System Informatics, 2025, # 29

Knowledge graphs have come a long way from a simple set of RDF triples to systems for acquiring new knowledge. While in previous years, the primary application of knowledge graphs used to be semantic search, nowadays, knowledge graphs are permeating all areas of modern industry. This paper presents a survey of new knowledge graph variants, such as virtual knowledge graphs, dynamic knowledge graphs, and executable knowledge graphs used in modern industrial applications, as well as their primary application: cognitive digital twins. The paper also briefly examines methods for constructing knowledge graphs using large language models and improving the performance of large language models through the use of knowledge graphs.

The intention of the paper is to present a framework for developing, studying and comparing testing equivalences in interleaving, step, partial-order and combined semantics with reverse, in the context of safe Dense-Time Petri Nets (TPNs). For representation of behavior of TPNs partial-order semantic of time causal processes are used. Reversibility means that single or concurrent actions can be undone only after caused actions are undone or not done yet. As result we establish relationships between testing equivalences under consideration.

Reversible computing, which has been widely studied in recent years, is an unconventional form of computing that can be performed in both directions: forward and backward. Any sequence of actions performed by the system can be later canceled for any reason (for example, in case of an error), allowing one to restore the system to its previous state, as if the canceled actions had never performed. Event structures are a fundamental model in concurrency theory, allowing us to comprehend the behavior of concurrent systems by describing system events and their relations. In the literature, there are two main approaches to constructing the semantics of transition systems for event structure models. One approach is based on configurations, i.e. sets of already executed events, and the other relies on residuals, i.e. model fragments that have not yet been executed. Configuration-based transition systems are mainly used to represent semantics and equivalences of concurrent models. Residual-based transition systems are actively involved to demonstrate the consistency between the operational and denotational semantics of algebraic calculi for concurrent processes, as well as to visualize the behavior of models. This article provides a category-theoretic characterization of these types of transition systems semantics for cause-respecting reversible prime event structures, and establishes the relationship between the semantics, which can be useful in constructing algebraic descriptions of the composition of reversible concurrent processes.

In the last 30 years, neural networks have been one of the most rapidly developing areas of artificial intelligence. They are widely used in sound and image processing, medicine, content analysis, generation problems and others. Significant increase in computing power, the ability to process large amounts of data, and the development of neural network theory itself made the advancement possible. The paper provides an analysis of the development of learning algorithms and neural network architectures from their emergence to the state-of-the-art. The most actively developing areas were reviewed including large language models, giant networks and multimodal models. Kolmogorov-Arnold networks were also indicated as perspective research area.

This paper presents an algorithm for recovering positions of expressions in source code Cloud Sisal programs. The relevance of this study is due to the importance of accurately mapping abstract nodes of the syntax tree to corresponding fragments of source code for creating development tools such as a source code editor, a visual debugger, and error diagnostic utilities. The proposed approach addresses the problem of incomplete positional information in the output of syntactic parsers, when it is difficult to modify existing tools. The paper describes a developed three-phase algorithm, which includes the stages of token sequence reconstruction, token position calculation, and abstract syntax tree node position calculation. The algorithm's asymptotic time complexity is linear relative to the input size and does not exceed O(n), where n is the number of characters in the source program.