Abstract:

In our quest to advance medical AI, we demonstrate that a pre-trained and frozen Vision Transformer paired with a linear classifier can achieve highly competitive performance in endoscopic image classification. Our central contribution is a systematic, layer-wise analysis that identifies the source of the most powerful features, challenging the common heuristic of using only the final layer. We uncover a distinct "peak-before-the-end" phenomenon, where a late-intermediate layer offers a more generalizable representation for the downstream medical task. On the Kvasir and HyperKvasir benchmarks, our parameter-light approach not only achieves excellent accuracy but also drastically reduces computational overhead. This work provides a practical roadmap for efficiently leveraging the power of general foundation models in clinical environments.

Abstract:

Automation of parallel programming is important at any stage of parallel program development. These stages include profiling of the original program, program transformation, which allows us to achieve higher performance after program parallelization, and, finally, construction and optimization of the parallel program. It is also important to choose a suitable parallel programming model to express parallelism available in a program. On the one hand, the parallel programming model should be capable to map the parallel program to a variety of existing hardware resources. On the other hand, it should simplify the development of the assistant tools and it should allow the user to explore the parallel program the assistant tools generate in a semi-automatic way. The SAPFOR (System FOR Automated Parallelization) system combines various approaches to automation of parallel programming. Moreover, it allows the user to guide the parallelization if necessary. SAPFOR produces parallel programs according to the high-level DVMH parallel programming model which simplify the development of efficient parallel programs for heterogeneous computing clusters. This paper focuses on the approach to semi-automatic parallel programming, which SAPFOR implements. We discuss the architecture of the system and present the interactive subsystem which is useful to guide the SAPFOR through program parallelization. We used the interactive subsystem to parallelize programs from the NAS Parallel Benchmarks in a semi-automatic way. Finally, we compare the performance of manually written parallel programs with programs the SAPFOR system builds.

Abstract: The system for automated parallelization SAPFOR (System FOR Automated Parallelization) includes tools for program analysis and transformation. The main goal of the system is to reduce the complexity of program parallelization. SAPFOR system is focused on the investigation of multilingual applications in Fortran and C programming languages. The low-level LLVM IR representation is used in SAPFOR for program analysis. This representation allows us to perform various IR-level optimizations to improve the quality of program analysis. At the same time, it loses some features of the program, which are available in its higher level representation. One of these features is the multi-dimensional structure of the arrays. Data dependence analysis is one of the main problems which should be solved to automate program parallelization. Moreover, such an analysis belongs to the class of NP-hard problems. Knowledge of the multidimensional structure of arrays allows in many cases to take into account the structure of index expressions in calls to arrays and reduce the complexity of the analysis. In addition, the use of multi-dimensional arrays allows us to use multi-dimensional processor matrix and to parallelize a whole loop nests, rather than a single loop in the nest. So, parallelism of a program is going to be increased. These opportunities are natively supported in the DVM system. This paper discusses the approach used in the SAPFOR system to recover the form of multi-dimensional arrays by their linearized representation in LLVM IR. The proposed approach has been successfully evaluated on various applications including performance tests from the NAS Parallel Benchmarks suite.

Abstract:

This paper introduces RusHallu-RAG, the first comprehensive benchmark for detecting hallucinations in Russian-language RAG systems. The dataset of 1,000 query-answer pairs covers general knowledge and scientific domains with controlled context relevance balancing. A fine-grained taxonomy of six hallucination types is proposed and used for annotation at both response and span levels with human experts and LLMs.

For training a hallucination detector, a balanced dataset of 3,000 examples is collected using synthetic generation methods. A detector based on YandexGPT-5-Lite-8B is trained via SFT. We evaluate 15 open-weight and proprietary models alongside the trained detector. Results show that model size does not guarantee performance: medium-sized models (20–33B) sometimes outperform larger ones. The trained detector achieves an average accuracy of 63.7%, ranking second after proprietary Gemini-2.5-Pro, with a gain of 47.2 p.p. over the baseline. The gap between proprietary and open-weight models is reduced from 20–35 to 8.4 p.p., confirming the effectiveness of the approach. The benchmark, code, and trained model are publicly released.

