Rashid Nasimov

Wearable surface electromyography (sEMG) biosensors enable non-invasive monitoring of hand and forearm activity and are promising for sign language recognition (SLR) in digital health applications. However, signals acquired with wearable sEMG devices are affected by noise and motion artifacts, and the preprocessing filter strongly shapes the extracted features, yet filtering methods are rarely compared under identical conditions. Here, eight preprocessing methods (Bessel, Butterworth, Chebyshev Type I and II, Elliptic, Equiripple FIR, Discrete Wavelet Transform (DWT), and Modified Wiener) were compared under controlled laboratory conditions using four-channel sEMG recorded with a wireless wearable acquisition system from 20 participants performing 10 sign classes. All methods shared the same segmentation, feature extraction, and classification pipeline with KNN, LDA, LR, RF, and SVM classifiers, evaluated under random split, cross-session, and subject-independent protocols. Butterworth filtering with KNN achieved the highest random split performance (Accuracy 96.66%, F1-score 96.12%) and the highest subject-independent accuracy (88.81%). Filter effectiveness depended on the classifier and validation scenario, and a higher SNR did not guarantee higher accuracy. LR and LDA required approximately 0.1 ms per sample for prediction in the evaluated computational environment. However, these measurements do not establish real-time performance on wearable hardware. The findings provide comparative evidence for selecting sEMG preprocessing methods under controlled offline conditions, while their applicability to real-time wearable systems requires further validation.

The rapid growth of corporate networks and the wide use of “Bring Your Own Device” (BYOD) policies have made these networks harder to secure and manage. One visible consequence is the spread of unauthorized, unmanaged switches and wireless routers. Network Management Systems (NMSs) differ in their polling, topology discovery, and multivendor capabilities. This paper presents an automated framework for multivendor network topology discovery and Shadow IT detection, together with a topological mapping architecture based on concurrent routines written in the Go programming language. The contribution lies in the integration and operational evaluation of established techniques on a heterogeneous campus network rather than in new individual algorithms. A hybrid asynchronous scanning algorithm combining the Internet Control Message Protocol (ICMP), Simple Network Management Protocol (SNMP), Link Layer Discovery Protocol (LLDP), and Address Resolution Protocol (ARP) is used to overlap discovery-stage response waits. We use graph-theoretic models and a heuristic “Health Score” function to classify network anomalies such as loopbacks, Spanning Tree Protocol (STP) conflicts, and unauthorized devices. Virtualization-prefix and cached ESXi-signature checks exclude candidate ports associated with virtualization; Time-To-Live (TTL) values separately estimate the operating system (OS) family. Experiments conducted on a live university campus network showed that the 100-worker asynchronous pool achieved a 94.9-fold discovery-stage speedup over sequential polling. In a replay of a recent forwarding database (FDB) snapshot, the virtualization checks reduced the number of candidate ports with multiple media access control (MAC) addresses from 96 to 72. Detection accuracy against independent labels remains to be evaluated. The framework was further evaluated through a fixed 24-h monitoring window and a functional comparison with established network management platforms. The results demonstrate faster discovery on the evaluated campus network; detection accuracy, resource overhead, and generalization require further validation.

Unmanned aerial vehicle imagery provides a practical and high-resolution data source for monitoring the spatial condition of urban and peri-urban green zones. However, the reliable detection of damaged, stressed, or declining vegetation areas remains challenging because UAV images often contain complex background patterns, illumination variations, shadows, mixed vegetation structures, and visually similar damage symptoms. Convolutional neural networks have shown strong potential for extracting hierarchical visual features from image data, yet their conventional fully connected classification heads may not always provide the most robust decision boundary, particularly when training data are limited, class distributions are imbalanced, or subtle vegetation damage patterns must be separated. This study presents a comparative architectural evaluation of CNN-based models in which the internal components of the network, including convolutional kernel size, pooling strategy, flattening mechanism, and classification head, are systematically modified and assessed. A custom CNN architecture is used as the feature extractor, while its conventional classification head is replaced with alternative machine learning classifiers, including XGBoost, Random Forest, Support Vector Machine, and Logistic Regression. In addition, the effects of different convolutional kernels, max and average pooling operations, flatten and global average pooling representations are experimentally investigated. The intermediate outputs of convolution, pooling, and feature extraction stages are visualized to interpret how architectural modifications influence the representation of vegetation texture, color degradation, canopy gaps, and damaged regions. The proposed experimental framework enables a detailed comparison between conventional CNN classifiers and hybrid CNN–machine learning models. The study evaluates each configuration using classification metrics, visual feature analysis, and model interpretability tools. The outcome is to identify an effective CNN-based architecture for UAV imagery that improves damaged green zone detection while providing a clearer understanding of how architectural choices affect model behavior.

