Vol 21, No 6 (2015)

The paper shows an incompleteness of theoretical foundations of the Computational Semantics branches called Semantic Role Labeling and Frame-Semantic Parsing. This situation is a consequence of a seeming lack of a semantic formalism allowing to describe semantic structures of complex sentences and discourses pertaining to arbitrary application domains. It is concluded that the theory of K-representations (knowledge representations) provides a formalism of the kind, determining a new class of formal languages — the class of SK-languages (standard knowledge languages). Some new expressive mechanisms of SK-languages are illustrated. The central ideas of a method of semantic parsing of natural language (NL) texts proposed by the theory of K-representations are set forth. The method employs the class of SK-languages for constructing semantic representations of texts. The final part of the paper considers the application of the method to designing NL-interfaces for software management. A file manager with a NL-interface NLC-1 (Natural Language Commander — Version One) has been developed, the system is implemented with the help of the functional programming language Haskell.

The mobile agent needs the developed receptors for existence in a dynamic environment with unknown characteristics. But the increase in the number of receptors leads to a significant increase in the complexity of the processing of the signals coming from these sensors. The problem of the dimension of the input sensory data of agent occurs. The article discusses the possibility of using of multisets to reduce the dimension of the input data vector from sensors and the possibility of using of multisets for further assessment of the situation by the mobile agent. On the example of the mobile agent reinforcement learning it is shown that the presentation of the properties of the environment in the form of multisets allows to aggregate the sensory data of agent, thereby reducing the dimension of the input data vector.

By numerical method of spatial characteristics there is solved the problem of impulsive destruction of thin-walled isotropic body. The impulse is made by point explosion in the center of body or by moving detonation point of explosive cord, located along the perimeter of the inner cavity of the body. Limit of destruction’s stress is determined by as function of velocity of detonation, thickness of walls and Puasson’s factor of a body material.

We review the new network informational system data securing aррroach based on newly created objects (file system objects and cliрboard) access control methods imрlementation, which allows to exclude object from access рolicy (with helр of created objects automatic labeling). Рractical imрlementation of such aррroach (while saving labeling directly in created file) allows to formulate and solve the task of imрlementing data (рrocessed in network informational system) access рolicy in a view of different рossible ways of data exchange between comрuters in such system. Herewith we imрlement data streams managing already within whole system. Reviewed рrotection method is based on рractical realization which was рatented by authors of "File objects access control system based on auto-labeling" solution. This solution allows to rethink known realization of both access control methods including discretionary and mandate ones. This not only dramatically simрlifies setting file objects access рolicy (by eliminating the "access object" essence from access control scheme), but also settings correct imрlementation in the general case is рrovided in the same time.

The text shows an approach to the creation of the software part of the satellite monitoring system. Also, it talks about the necessity to develop their own solutions to a wider variety of use cartographic information and other data information. The proposed conceptual approach to the creation of a monitoring system, which is based on web technology, will greatly simplify the development of such systems, aimed at addressing the problems of any possible type. In particular, realization of the main functions which any cartographical cursor — scalings and movements of the card has to realize is given in this article. Also the example ot comparison of pixels of the card to geographical coordinates of the objects which are stored in a database that allows to connect them in program functions is given. Code examples are given. The code is given in the C# language as server part and JavaScript as client part.

The algorithm of parallel calculations for a problem of the spectral-time analysis (STAN) on basic polyharmonic functions is offered. The analysis of consecutive algorithm for STAN showed that on temporary complexity it belongs to the class of exponential algorithms, it formed the basis for development of parallel algorithm. The method of a task of vectors of search frequencies is created. It is known that the speedup and efficiency of parallel algorithms heavily depends on its balancing, i. e. how evenly distributed computational load between processes executing. Balancing the developed parallel algorithm is based on partitioning the sequence of numbers of the vectors of the search frequencies at equal length sections. Introduced a splitting sequence of vectors of indices into equal parts correctly because: does not increase the amount of computation and memory required for the implementation of the algorithm; ensures uniform loading of the compute nodes by partitioning the sequence of index numbers as equal parts of the index; provides scalability when increasing the number of compute nodes by reducing the lengths of the parts of the index. In the offered parallel algorithm effective frequency search ls used. Results of computing experiment with model realization of parallel algorithm for signal STAN with amplitude and frequency modulation are given.

The paper presents experimental results of multidimensional data clustering and classification by help of Kohonen’s cellular neural network (CNN). An important feature of our study is that Kohonen’s CNN and visualization tools (self-organizing map (SOM), U-, H- and P-matrixes, coordinate maps, map of data classes) have been implemented in Microsoft Excel spreadsheet, without programming in VBA. The user interface of this spreadsheet model makes it easy to change the configurable parameters and visually observe the neural network learning process and data classification results using cluster grouping method. The experimental studies have shown high temporal efficiency of Kohonen’s CNN learning algorithm. Having the time complexity of the algorithm is T = O(n•N•d•i), where n — the number of neurons, N — the size of training sample, d — the input space dimension, i — the number of training epochs, the experimental run time of one training epoch was t ≈ 100 s for n = 400, N = 8000, d = 4. That is, the learning algorithm of Kohonen’s CNN works in Excel about 2 times slower than the classic learning algorithm in the SOM_PAK program. This slight training time increase in Excel is repeatedly paid by clarity, analysis flexibility and effective visualization of multi-dimensional data clustering patterns. A useful application of the edge effect and multilinked SOMs for reliable identification of clusters grouping boundaries in linear and nonlinear separable input spaces and the possibility of solving a well-known problem of "dead" neurons are shown.