Abstract:

Task-specific models have long dominated natural language processing tasks. However, general-purpose large language models (LLMs) have recently begun to successfully compete with highly specialized solutions across various NLP domains. In this paper, we investigate the applicability of LLMs to the task of estimating the semantic proximity of a word's meanings in a pair of usages, known as Word-in-Context (WiC). Drawing on the multilingual CoMeDi benchmark, we propose novel approaches to building automated WiC-systems based on LLMs. We conduct a systematic comparison of five different configurations in terms of quality and computational costs. In particular, we propose a configuration where LLM predictions are adjusted using a training set, without the need to fine-tune the LLM itself. Results on test sets across seven languages show that our proposed approaches enable LLMs to outperform all existing specialized systems, establishing a new state-of-the-art (SOTA) on the CoMeDi benchmark. Nevertheless, the achieved high quality comes with a significant increase in computational costs: LLM-based systems require several orders of magnitude more computations compared to compact specialized models (such as XL-DURel). This work represents a step towards understanding the trade-off between accuracy and resource efficiency when using modern LLMs in lexical semantics tasks.

Abstract: The use of pointers and indirect memory accesses in the program, as well as the complex control flow are some of the main weaknesses of the static analysis of programs. The program properties investigated by this analysis are too conservative to accurately describe program behavior and hence they prevent parallel execution of the program. The application of dynamic analysis allows us to expand the capabilities of semi-automatic parallelization. In the SAPFOR system (System FOR Automated Parallelization), a dynamic analysis tool has been implemented, based on on the instrumentation of the LLVM representation of an analyzed program, which allows the system to explore programs in both C and Fortran programming languages. The capabilities of the static analysis implemented in SAPFOR are used to reduce the overhead program execution, while maintaining the completeness of the analysis. The use of static analysis allows to reduce the number of analyzed memory accesses and to ignore scalar variables, which can be explored in a static way. The developed tool was tested on performance tests from the NAS Parallel Benchmarks package for C and Fortran languages. The implementation of dynamic analysis, in addition to traditional types of data dependencies (flow, anit, output), allows us to determine privitizable variables and a possibility of pipeline execution of loops. Together with the capabilities of DVM and OpenMP these greatly facilitates program parallelization and simplify insertion of the appropriate compiler directives.

Abstract:

Large language models (LLMs) are increasingly being used for code generation. However, both the generated code and the underlying LLM-based systems demand rigorous security evaluation. A common approach to enhancing code generation quality involves multi-agent systems composed of multiple models. This paper evaluates the performance of GPT-OSS 20B, GPT-OSS 120B, and Qwen3-Coder 480B models in single-agent and multi-agent configurations, using two code security benchmarks, SecurityEval and CyberSecEval. The key contribution is SafeAICoder, an extensible and scalable framework for testing LLMs enabling distributed server-side generation of multi-module programs and tests, independent of client-side code.

Abstract:

Many lines have been written about the wealth of the holdings of the National Library of the Republic of Tatarstan. When speaking of the rare books stored in the Department of Manuscripts and Rare Books of our Library, it is worth mentioning the lifetime editions of some Russian classics. Many of these editions are bibliographic rarities. The history of the publication of A.S. Pushkin’s "Eugene Onegin" and "Poems", F.I. Tiutchev’s "Poems", and A.A. Akhmatova’s "The evening", is especially interesting. It would be impossible to have all the lifetime editions of famous poets and writers within one library’s holdings. Nevertheless, the staff of the Department of Manuscripts and Rare Books continue the search for such rarities among our Library’s holdings.

Abstract:

The article addresses the issue of designing a non-violent combat system for a children's RPG game. It presents the results of an empirical research. Its purpose is to obtain data to determine the balance between tactical control and direct damage mechanics.

The scientific novelty of the study lies in the empirical identification of the structure of perception by children aged 6–12 of various types of game mechanics: tactical (indirect impact on the enemy), tactical-force (neutralization without direct damage), and force (direct damage). Based on a combination of card sorting and associative analysis methods, a matrix of similarity of mechanics reflecting their categorization in children's minds was constructed, and stable emotional reactions and associations of children regarding key elements of the combat system (ice – control, wind – humor, electricity – ingenuity, fire – danger, etc.) were identified.

As a result of the study, three clusters of mechanics were identified (“Turn Skip,” “Weakening and Restriction,” “Direct Damage”) and a preference rating was compiled, in which tactical mechanics took the leading positions, while force (direct damage) mechanics were not a priority and were described by children as boring. Thus, the hypothesis regarding the preferences of the children's audience was fully confirmed: the core of a safe combat system should consist of non-violent tactical mechanics (70-80%) that ensure the intellectual engagement of players, combined with tactical-force mechanics aimed at their emotional release.

The practical significance of the study is determined by the possibility of applying the developed recommendations in the design of children's video games, as well as the adaptation of the proposed methodology for similar studies.

