No 3 (2023)

This study presents a classification and comparative analysis of intelligent system event methods for the detection of multi-step cyber-attacks. Such attacks are a sequence of interrelated steps of an attacker pursuing a specific goal of intrusion. The paper analyzes approaches to multistep cyber-attack detection based on system event learning methods, including supervised learning, unsupervised learning, and semi-supervised learning. The approaches considered are analyzed according to the following criteria: the method of extracting knowledge about scenarios of system events and attacks, the method for scenario knowledge representation, the method for security events analysis, the security problem to be solved, and the data set used. The paper gives the main advantages and disadvantages of learning-based approaches to the detection of multi-step cyberattacks, as well as possible directions of research in this area.

It is proposed to base Low-code platforms on an interdisciplinary language of spatial methodology for constructing semantic graphic forms of rules. A convergent approach is proposed for the convergence of the properties of the "intelligence of space" into the rules of a spatial methodology, which operates with the entities of space in the form of geometric forms of rules and matter. The spatial methodology will allow formally presenting the process of self-development of the natural language of the customer's reasoning into geometric forms of the rule of elementary distinction in undertanding, and the latter into semantic executable graphical algorithms for overcoming the gap or solving problems of creating physical objects. In the language of spatial methodology, the problem of formalizing the process of overcoming the gap between the image of the target matter (consciousness) and its physical realization (matter) is solved. The need for a transition from subject to spatial methodology is substantiated. A convergent approach to the formation of its rules is proposed.

The article proposes new methods for assessing the strength of emotions that affect the decision-making of an intelligent robot when optimizing a route. The methods offer approaches to assess the strength of emotion based on the use of sensory signals and the analysis of images in the choice environment. The images are formed after the processing of sensory signals and, on their basis, lines of parameters and characteristics of the reference sections of movement are created. They are compared with standards that characterize the logical-probabilistic and/or logical-linguistic parameters and characteristics of the reference sections of the routes. Algorithms for choosing the optimal routes for an intelligent robot based on the proposed methods for assessing the strength of emotions are described. 

To organize the interaction of experts in the joint solution of problems, digital platforms are increasingly used. They provide many opportunities for structuring tasks, attracting experts, assessing the competencies of experts and managing the process of joint problem solving. However, when managing the solution process, the question of choosing a visualization that provides simplicity and visibility of progress remains open. The paper analyzes the most common visualizations for managing the problem solving process, evaluates their advantages and disadvantages in relation to typical tasks available on existing services for joint problem solving. The result of the work is a model for evaluating and comparing visualizations of the problem solving process to make a choice of visualization for use on a platform for collaborative problem solving in a human-machine cloud.

The correspondence of some risk indicators to the requirement of their coordinated change with the associated preference indicators is studied for problems of comparing interval alternatives. A coordinated change is such a change, in which the value of the corresponding risk indicator increases with the growth of the preference indicator. It is shown that in methods of individual risk, the “mean-risk” type, left-sided risk indicators are coordinated for choosing the distribution mode as an indicator of preference, and also, with known limitations, for choosing the distribution median as a measure of preference. It has been established that the indicator of the mean semi-deviation, which is recommended as an indicator of risk for choosing the mathematical expectation of the distribution as a measure of preference, does not meet this requirement, and therefore cannot, generally speaking, be considered as adequate for problems of comparing interval alternatives.

The reinforcement learning approach offers a wide range of methods for solving problems of intelligent agents’ control. However, the problem of training an agent from sparse rewards remains relevant. One of the possible solutions is to use methods of intrinsic motivation – an idea came from developmental psychology. Intrinsic motivation explains human behavior in the absence of extrinsic control stimulate. In this article, we review the existing methods of determining intrinsic motivation based on the learned world model. The method systematization consisting of three classes is proposed. These classes differ by the application of the word model to agent components: reward system, exploration policy and intrinsic goals. We present a unified framework for describing the architecture of an agent using a world model and intrinsic motivation to improve learning. The prospects for the development in this field of study are analyzed.

The paper proposes a technology for analyzing data from camera traps using two-stage neural network processing. The task of the first stage is to separate empty images from non-empty ones. To solve the problem, a comparative analysis of the YOLOv5, YOLOR, YOLOX architectures was carried out and the most optimal detector model was identified. The task of the second stage is to classify the objects found by the detector. Models such as EfficientNetV2, SeResNet, ResNeSt, ReXNet, ResNet were compared. To train the detector model and the classifier, a data preparation approach was developed, which consists in removing duplicate images from the sample. The method was modified using agglomerative clustering to divide the sample into training, validation, and test. In the task of object detection, the YOLOv5-L6 algorithm was the best with an accuracy of 98.5% on the data set. In the task of classifying the found objects, the ResNeSt-101 architecture was the best of all with a recognition quality of 98.339% on test data.

This article presents the results of compiling a list of dissertation studies defended for the degree of candidate of sciences for the period from 2016 to 2022, the topics of which relate to the artificial intelligence field (AI). A thematic analysis of these dissertations and characteristics of the community of candidates of science belonging to the category “Artificial Intelligence Researchers” are provided. The generated list of dissertations is structured by areas of artificial intelligence technologies and subtechnologies. It is shown that the training of highly qualified scientific personnel fully meets the needs of the artificial intelligence field in the category of "AI Researchers" on the short-term planning horizon. “AI Competence Centers” are identified according to the criterion of the number of dissertations completed in scientific and educational organizations on the AI subject.