No 3 (2024)
Articles
Architecture of a three-dimensional convolutional neural network for detecting the fact of falsification of a video sequence
Abstract
The article reflects the use of neural network technologies to determine the facts of falsification of the contents of video sequences. In the modern world, new technologies have become an integral part of the multimedia environment, but their proliferation has also created a new threat – the possibility of misuse to falsify the contents of video sequences. This leads to serious problems, such as the spread of fake news and misinformation of society. The scientific article examines this problem and determines the need to use neural networks to solve it. In comparison with other existing models and approaches, neural networks have high efficiency and accuracy in detecting video data falsification due to their ability to extract complex features and learn from large amounts of source data, which is especially important when reducing the resolution of the analyzed video sequence. Within the framework of this work, a mathematical model for identifying the falsification of audio and video sequences in video recordings is presented, as well as a model based on a three-dimensional convolutional neural network to determine the fact of falsification of a video sequence by analyzing the contents of individual frames. Within the framework of this work, it was proposed to consider the problem of identifying falsifications in video recordings as a joint solution to two problems: identification of falsification of audio and video sequences, and the resulting problem itself was transformed into a classical classification problem. Any video recording can be assigned to one of the four groups described in the work. Only the videos belonging to the first group are considered authentic, and all the others are fabricated. To increase the flexibility of the model, probabilistic classifiers have been added, which allows to take into account the degree of confidence in the predictions. The peculiarity of the resulting solution is the ability to adjust the threshold values, which allows to adapt the model to different levels of rigor depending on the task. The architecture of a three-dimensional convolutional neural network, including a preprocessing layer and a neural network layer, is proposed to determine fabricated photoreceads. The resulting model has a sufficient degree of accuracy in determining falsified video sequences, taking into account a significant decrease in frame resolution. Testing of the model on a training dataset showed the proportion of correct detection of video sequence falsification above 70%, which is noticeably better than guessing. Despite the sufficient accuracy, the model can be refined to more significantly increase the proportion of correct predictions.
Software systems and computational methods. 2024;(3):1-11
1-11
Configuration of memory-oriented motion control system
Abstract
The paper investigates the possibilities of configuring the control cycle, i.e., determining the distribution of time intervals required for the execution of individual control operations across execution threads, which ensures the realizability of control. The object of research in this article are control systems with object-oriented architecture, assuming a combined vertical-horizontal integration of functional blocks and modules that distribute all control tasks among themselves. This architecture is realized by means of an actor instrumental model using metaprogramming. Such control systems are best at reducing control cycle time by performing computational and other control operations in parallel. Several approaches to control cycle configuration are considered: without optimization, with combinatorial optimization in time, with combinatorial optimization in system resources. Also, achieving a near-optimal configuration can be achieved by using adaptive configuration. Research shows that the control system cycle configuration problem has several solutions. Practical obtaining a solution to the configuration problem in the case of combinatorial optimization is associated with significant difficulties due to the high algorithmic complexity of the problem and a large amount of required computations, rapidly growing as the number of operations at the stages of the control cycle. A possible means of overcoming these difficulties is the use of stochastic methods, which sharply reduce the required amount of computation. Also, a significant reduction in the complexity of the task of configuring the control system cycle can be achieved by using adaptive configuration, which has two variants of realization. The first variant is the real-time configuration of the control system cycle. The second variant is the determination of quasi-optimal configuration on the basis of multiple configurations with different initial data and subsequent comparison of the obtained results.
Software systems and computational methods. 2024;(3):12-25
12-25
Fine-tuning neural networks for the features of a dataset in the speaker verification task using transfer learning
Abstract
The subject of this study is neural networks, trained using transfer learning methods tailored to the specific characteristics of the dataset. The object of the study is machine learning methods used for solving speaker verification tasks. The aim of the research is to improve the efficiency of neural networks in the task of speaker verification. In this work, three datasets in different languages were prepared for the fine-tuning process: English, Russian, and Chinese. Additionally, an experimental study was conducted using modern pre-trained models ResNetSE34L and ResNetSE34V2, aimed at enhancing the efficiency of neural networks in text-independent speaker verification. The research methodology includes assessing the effectiveness of fine-tuning neural networks to the characteristics of the dataset in the speaker verification task, based on the equal error rate (EER) of Type I and Type II errors. A series of experiments were also conducted, during which parameters were varied, and layer freezing techniques were applied. The maximum reduction in the equal error rate (EER) when using the English dataset was achieved by adjusting the number of epochs and the learning rate, reducing the error by 50%. Similar parameter adjustments with the Russian dataset reduced the error by 63.64%. When fine-tuning with the Chinese dataset, the lowest error rate was achieved in the experiment that involved freezing the fully connected layer, modifying the learning rate, and changing the optimizer—resulting in a 16.04% error reduction. The obtained results can be used in the design and development of speaker verification systems and for educational purposes. It was also concluded that transfer learning is effective for fine-tuning neural networks to the specific characteristics of a dataset, as a significant reduction in EER was achieved in the majority of experiments, indicating improved speaker recognition accuracy.
