Volume 21, Nº 3 (2022)
Robotics, automation and control systems
Trajectory Planning Algorithms in Two-Dimensional Environment with Obstacles
Resumo
This article proposes algorithms for planning and controlling the movement of a mobile robot in a two-dimensional stationary environment with obstacles. The task is to reduce the length of the planned path, take into account the dynamic constraints of the robot and obtain a smooth trajectory. To take into account the dynamic constraints of the mobile robot, virtual obstacles are added to the map to cover the unfeasible sectors of the movement. This way of accounting for dynamic constraints allows the use of map-oriented methods without increasing their complexity. An improved version of the rapidly exploring random tree algorithm (multi-parent nodes RRT – MPN-RRT) is proposed as a global planning algorithm. Several parent nodes decrease the length of the planned path in comprise with the original one-node version of RRT. The shortest path on the constructed graph is found using the ant colony optimization algorithm. It is shown that the use of two-parent nodes can reduce the average path length for an urban environment with a low building density. To solve the problem of slow convergence of algorithms based on random search and path smoothing, the RRT algorithm is supplemented with a local optimization algorithm. The RRT algorithm searches for a global path, which is smoothed and optimized by an iterative local algorithm. The lower-level control algorithms developed in this article automatically decrease the robot’s velocity when approaching obstacles or turning. The overall efficiency of the developed algorithms is demonstrated by numerical simulation methods using a large number of experiments.
Informatics and Automation. 2022;21(3):459-492
459-492
Artificial intelligence, knowledge and data engineering
Application of a Compartmental Spiking Neuron Model with Structural Adaptation for Solving Classification Problems
Resumo
The problem of classification using a compartmental spiking neuron model is considered. The state of the art of spiking neural networks analysis is carried out. It is concluded that there are very few works on the study of compartmental neuron models. The choice of a compartmental spiking model is justified as a neuron model for this work. A brief description of such a model is given, and its main features are noted in terms of the possibility of its structural reconfiguration. The method of structural adaptation of the model to the input spike pattern is described. The general scheme of the compartmental spiking neurons’ organization into a network for solving the classification problem is given. The time-to-first-spike method is chosen for encoding numerical information into spike patterns, and a formula is given for calculating the delays of individual signals in the spike pattern when encoding information. Brief results of experiments on solving the classification problem on publicly available data sets (Iris, MNIST) are presented. The conclusion is made about the comparability of the obtained results with the existing classical methods. In addition, a detailed step-by-step description of experiments to determine the state of an autonomous uninhabited underwater vehicle is provided. Estimates of computational costs for solving the classification problem using a compartmental spiking neuron model are given. The conclusion is made about the prospects of using spiking compartmental models of a neuron to increase the bio-plausibility of the implementation of behavioral functions in neuromorphic control systems. Further promising directions for the development of neuromorphic systems based on the compartmental spiking neuron model are considered.
Informatics and Automation. 2022;21(3):493-520
493-520
Experimental Study of Language Models of "Transformer" in the Problem of Finding the Answer to a Question in a Russian-Language Text
Resumo
The aim of the study is to obtain a more lightweight language model that is comparable in terms of EM and F1 with the best modern language models in the task of finding the answer to a question in a text in Russian. The results of the work can be used in various question-and-answer systems for which response time is important. Since the lighter model has fewer parameters than the original one, it can be used on less powerful computing devices, including mobile devices. In this paper, methods of natural language processing, machine learning, and the theory of artificial neural networks are used. The neural network is configured and trained using the Torch and Hugging face machine learning libraries. In the work, the DistilBERT model was trained on the SberQUAD dataset with and without distillation. The work of the received models is compared. The distilled DistilBERT model (EM 58,57 and F1 78,42) was able to outperform the results of the larger ruGPT-3-medium generative network (EM 57,60 and F1 77,73), despite the fact that ruGPT-3-medium had 6,5 times more parameters. The model also showed better EM and F1 metrics than the same model, but to which only conventional training without distillation was applied (EM 55,65, F1 76,51). Unfortunately, the resulting model lags further behind the larger robert discriminative model (EM 66,83, F1 84,95), which has 3,2 times more parameters. The application of the DistilBERT model in question-and-answer systems in Russian is substantiated. Directions for further research are proposed.
Informatics and Automation. 2022;21(3):521-542
521-542
Data Analysis and Visualization in the Tasks of the Project Solutions Multicriteria Optimization
Resumo
The accumulation of data on project management processes and standard solutions has made relevant research related to the use of knowledge engineering methods for a multi-criteria search for options that set optimal settings for project environment parameters. Purpose: Development of a method for searching and visualizing groups of projects that can be evaluated based on the concept of dominance and interpreted in terms of project variables and performance indicators. Methods: The enrichment of the sample while maintaining an implicit link between the project variables and performance indicators is carried out using a predictive neural network model. A set of genetic algorithms is used to detect the Pareto front in the multidimensional criterion space. The ontology of projects is determined after clustering options in the solution space and transforming the cluster structure into the criterion space. Automation of the search in the multidimensional space of the Pareto front greatest curvature zone, which determines the equilibrium design solutions, their visualization and interpretation are carried out using a tree map. Results: A tree map is constructed at any dimension of the criterion space and has a structure that has a topological correspondence with projections of shared cluster images from a multidimensional space onto a plane. For various types of transformations and correlations between performance indicators and project variables, it is shown that the areas of the Pareto front greatest curvature are determined either by the contents of the whole cluster or by part of the variants representing the "best" cluster. If an undivided rectangle of a cluster is adjacent to the upper right corner of a tree map, then its representatives in the criterion space are well separated from the rest of the clusters and, when maximizing performance indicators, are closest to the ideal point. All representatives of such a cluster are effective solutions. If the winning cluster contains dominant options inside the decision tree, then the ”best" cluster is represented by the remaining options that set the optimal settings for the project variables. Practical relevance: The proposed methods of searching and visualizing groups of projects can be used when choosing the conditions of resource and organizational and economic modeling of the project environment, ensuring the optimization of risks, cost, functional, and time criteria.
