Software & Systems

Articles of journal № 2 at 2019 year.

Order result by: Public date | Title | Authors

21. Comparison of neural network models for classifying text fragments containing biographical information [№2 за 2019 год]
Author: Glazkova A.V.
Visitors: 5715
The paper presents the results of comparison of neural network architectures to solve the automatic text classification problem. The paper focuses on the topic classification of sentences containing bio-graphical facts. The data set was collected on the basis of Wikipedia and represented as a corpus of bi-ographical texts with meta tagging. Each sentence in the corpus is marked according to whether it con-tains any biographical information and what type of information it contains. The paper describes the results of topic classification of the corpus sentences for five neural net-work architectures. They are: feedforward network (FNN), recurrent neural network (RNN), gated re-current unit RNN (GRU), long short-term memory RNN (LSTM) and bidirectional LSTM (BLSTM). These models were trained on the word embedding features obtained for the corpus sentences words. We have tried two types of classification: binary (depending on whether a sentence contain biograph-ical information or not) and multi-class classification (definition of the type of biographical fact). LSTM and BLSTM models have shown the best results. The obtained results will be used to implement biographical information retrieval methods for natural language texts.

22. A set-theoretic model of a functional approach to intellectualization of building and construction management processes [№2 за 2019 год]
Author: R.V. Dushkin
Visitors: 6106
The paper describes a functional approach to managing the internal environment of an intelligent building. Such approach implies using and transferring unchangeable states between the peripheral equipment and central engineering systems, the absence of side effects during control actions, as well as the construction of a decentralized network of computations and decision making on an end-user device side based on the Internet of things. This allows simultaneous obtaining of all benefits of differ-ent consideration paradigms of management processes and intellectualization. Additionally, there is an possibility of emergent showing new features of the general approach to increase the degree of control-lability and the operation efficiency of controlled objects. The paper considers mathematical foundations of the organization of a distributed computing envi-ronment for implementation of a functional approach. It also briefly describes possible control scenar-ios in various operation modes of engineering systems of intelligent buildings. The authors describe a set-theoretic model of an intellectualized building management system. This model is a theoretical ba-sis for recommendations on changing the paradigm of constructing a general automated control system for an intelligent building from the automation of engineering processes to the management intellectu-alization and the creation of an artificial intelligence system that implements a complete and autono-mous cycle of building management. The functional approach complete with new technologies (artificial intelligence, decentralized da-tabases, the Internet of things) for managing such objects as intelligent buildings, allows transferring the operation of such buildings to a higher level of service availability, sustainability, environmental friendliness and comprehensive development of not only a controlled object, but also the hierarchy of its supersystems (municipality, region, state).

◄ ← Preview | 1 | 2 | 3