Abstract:

This paper presents the development of Tula, a tool for video game balancing. The necessity for such a tool is substantiated by the growing requirements for quality and cost-effectiveness in the video game industry, particularly in managing in-game economy and game world logic. The study analyzes existing tools and approaches to game balancing, identifying their limitations, which informed the design of the new tool's functionality. The presented tool integrates features of contemporary solutions while providing enhanced capabilities for game parameter analysis and testing, including prototype generation via class descriptions and real-time simulation. The technological foundation and architecture of the tool are described in detail. Key implementation aspects are discussed: interface responsiveness, continuous data synchronization, and security. Comparative analysis with Machinations revealed advantages in data processing correctness, interface convenience, and prototype modification flexibility.

Abstract:

Using Vision Transformer (ViT) models in real medical practice – for example, in hospitals or diagnostic centers – is often difficult because doctors' work computers usually do not have powerful graphics processors (GPUs), and computing resources are limited. This work investigates a complete practical pipeline for model inference, aimed at reducing computational costs without significant loss of predictive performance.

The proposed approach combines several optimization techniques. First, knowledge distillation (KD) is used, where a compact student model learns to mimic the behavior of a larger, more accurate teacher model. Second, Exponential Moving Average (EMA) of the model weights is applied to stabilize training and improve generalization. Third, post-training INT8 quantization (PTQ) is explored to reduce model size and accelerate inference. Additionally, a simplified quantization-aware training variant (QAT-lite) is considered, where the effects of quantization are partially incorporated during fine-tuning.

Experiments are conducted on the ISIC dataset, which contains dermoscopic images of skin lesions. Model performance is evaluated using standard classification metrics, including accuracy, macro-averaged F1 score, and area under the ROC curve (ROC-AUC). CPU performance is also analyzed, including inference latency, throughput, memory consumption, and the final model size.

The results show that post-training INT8 quantization preserves performance close to the FP32 baseline while substantially reducing memory and computational requirements. In contrast, QAT-lite does not consistently provide reproducible improvements over PTQ.

Abstract: Digitalization of the Russian economy requires appropriate changes in the system of education and training. Transformations in the system of secondary education should begin in the subject area «Mathematics and informatics». The paper shows the experience of training and participation of school students in the sections of mathematics, informatics, bioinformatics of scientific conferences.

Abstract:

The paper is devoted to the development and experimental verification of a modular user interface architecture for game projects in Unreal Engine. The study identifies typical UI design antipatterns (monolithic widget structures, hard references, lack of abstraction, platform dependency) and provides an analysis of existing architectural approaches (Data-Driven UI, MVC, MVVM), demonstrating their limitations. The authors propose an architecture that integrates the Common UI framework, an asynchronous loading mechanism with soft references, the Gameplay Tags system, and a data registry for dynamic settings menu generation. The architecture is implemented as the AdvancedGameUserInterface plugin. Its experimental evaluation involving 12 developers and 5 UI designers confirmed the effectiveness of the proposed solution across three key metrics: maintainability, performance, and scalability. Based on the research results, practical recommendations are formulated for applying the developed architecture depending on the scale of the game project. The research findings can be used by developers to improve quality and reduce maintenance costs for user interfaces in game projects on Unreal Engine.

Abstract:

The article is devoted to the study of the gaming experience of users and the formation of a cognitive model of their interaction with the new video game "Koshchejskie prodelki", which is at an early stage of development. The UX research methodology proposed by the authors, as well as the results obtained during it, demonstrate how qualitative data on the perception, emotions, feelings, behavior patterns and achievements of players help optimize the process of creating video games, making them not only entertaining, but also psychologically effective. The results of the work have practical value for indie developers who, with limited resources, strive to create high-quality gaming products.

Abstract:

Amidst the rapid growth of e-commerce and increasing demand for personalization, the Russian market for customized clothing faces a shortage of technological and widely accessible solutions. This paper presents the results of a research and implementation project focused on developing a multi-brand digital platform with an integrated 3D configurator, aimed at transforming the pre-order cycle. The solution enables customers to interactively create garment designs in a web environment, while allowing designers to optimize logistics and minimize overproduction.

The primary scientific and technical contribution of this work lies in its detailed description of the platform's target architecture and a scalable 3D model processing pipeline that ensures model optimization and correct browser-based rendering. An additional contribution is the developed methodology for preparing and optimizing 3D garment models for web visualization. Formalized as a set of technical requirements, this methodology achieves a balance between visual quality and performance.

As a result of this research, the authors have addressed the challenge of unifying 3D model formats from different designers within a multi-brand digital platform—a key distinction from existing single-brand solutions. Furthermore, the implemented technology enables the customization of 3D clothing models with interactive real-time visualization of all design modifications on a single screen.

The technological feasibility and effectiveness of the solution are substantiated by a comparative analysis of existing alternatives, a market analysis using the PAM-TAM-SAM-SOM model, and an assessment of functional requirements.