Informatics and Automation. 2022;21(3):543-571
543-571
Machine Learning in Base-Calling for Next-Generation Sequencing Methods
Resumo
The development of next-generation sequencing (NGS) technologies has made a significant contribution to the trend of reducing costs and obtaining massive sequencing data. The Institute for Analytical Instrumentation of the Russian Academy of Sciences is developing a hardware-software complex for deciphering nucleic acid sequences by the method of mass parallel sequencing (Nanofor SPS). Image processing algorithms play an essential role in solving the problems of genome deciphering. The final part of this preliminary analysis of raw data is the base-calling process. Base-calling is the process of determining a nucleotide base that generates the corresponding intensity value in the fluorescence channels for different wavelengths in the flow cell image frames for different synthesis sequencing runs. An extensive analysis of various base-calling approaches and a summary of the common procedures available for the Illumina platform are provided. Various chemical processes included in the synthesis sequencing technology, which cause shifts in the values of recorded intensities, are considered, including the effects of phasing / prephasing, signal decay, and crosstalk. A generalized model is defined, within which possible implementations are considered. Possible machine learning (ML) approaches for creating and evaluating models that implement the base-calling processing stage are considered. ML approaches take many forms, including unsupervised learning, semi-supervised learning, and supervised learning. The paper shows the possibility of using various machine learning algorithms based on the Scikit-learn platform. A separate important task is the optimal selection of features identified in the detected clusters on a flow cell for machine learning. Finally, a number of sequencing data for the MiSeq Illumina and Nanofor SPS devices show the promise of the machine learning method for solving the base-calling problem.
Informatics and Automation. 2022;21(3):572-603
572-603
Mathematical modeling and applied mathematics
Dynamic Model of Population Invasion with Depression Effect
Resumo
The article is devoted to the study of one of the current scenarios for thedevelopment of population processes in contemporary ecological systems. Biological invasionshave become extremely common due to climate change, economic activities to improve ecosystemproductivity, and random events. The invader does not always smoothly occupy an ecological niche,as in logistic models. The dynamics of the situations we have chosen after the introduction of analien species is extremely diverse. In some cases, the phenomenon of an outbreak of abundanceis quickly realized up to the beginning of the destruction by the species of its new range. Thedevelopment of the situation in the process of invasion depends on the superposition of bioticand abiotic factors. The dynamics of the abundance of the invader is affected by the favorableconditions and, to a greater extent, by the possibility of realizing the reproductive potential andthe resistance of the biotic environment. Counteraction develops with a delay and manifests itselfwhen the invader reaches a significant number. In the work, a continuous model of the invasiveprocess with a sharp transition to a state of population depression has been developed. The stageof the population crisis ends with the transition to equilibrium, since the resistance in the modelscenario depends adaptively and in a threshold way on the number. The problem of computationaldescription of a scenario with active but delayed environmental resistance is practically relevantfor situations of developing measures of artificial resistance to an undesirable invader. In thesolution of our model, there is a mode of prolonged stable fluctuations after exiting the depressionstage.
Informatics and Automation. 2022;21(3):604-623
604-623
Numerical Solution of the Problem of Filtering Estimates Information Impact on the Electorate
Resumo
The formulation and numerical scheme for solving the problem of filtering estimates of the informational impact of mass media on the electorate, allowing with a high degree of accuracy at a given observation interval to estimate the number of individuals in society who prefer a certain political subject (opinion), are proposed in the article. A mathematical model for assessing the information impact on the electorate during election campaigns, which boils down to solving a stochastic differential equation – the equation of state, forms the basis of the formulation of the problem. When compiling a model for filtering information impact estimates, it is proposed to reduce the study of the equation of state to a numerical solution of the Duncan–Mortensen–Zakai equation by introducing an additional observation equation, which is obtained from the equation of state when evaluating its stochastic components (observed agitation intensities) by methods of polyspectral analysis. In the projection formulation of the Galerkin method, when reducing to a system of linear differential equations and obtaining its solution in a recursive estimation scheme when sampling the analysis interval into subintervals and using the matrix exponential method, the Duncan–Mortensen–Zakai equation is solved. For a visual comparison of the effectiveness of the generated numerical solution to the problem of filtering information impact assessments, calculations were carried out on test examples.
Informatics and Automation. 2022;21(3):624-652
624-652