form modern practice of retail operations, reflecting the peculiarities of math control methods application: \item{Statistical estimate of the demand dynamics in distributed retail network with wide and varying assortment of rarely sold items;} -joint efficacy of goods and cross-services portfolio at each salepoint location with its own specifics; -the control of the portfolio of merchant wholesale discounts and premium for the volume and terms of sales; -optimal stock volume dynamis during the product lifecycle, in the case of forward replenishment and supply; -the portfolio management of credit proposals, discounts, and other demand stimulation actions; -the control of salepoints network staff under the constraints of time limits of sale operations and salaries budget in the dynamic demand conditions; -complex estimate of expenses and pricing of transport logistics contracts in the regional contractors markets.

Friday, October 11 10:30 – 12:00

Lecture-hall Физтех.Био, ауд. 107

Chair: Prof. MALSAGOV MAGOMED

This lecture is an elementary introduction into genetic algorithms. It covers the following topics: Genetic algorithms (GA) as a method for solving optimization problems. Com-parison of GA with the main classes of optimization methods: local evaluation, local gradient, bruteforce. Features of GA, their advantages and disadvantages. Terminology of GA. Methods of coding in GA. Continuous chromosomes. Enumerated chromosomes with repeating and unique genes. Basic genetic op-erators. Tasks with restriction. Multi-criterial optimization. Variety of imple-mentations of GA. The lecture is illustrated by demonstration examples.

8. TIUMENTSEV YU.V., EGORCHEV M.V.

One of the critical elements of the process of creating new engineering systems is the formation of mathematical and computer models that provide solutions to the problems of creating and using such systems. For such systems, typical is a high level of complexity of the objects and processes being modeled, their multidimensionality, non-linearity and non-stationarity, the diversity and complexity of the functions implemented by the simulated object. The solution to the problems of modeling for objects of this kind is significantly complicated by the fact that the corresponding models have to be formed in the presence of multiple and diverse uncertainties, such as incomplete and inaccurate knowledge of the characteristics and properties of the object being modeled, as well as the conditions in which the object will operate . Besides, during operation, the properties of the object being modeled may change, including sharp and significant, for example, due to equipment failures and/or structural damages. An approach to the formation of gray box models (semi-empirical models) for systems of this kind, based on combining theoretical knowledge about the object of modeling with the methods and tools of neural network modeling, is considered. As an example, we demonstrate the formation of a model for the longitudinal angular motion of a maneuverable aircraft, as well as the identification of the aerodynamic characteristics for the aircraft included in this model.