The article also outlines a practical strategy for implementing the digital platform, making it a valuable resource for researchers and practitioners working at the intersection of e-commerce, computer graphics, and the digital transformation of business processes.

Abstract:

This article presents a comprehensive analysis of the application of virtual reality (VR) technologies in medical education, with a particular focus on obstetrics and gynecology. It examines the pedagogical rationale for introducing VR (constructivism, Kolb’s experiential learning theory, cognitive load theory), modern technological platforms with varying levels of immersion, as well as the geography and practice of introducing VR technologies in leading medical universities in Russia and around the world. Particular attention is paid to specialized VR simulators for practicing obstetric care, emergency obstetric conditions (shoulder dystocia, postpartum hemorrhage), and minimally invasive gynecological interventions. The results of randomized controlled studies confirming the effectiveness of VR training in the formation of psychomotor skills, clinical thinking, and team interaction are presented. The main barriers to the implementation of VR are analyzed: financial, technical, personnel, as well as specific problems of cyber sickness and ethical challenges. Organizational approaches to expanding the use of VR are reviewed: models for the collective use of equipment, the development of verified educational content repositories, and integration with learning management systems. The potential for convergence of VR with artificial intelligence (AI) to create adaptive, personalized educational environments is substantiated.

Abstract:

The incredible growth in the popularity of digital technologies and education has led to the emergence of “computational thinking”. The dramatic demand and popularization of this notion ignited the redesign of educational standards and the upsurge of literature on this subject. During the extensive literature review, we found that it is not always clear what computational thinking really is, what aspects it entails, and how to think about it in simple yet robust terms. In this paper, we have examined the meaning of this concept, its importance in modern digital education, and drawn an analogy between computational thinking and the skills of writing ordinary text to facilitate its reliable understanding.

Abstract:

The article investigates cognitive load in users of a multi-role 3D platform. An experiment (71 participants: 34 designers, 37 customers; 639 observations) using objective metrics and subjective scales revealed different load profiles: designers are slowed down by data entry forms, while customers are hindered by 3D content. Based on 34 problem areas, seven interface design principles and a task classification by cognitive complexity level, accounting for role-specific factors, were formulated. Retesting (21 participants) confirmed their effectiveness: task completion time decreased by 21–28%, errors by 33–44%, frustration by 25–35%, and the use of 3D features increased from 40.5% to 90.9%. The developed three-component model (taxonomic, diagnostic, and design levels) is recommended for designing similar multi-role platforms with 3D functionality.

Abstract:

In recent years, large language models (LLMs) have achieved substantial progress in natural language processing tasks and have become key instruments for addressing a wide range of applied and research problems. However, as their scale and capabilities grow, the issue of hallucinations — i.e., the generation of false, unreliable, or nonexistent information presented in a credible manner—has become increasingly acute. Consequently, analyzing the nature of hallucinations and developing methods for their detection has acquired both scientific and practical significance.

This study examines the phenomenon of hallucinations in large language models, reviews their existing classification, and investigates potential causes. Using the Flan-T5 model, we analyze differences in the model’s internal states when generating hallucinations versus correct responses. Based on these discrepancies, we propose two approaches for hallucination detection: one leveraging attention maps and the other utilizing the model’s hidden states. These methods are evaluated on data from HaluEval and Shroom 2024 benchmarks in tasks such as summarization, question answering, paraphrasing, machine translation, and definition generation. Additionally, we assess the transferability of the trained detectors across different hallucination types, in order to evaluate the robustness of the proposed methods.

Abstract:

This paper presents the development and evaluation of a toolkit designed to optimize the workflow of game designers working with narrative content and game design for indie RPGs in the Godot engine. The relevance of this study stems from the growing number of independent studios utilizing Godot and similar game engines, which necessitates the creation of specialized tools to compensate for the lack of built-in solutions for RPG development, such as dialogue systems, turn-based combat design, cutscene management, and complex user interface implementation.

The aim of the work is to design a toolkit based on an architectural analysis of the reference game Persona 5 Royal, ensuring an optimized workflow for the game designer and reducing their dependence on the programmer. Methodologically, the research relied on identifying player states, independent modules, and areas of professional responsibility, which led to the development of five tools. Their effectiveness was evaluated with the participation of 12 novice game designers, including task completion time measurements, interviews, and expert ranking.

The results of the study showed that tools without analogues in Godot ("Combat Segments," "Complex User Interface," "Dialogue System") received high expert evaluations. The "Interactive Location Objects" tool, which has an analogue in Godot, reduced task completion time by an average of 65.5%. The "Commands to the Game World" tool did not demonstrate effectiveness and requires further improvement.