Software systems and computational methods. 2024;(3):26-36
26-36
The evolution of the Semantic Web technologies: problems and prospects
Abstract
The article is devoted to the consideration of key Semantic Web technologies, the analysis of their features, problematic aspects and growth points, which seems especially relevant in the context of import substitution and improving national information security. Special attention is paid to RDF graphs, which are based on an ontology-oriented approach, as well as the OWL language as the main tool for organizing machine-readable data structures with complex relationships between entities, a hierarchy of classes and properties. Attention is also paid to the limitations associated with the security of semantic databases, the need for their simplification, standardization and development of specialized software that meets usability criteria are analyzed. In addition, the prospects for further improvement of these technologies in the context of the Internet of Things and artificial intelligence are outlined. The article uses a comprehensive methodological framework, which implies the use of mainly general scientific methods, in particular, systematic and analytical. The article summarizes and analyzes current developments related to the Semantic Web technologies, which made it possible to identify a number of problems that need to be solved. First of all, the tools available today often have a high entry threshold, are characterized by an excessively complex, featureless interface without functions of complementary prompts and query visualization. Moreover, the Semantic Web languages need standardization and the introduction of a common protocol in order to simplify the process of working with multiformat data aggregated from different sources. Other important issues are ensuring the reliability and relevance of information, its integrity and confidentiality, as well as the contextual conditionality of logical conclusions and compliance with user requests. Among the key prospects is the creation of an intelligent autonomous environment in which devices can freely exchange data and interact with each other at the semantic level in order to provide high-quality personalized services. The provisions of the article can be taken as a basis for the development of domestic systems for structuring and describing data available for machine processing, as well as specialized lecture courses in higher education institutions.
Software systems and computational methods. 2024;(3):37-43
37-43
On clarifying the concept of "power of attorney" of artificial intelligence systems
Abstract
The subject of the study is the concept of "power of attorney" of artificial intelligence controlling robotic means of varying degrees of autonomy. The relevance of the choice of the subject of research as the principles of the use of robotic systems for various purposes, including in a group, and the object of research, which are algorithmic problems arising in the implementation of group action algorithms, is determined by the existing contradiction between the need for joint use of robotic systems, primarily autonomous, and the complexity of the software implementation of this requirement. The implementation of robotics generates certain technological problems related to the efficiency and safety of algorithmic support of autonomous and controlled robotic systems. The manifestation of such problems may be application errors that reduce the effectiveness of joint actions. In robotics, the main potential reason for the appearance of such a situation is the insufficient effectiveness of existing group application control algorithms, determined by the low level of problem study. The article formulates a list of typical situations that determine the use of autonomous and controlled robots in a group with a leader. On the basis of the proposed classification, possible algorithms for controlling movement in the group are analyzed: both calculations for target maneuvering and for ensuring mutual security. The main situations concerning the types of maneuver are formulated and the mathematical apparatus for their calculations is described. Based on the overview analysis of typical algorithms for controlling movement in space, the formulation of the scientific problem of solving the problem of developing group algorithms and mathematical methods is synthesized.
Software systems and computational methods. 2024;(3):44-54
44-54
Analysis of Idioms in Neural Machine Translation: A Data Set
Abstract
There has been a debate in various circles of the public for decades about whether a "machine can replace a person." This also applies to the field of translation. And so far, some are arguing, others are "making a dream come true." Therefore, now more and more research is aimed at improving machine translation systems (hereinafter MP). To understand the advantages and disadvantages of MP systems, it is necessary, first of all, to understand their algorithms. At the moment, the main open problem of neural machine translation (NMP) is the translation of idiomatic expressions. The meaning of such expressions does not consist of the meanings of their constituent words, and NMT models tend to translate them literally, which leads to confusing and meaningless translations. The research of idioms in the NMP is limited and difficult due to the lack of automatic methods. Therefore, despite the fact that modern NMP systems generate increasingly high-quality translations, the translation of idioms remains one of the unsolved tasks in this area. This is due to the fact that idioms, as a category of verbose expressions, represent an interesting linguistic phenomenon when the general meaning of an expression cannot be made up of the meanings of its parts. The first important problem is the lack of special data sets for learning and evaluating the translation of idioms. In this paper, we solve this problem by creating the first large-scale dataset for translating idioms. This data set is automatically extracted from the German translation corpus and includes a target set in which all sentences contain idioms, and a regular training corpus in which sentences containing idioms are marked. We have released this dataset and are using it to conduct preliminary experiments on NMP as a first step towards improving the translation of idioms.
Software systems and computational methods. 2024;(3):55-63
55-63
Numerical methods for finding the roots of polynomials with real and complex coefficients
Abstract
The subject of the article is the consideration and analysis of a set of algorithms for numerically finding the roots of polynomials, primarily complex ones based on methods for searching for an approximate decomposition of the initial polynomials into multipliers. If the numerical finding of real roots usually does not cause difficulties, then a number of difficulties arise with finding complex roots. This article proposes a set of algorithms for sequentially finding multiple roots of polynomials with real roots, then real roots by highlighting intervals that potentially contain roots and obviously do not contain them, and then complex roots of polynomials. To find complex roots, an iterative approximation of the original polynomial by the product of a trinomial by a polynomial of a lesser degree is used, followed by the use of the tangent method in the complex domain in the vicinity of the roots of the resulting trinomial. To find the roots of a polynomial with complex coefficients, we propose a solution to an equivalent problem with real coefficients. The implementation of the tasks is carried out by step-by-step application of a set of algorithms. After each stage, a group of roots is allocated and the same problem is solved for a polynomial of lesser degree. The sequence of the proposed algorithms makes it possible to find all the real and complex roots of the polynomial. To find the roots of a polynomial with real coefficients, an algorithm is constructed that includes the following main steps: determining multiple roots with a corresponding decrease in the degree of the polynomial; allocating a range of roots; finding intervals that are guaranteed to contain roots and finding them, after their allocation, it remains to find only pairs of complex conjugate roots; iterative construction of trinomials that serve as an estimate of the values of such pairs with minimal the accuracy sufficient for their localization; the actual search for roots in the complex domain by the tangent method. The computational complexity of the proposed algorithms is polynomial and does not exceed the cube of the degree of the polynomial, which makes it possible to obtain a solution for almost any polynomials arising in real problems. The field of application, in addition to the polynomial equations themselves, is the problems of optimization, differential equations and optimal control that can be reduced to them.
Software systems and computational methods. 2024;(3):64-76
64-76