All issues

Explainable Artificial Intelligence (XAI) is a field of artificial intelligence aimed at creating methods and tools for generating interpretable and human-understandable explanations of AI decisions. The relevance of model explainability increases with the deployment of artificial intelligence in critical domains (healthcare, finance, law), where algorithmic opacity can lead to serious consequences for users and society. This work presents an analytical review of the current state of the XAI field, covering theoretical foundations, methodology, and practical applications.

The examined explainable AI methods were selected and systematized based on a multi-level classification of XAI methods by problem formulation (goal, target audience, data type), methodology (application stage, model-specificity, methods, scale), and result form (representation, presentation, evaluation metrics).

A comparative analysis of explainable AI methods for various application domains is conducted. For classical machine learning, SHAP and LIME are examined in detail, revealing their theoretical foundations, computational characteristics, and limitations. For computer vision, gradient-based methods (SmoothGrad, Integrated Gradients), activation visualization methods (Grad-CAM, Grad-CAM++), perturbation-based methods (RISE, Occlusion), and conceptual explanations (TCAV, Network Dissection) are systematized. Special attention is paid to the specifics of applying XAI to natural language processing and large language models, including analysis of the faithfulness of Chain-of-Thought reasoning, natural language explanations, and attribution graph methods. Fundamental limitations of existing approaches to LLM explainability are identified and directions for future research are defined.

The review results demonstrate that XAI methods have reached significant maturity in classical machine learning and computer vision, however, their application to large language models remains an open research problem requiring the development of new explanation paradigms.

The paper presents various modifications of the Frank–Wolfe algorithm in the equilibrium traffic assignment problem. The Beckman model is used as a model for experiments. In this article, first of all, attention is paid to the choice of the direction of the basic step of the Frank–Wolfe algorithm. Algorithms will be presented: Conjugate Frank–Wolfe (CFW), Bi-conjugate Frank–Wolfe (BFW), Fukushima Frank –Wolfe (FFW). Each modification corresponds to different approaches to the choice of this direction. Some of these modifications are described in previous works of the authors. In this article, following algorithms will be proposed: N-conjugate Frank–Wolfe (NFW), Weighted Fukushima Frank–Wolfe (WFFW). These algorithms are some ideological continuation of the BFW and FFW algorithms. Thus, if the first algorithm used at each iteration the last two directions of the previous iterations to select the next direction conjugate to them, then the proposed algorithm NFW is using more than $N$ previous directions. In the case of Fukushima Frank–Wolfe, the average of several previous directions is taken as the next direction. According to this algorithm, a modification WFFW is proposed, which uses a exponential smoothing from previous directions. For comparative analysis, experiments with various modifications were carried out on several data sets representing urban structures and taken from publicly available sources. The relative gap value was taken as the quality metric. The experimental results showed the advantage of algorithms using the previous directions for step selection over the classic Frank–Wolfe algorithm. In addition, an improvement in efficiency was revealed when using more than two conjugate directions. For example, on various datasets, the modification 3FW showed the best convergence. In addition, the proposed modification WFFW often overtook FFW and CFW, although performed worse than NFW.

Analysis of regional economic indicators plays a crucial role in management and development planning, with Gross Regional Product (GRP) serving as one of the key indicators of economic activity. The application of artificial intelligence, including neural network technologies, enables significant improvements in the accuracy and reliability of forecasts of economic processes. This study compares three neural network algorithm models for predicting the GRP of a typical region of the Russian Federation — the Udmurt Republic — based on time series data from 2000 to 2023. The selected models include a neural network with the Bat Algorithm (BA-LSTM), a neural network model based on backpropagation error optimized with a Genetic Algorithm (GA-BPNN), and a neural network model of Elman optimized using the Particle Swarm Optimization algorithm (PSO-Elman). The research involved stages of neural network modeling such as data preprocessing, training model, and comparative analysis based on accuracy and forecast quality metrics. This approach allows for evaluating the advantages and limitations of each model in the context of GRP forecasting, as well as identifying the most promising directions for further research. The utilization of modern neural network methods opens new opportunities for automating regional economic analysis and improving the quality of forecast assessments, which is especially relevant when data are limited and for rapid decision-making. The study uses factors such as the amount of production capital, the average annual number of labor resources, the share of high-tech and knowledge-intensive industries in GRP, and an inflation indicator as input data for predicting GRP. The high accuracy of the predictions achieved by including these factors in the neural network models confirms the strong correlation between these factors and GRP. The results demonstrate the exceptional accuracy of the BA-LSTM neural network model on validation data: the coefficient of determination was 0.82, and the mean absolute percentage error was 4.19%. The high performance and reliability of this model confirm its capacity to predict effectively the dynamics of the GRP. During the forecast period up to 2030, the Udmurt Republic is expected to experience an annual increase in Gross Regional Product (GRP) of +4.6% in current prices or +2.5% in comparable 2023 prices. By 2030, the GRP is projected to reach 1264.5 billion rubles.