The practical significance of the research results lies in the creation of a ready-to-implement toolkit that enables indie teams to reduce RPG development time and increase the autonomy of their game designers.

Abstract:

This study investigates the ability of large language models to detect greeting and goodbye sequences in Russian-language dialogues and determine their precise pragmatic boundaries. YandexGPT 5.1 Pro, GigaChat 2 Max, and GPT-5 are evaluated on two tasks: classifying dialogues based on the presence of these sequences and extracting individual utterances, taking their intentional boundaries into account. The experiment is conducted on 216 dialogues from the Russian Multimedia Politeness Corpus, thereby accounting for the broad communicative and social context of interaction. GPT-5 achieves the strongest dialogue-level classification performance, whereas YandexGPT and GigaChat attain near-perfect precision through conservative prediction strategies. Performance declines markedly for all models when exact boundaries must be extracted. Error analysis shows that the models systematically over-include contextual nonverbal actions and miss non-conventional markers, including gestures, address forms, discourse markers, and implicatures indicating future contact. The article presents an updated description of the corpus and its web interface, as well as facilities for constructing diagnostic subcorpora and benchmarks. The findings underscore the need to model discourse structure and the contextual variation of everyday communication more precisely.

Abstract:

Despite recent progress, Optical Character Recognition (OCR) on historical newspapers still leaves 5–10% character errors. We present a fully automated post-OCR normalization pipeline that combines lightweight 7–8B instruction-tuned LLMs quantized to 4-bit (INT4) with a small set of regex rules. On the BLN600 benchmark (600 pages of 19th-century British newspapers), our best model YandexGPT-5-Instruct Q4 reduces Character Error Rate (CER) from 8.4% to 4.0% (–52.5%) and Word Error Rate (WER) from 20.2% to 6.5% (–67.8%), while raising semantic similarity to 0.962. The system runs on consumer hardware (RTX-4060 Ti, 8 GB VRAM) at about 35 seconds per page and requires no fine-tuning or parallel training data. These results indicate that compact INT4 LLMs are a practical alternative to large checkpoints for post-OCR cleanup of historical documents.

Abstract:

Large language models (LLMs) are increasingly being used for code generation. Yet, both the generated code and the underlying LLM-based systems demand rigorous security evaluation. A common approach to enhancing code generation quality involves multi-agent systems composed of multiple models. This paper evaluates the performance of GPT-OSS 20B, GPT-OSS 120B, and Qwen3-Coder 480B models in single-agent and multi-agent configurations, using two code security benchmarks, SecurityEval and CyberSecEval. The key contribution is SafeAICoder, an extensible and scalable framework for testing LLMs enabling distributed server-side generation of multi-module programs and tests, independent of client-side code.

Abstract:

The complexity of the problem of extracting argumentative structures is associated with such problems as selecting argumentative segments, predicting long-range connections between non-contact segments, and training on data labeled with a low degree of inter-annotator consistency. In this paper, we consider an approach to extracting argumentative relations from fairly large texts related to scientific communication. A comparative analysis was performed of fine-tuning methods using a pre-trained Longformer-type language model that takes into account long contexts and two methods that take into account annotator discrepancies in argument labeling by using the so-called soft labels obtained by uniformly smoothing labels and averaging expert assessments. The experiments were conducted on four datasets containing positive and negative examples of statement pairs (premise, conclusion) and differing in segmentation methods and average text size. The best results were obtained using the model with averaging expert assessments. At the same time, it is noted that the model using smoothed labels also increases the accuracy of classifiers, but worsens the recall.

Abstract:

This paper examines the challenges of assessing the reliability of information extracted from popular science texts. Popular science is known to be characterized by a free style of presentation, loose references to primary sources and expert opinions, and the presence of controversial statements and conclusions. Clarifying these components of the author's reasoning can significantly increase the credibility among critical readers or, conversely, refute the authors' assertions.

This paper proposes a new approach to assessing the reliability of information contained in popular science publications. This approach utilizes D. Walton's argumentation theory to identify potentially unreliable claims and logical fallacies. Searching scientific databases and calculating trust metrics for scientific publications based on their metadata and scientific ratings will enable the verification of controversial claims and an assessment of their reliability. The use of large language models will enable the generation of explanations for decisions made.

To support the development and validation of this approach, two data collection methods are proposed. The first uses existing corpora of Russian-language popular science texts and commentaries with annotated argumentation. The annotation includes, among other things, polemical argumentation schemes that challenge the author's thesis or argument. The second method utilizes LLM for generating synthetic theses and popular science articles, based on actual scientific publications.