Formation of the qualified, international recognized IT specialists in Kazakhstan became the purpose of creation of a higher educational institution of a similar profile. International IT University provided with grants from the government of Kazakhstan and national infocommunication companies, which cover disciplines by Kazakhstan and the U.S. educational systems.
Automatic processing of morphologically rich, agglutinative, and low-resource languages remains challenging because productive affixation increases lexical sparsity, weakens statistical generalization, and often produces inconsistent predictions across related linguistic annotation tasks. This study presents QazNLP, a constraint-aware multi-task framework for Kazakh that jointly performs morphological tagging, part-of-speech tagging, and named entity recognition using a shared transformer encoder with task-specific prediction heads. To improve structural reliability, the framework introduces differentiable cross-task compatibility penalties that discourage linguistically invalid label combinations during training and constrained decoding. The study further provides a reproducible evaluation setting based on a cleaned Kazakh news corpus with fixed data splits and robustness diagnostics for out-of-vocabulary tokens, long agglutinative word forms, reduced-data regimes, and cross-task contradiction analysis. In addition, the manuscript explicitly documents the encoder and tokenization setup, justifies the choice of pretrained encoder, reports decoding complexity, and situates the NER component with respect to the KazNERD benchmark. Experimental results show that the proposed model consistently outperforms competitive single-task and shared multi-task baselines in joint average F1, robustness under sparse-data conditions, and structural consistency of predicted labels. The findings indicate that explicit compatibility-aware optimization offers a practical and extensible direction for sequence labeling in morphologically rich low-resource languages.
This work investigates the deployment and comparative performance of the YOLOv8s and YOLOv11s object detection models on the Radxa 4D platform, utilizing the Rockchip RK3576 system-on-chip with a dedicated neural processing unit (NPU). The research focuses on the end-to-end adaptation and evaluation of modern convolutional neural networks for real-time inference under strict computational constraints typical for edge devices. The main challenge addressed is achieving real-time detection while maintaining accuracy after INT8 quantization and hardware-specific compilation. The proposed solution employs a conversion pipeline from PyTorch to ONNX and RKNN formats, followed by asymmetric affine INT8 quantization and NPU execution. Experimental evaluation was conducted on a combined 5,981-image aerial dataset specifically targeting human detection amid small objects and complex backgrounds. The results demonstrate significant performance improvements for both architectures: inference speed increased from 1.1-1.4 FPS on CPU to 25.5-26.4 FPS on NPU, while average latency decreased from ${7 4 0}-{9 0 9} \text{ms}$ to ${3 7}-{3 9} \text{ms}$ per frame. Detection accuracy remained stable; notably, INT8 quantization acted as an implicit regularization mechanism, reducing False Positives across both models, although a moderate decrease in strict spatial metrics (mAP@0.5:0.95) was observed. A distinctive feature of this work is the empirical comparison of different YOLO generations, confirming reliable detection under challenging backgrounds. The findings validate the practical application of this approach in edge AI systems, including UAV-based monitoring and autonomous robotics.
The integration of technology has transformed education, introducing tools that make learning more interactive and engaging. In language learning, digital applications like Quizlet have become popular for addressing diverse learner needs. Quizlet provides flashcards, quizzes, and games, enabling students to engage actively with vocabulary while maintaining motivation. Vocabulary proficiency is essential for effective communication and is linked to reading comprehension, academic performance, and overall language skills. Traditional methods, such as rote memorization, often fail to engage learners. This study investigates the effectiveness of Quizlet in enhancing vocabulary proficiency among students at the International Information Technology University. Using a quantitative design, four groups of 15 students each participated to assess the tool's impact. The results are expected to contribute to research on educational technology and its role in improving language learning outcomes.
This research is dedicated to developing an integrated mathematical model for the comprehensive assessment of Microservice Architecture (MSA) dependability, accounting for both technical failures and cyber threats. 1 The traditional separate analysis of fault tolerance and security is insufficient for adequately evaluating the overall resilience of distributed systems. The methodology employs the apparatus of stochastic modeling (Markov chains), combining reliability metrics (failure rate, restoration rate) and security parameters (attack probability, defense effectiveness). The model calculates the steady-state probability of a single service being operational and the overall system being operational considering architectural complexity. Numerical modeling demonstrated that systemic resilience is exponentially sensitive to security factors and architectural sprawl. Investment in increasing defense effectiveness is identified as a critical multiplicative factor for ensuring system availability in scalable MSA. The proposed model is a practical tool for Site Reliability Engineering (SRE) and cyber resilience assurance.
This article addresses the critical challenge of decision support for ensuring the reliability and ergonomics of cyber-physical systems. The study analyzes the evolving role of the human operator amidst the transition to “Industry 5.0” and “Operator 5.0” paradigms, providing a substantiated methodological framework for the reliability-driven design of cyber-physical-human systems. To evaluate human-machine interaction, the research utilizes the functional network apparatus integrated with the principle of algorithmic recognition for typical functional configurations. The study results in a logic programming-based information technology that enables real-time assessment of human-computer interaction scenarios, facilitating enhanced decision-making in complex industrial environments.