Social media posts play an important role in demonstration of financial market state, and their analysis is a powerful tool for trading. The article describes the result of a study of the impact of social media activities on the movement of the financial market. The top authoritative influencers are selected. Twitter posts are used as data. Such texts usually include slang and abbreviations, so methods for preparing primary text data, including Stanza, regular expressions are presented. Two approaches to the representation of a point in time in the format of text data are considered. The difference of the influence of a single tweet or a whole package consisting of tweets collected over a certain period of time is investigated. A statistical approach in the form of frequency analysis is also considered, metrics defined by the significance of a particular word when identifying the relationship between price changes and Twitter posts are introduced. Frequency analysis involves the study of the occurrence distributions of various words and bigrams in the text for positive, negative or general trends. To build the markup, changes in the market are processed into a binary vector using various parameters, thus setting the task of binary classification. The parameters for Binance candlesticks are sorted out for better description of the movement of the cryptocurrency market, their variability is also explored in this article. Sentiment is studied using Stanford Core NLP. The result of statistical analysis is relevant to feature selection for further binary or multiclass classification tasks. The presented methods of text analysis contribute to the increase of the accuracy of models designed to solve natural language processing problems by selecting words, improving the quality of vectorization. Such algorithms are often used in automated trading strategies to predict the price of an asset, the trend of its movement.

The technology of creating patterns from natural language words (concepts) based on text data in the bag of words model is considered. Patterns are used to reduce the dimension of the original space in the description of documents and search for semantically related words by topic. The process of dimensionality reduction is implemented through the formation of patterns of latent features. The variety of structures of document relations is investigated in order to divide them into themes in the latent space.

It is considered that a given set of documents (objects) is divided into two non-overlapping classes, for the analysis of which it is necessary to use a common dictionary. The belonging of words to a common vocabulary is initially unknown. Class objects are considered as opposition to each other. Quantitative parameters of oppositionality are determined through the values of the stability of each feature and generalized assessments of objects according to non-overlapping sets of features.

To calculate the stability, the feature values are divided into non-intersecting intervals, the optimal boundaries of which are determined by a special criterion. The maximum stability is achieved under the condition that the boundaries of each interval contain values of one of the two classes.

The composition of features in sets (patterns of words) is formed from a sequence ordered by stability values. The process of formation of patterns and latent features based on them is implemented according to the rules of hierarchical agglomerative grouping.

A set of latent features is used for cluster analysis of documents using metric grouping algorithms. The analysis applies the coefficient of content authenticity based on the data on the belonging of documents to classes. The coefficient is a numerical characteristic of the dominance of class representatives in groups.

To divide documents into topics, it is proposed to use the union of groups in relation to their centers. As patterns for each topic, a sequence of words ordered by frequency of occurrence from a common dictionary is considered.

The results of a computational experiment on collections of abstracts of scientific dissertations are presented. Sequences of words from the general dictionary on 4 topics are formed.

Posts on social networks can both predict the movement of the financial market, and in some cases even determine its direction. The analysis of posts on Twitter contributes to the prediction of cryptocurrency prices. The specificity of the community is represented in a special vocabulary. Thus, slang expressions and abbreviations are used in posts, the presence of which makes it difficult to vectorize text data, as a result of which preprocessing methods such as Stanza lemmatization and the use of regular expressions are considered. This paper describes created simplest machine learning models, which may work despite such problems as lack of data and short prediction timeframe. A word is considered as an element of a binary vector of a data unit in the course of the problem of binary classification solving. Basic words are determined according to the frequency analysis of mentions of a word. The markup is based on Binance candlesticks with variable parameters for a more accurate description of the trend of price changes. The paper introduces metrics that reflect the distribution of words depending on their belonging to a positive or negative classes. To solve the classification problem, we used a dense model with parameters selected by Keras Tuner, logistic regression, a random forest classifier, a naive Bayesian classifier capable of working with a small sample, which is very important for our task, and the k-nearest neighbors method. The constructed models were compared based on the accuracy metric of the predicted labels. During the investigation we recognized that the best approach is to use models which predict price movements of a single coin. Our model deals with posts that mention LUNA project, which no longer exist. This approach to solving binary classification of text data is widely used to predict the price of an asset, the trend of its movement, which is often used in automated trading.

This paper is dedicated to the analysis of medical and biological data obtained through locomotor training and testing of astronauts conducted both on Earth and during spaceflight. These experiments can be described as the astronaut’s movement on a treadmill according to a predefined regimen in various speed modes. During these modes, not only the speed is recorded but also a range of parameters, including heart rate, ground reaction force, and others, are collected. In order to analyze the dynamics of the astronaut’s condition over an extended period, it is necessary to perform a qualitative segmentation of their movement modes to independently assess the target metrics. This task becomes particularly relevant in the development of an autonomous life support system for astronauts that operates without direct supervision from Earth. The segmentation of target data is complicated by the presence of various anomalies, such as deviations from the predefined regimen, arbitrary and varying duration of mode transitions, hardware failures, and other factors. The paper includes a detailed review of several contemporary retrospective (offline) nonparametric methods for detecting multiple changepoints, which refer to sudden changes in the properties of the observed time series occurring at unknown moments. Special attention is given to algorithms and statistical measures that determine the homogeneity of the data and methods for detecting change points. The paper considers approaches based on dynamic programming and sliding window methods. The second part of the paper focuses on the numerical modeling of these methods using characteristic examples of experimental data, including both “simple” and “complex” speed profiles of movement. The analysis conducted allowed us to identify the preferred methods, which will be further evaluated on the complete dataset. Preference is given to methods that ensure the closeness of the markup to a reference one, potentially allow the detection of both boundaries of transient processes, as well as are robust relative to internal parameters.

The method for mapping of intermediate representations (IR) set of C, C++ programs to vector embedding space is considered to create an empirical estimation framework for static performance prediction using LLVM compiler infrastructure. The usage of embeddings makes programs easier to compare due to avoiding Control Flow Graphs (CFG) and Data Flow Graphs (DFG) direct comparison. This method is based on transformation series of the initial IR such as: instrumentation — injection of artificial instructions in an instrumentation compiler’s pass depending on load offset delta in the current instruction compared to the previous one, mapping of instrumented IR into multidimensional vector with IR2Vec and dimension reduction with t-SNE (t-distributed stochastic neighbor embedding) method. The D1 cache miss ratio measured with perf stat tool is considered as performance metric. A heuristic criterion of programs having more or less cache miss ratio is given. This criterion is based on embeddings of programs in 2D-space. The instrumentation compiler’s pass developed in this work is described: how it generates and injects artificial instructions into IR within the used memory model. The software pipeline that implements the performance estimation based on LLVM compiler infrastructure is given. Computational experiments are performed on synthetic tests which are the sets of programs with the same CFGs but with different sequences of offsets used when accessing the one-dimensional array of a given size. The correlation coefficient between performance metric and distance to the worst program’s embedding is measured and proved to be negative regardless of t-SNE initialization. This fact proves the heuristic criterion to be true. The process of such synthetic tests generation is also considered. Moreover, the variety of performance metric in programs set in such a test is proposed as a metric to be improved with exploration of more tests generators.

The traditional approach to studying inter-journal communication involves analyzing journal citation graphs. This paper proposes a method for analyzing journal networks using a new type of bibliometric graph — a journal intersection graph based on the binary operation of set intersection — employing techniques grounded in the friendship index and the disparity function. The approach is demonstrated using a relatively small example of a real journal network, with data sourced from the All-Russian portal Math-Net.Ru information system: 63 journals from 2008–2021 meeting specific criteria, containing almost 69 thousand articles authored by 54 thousand individuals. The mathematical model of this real-world network is represented as an intersection graph using the Jaccard coefficient, which exhibits specific properties: low dimensionality, high graph density, and an edge weight distribution that is not approximated by a power law function. The obtained results include the network structure of connections within the studied set of journals, accounting for their degree of interaction, and the identification of significant vertices using the friendship index. This captures the graph’s structural properties, offers an obvious substantive interpretation, and allows for ranking journals by this metric. Thus, the method implements a tool for distinguishing between vertices that are leaders in terms of the friendship index and “network integrators” (based on closeness/betweenness centrality). It also demonstrates a qualitative change in structural properties when reducing graph density while maintaining connectivity, achieved by applying the disparity function. The sequential application of the disparity function while lowering the significance threshold allows for the identification of the graph’s core, containing the most strongly connected vertices. This, in turn, enables the determination of a set of vertices (and corresponding journals) that are simultaneously part of the core and have the highest significance according to the friendship index. An analysis of the levels of this resulting journal set within the “Belyi Spisok” (“White List”) shows these journals have a high rating. The findings provide a deeper understanding of the relationship structure within scientific journal networks and define new approaches for their study.