Detection breaks down on objects a few pixels wide for a structural reason: as box side length shrinks, intersection over union (IoU) degenerates toward an indicator of exact pixel coincidence, so a one-pixel error can rewrite both the label assignment and the regression gradient. We avoid this regime by discarding pairwise box matching altogether. A scene is treated instead as an unnormalized positive measure over object centers, log-scales, and categories, and training minimizes a debiased, uncertainty-aware unbalanced optimal transport divergence between the predicted measure and a Gaussian-smoothed annotation measure. Soft correspondences replace hard assignment, and because the marginals are left free, total confidence mass survives as count supervision. A second training-only signal is drawn from the image itself predicted instances are rendered back to pixels through anisotropic Gaussian footprints, a shared blur kernel, pixel integration, and a capacity limited background, then compared with the observation under a Poisson–Gaussian noise model. Scene priors enter through a clipped log-space gate that caps how far context may suppress appearance evidence, a bound we audit by reporting recall on atypical placements alongside aggregate gains. The method reaches 31.7 AP on the AI-TOD-v2 validation split, including 23.4 AP on 2–8-pixel objects, and 36.8, 37.1, and 31.2 AP on SODA-D, SODA-A, and AI-TOD-R, with 53.6 AP on tiny persons in TinyPerson. Ablations isolate the contribution of each component.

In most railway divisions the results of scheduled inspections of automation and telemechanics field devices are still recorded on paper. Such records reach engineering management with a delay, are easy to lose, and are difficult to verify. This paper examines the data transmission core of a digital inspection complex in which passive RFID tags identify both the equipment and the personnel, and the inspection record is delivered to a cloud server over a hybrid cellular architecture combining a failure-prone GSM channel with a reliable CDMA channel. To quantify the stability of such a system, a continuous-time Markov chain model is constructed in which the link is represented as an M/M/2/K queue with one unreliable server: both channels carry traffic in parallel, and during a GSM outage the CDMA channel alone sustains service. Records already admitted are preserved across a channel switch; only records arriving at a full shared buffer are rejected. This residual overflow loss stays below 0.1% at routine load with a buffer of m ≥ 5 and reaches about 2.8% only under post-incident overload. The model parameters were measured on an operating ESP32-based scanner complex piloted at Hamza station on 46 point machines. Calculations for three load scenarios show that increasing the local buffer beyond m = 5 yields diminishing returns while the delay grows, and that resilience is governed primarily by the presence of the redundant channel and adequate buffering, with the primary-channel recovery rate a secondary factor.

Electromyography (EMG)-based sensing is an important tool for assessing neuromuscular activity and monitoring athlete development; its reliability depends on electrode placement, signal quality, and accurate identification of muscle activation periods. This study proposes an intelligent EMG sensing framework integrating preliminary electrode placement assessment, muscle activity detection, feature extraction, and regression-based progression prediction. A placement assessment indicated that positioning the electrode adjacent to the innervation zone produced the highest RMS under the tested conditions. A two-stage activity detection method based on clustering and probabilistic modeling achieved an average error of 1.5% and a temporal deviation of 19 ms. Nine time-domain EMG features extracted from the detected activity segments were used to characterize athlete progression and estimate the time required to reach a reference neuromuscular profile. Among the methods, Linear Regression provided the best fit to the data, obtaining R² = 0.987 and RMSE = 4.21 and suggesting a predominantly linear relationship between the EMG-derived features and training duration within the dataset. However, these results were obtained from only six longitudinal observation periods for a single representative athlete, with each period represented by a 90-dimensional EMG feature vector derived from the ten movement classes. Therefore, the results should be interpreted as preliminary, athlete-specific goodness-of-fit findings rather than evidence of generalizable predictive performance. Validation using larger longitudinal cohorts and independent datasets is required. The proposed framework is compatible with future IoT-enabled wearable and edge-computing architectures; however, hardware-level implementation was beyond the scope of this study.

This study examines how full-time undergraduate and graduate students at a leading technical university in Uzbekistan engage with generative AI tools and how this relates to cognitive regulation, motivation, self-regulation, and ethical awareness. Using a quantitative exploratory design, an online survey was administered to 381 participants during the first (autumn) semester of the 2025–2026 academic year. A preliminarily screened 30-item instrument was developed across seven scales AI Use, Digital Attention, Cognitive Load, Motivation, Self-Regulation, Neuro-AI Pedagogical Competence, and Ethics and Academic Autonomy with internal consistency ranging from α = 0.717 to α = 0.895; content and discriminant validity have not yet been confirmed on independent samples. Group comparisons were conducted using Mann–Whitney U with Cohen's d on a balanced subsample ( n = 302; 151 per group), constructed to avoid confounding group-size asymmetry with substantive group differences. Significant between-group differences were found on six of seven scales ( p < .001): Ethics and Academic Autonomy (d = 1.47), AI Use (d = 1.29), Self-Regulation (d = 1.14), Neuro-AI Pedagogical Competence (d = 1.14), and Digital Attention (d = 0.99) showed large effects; Motivation showed a medium effect (d = 0.70). Cognitive load did not differ significantly ( p = .296). Multiple regression ( n = 372) identified Ethics and Academic Autonomy ( β = 0.446), Self-Regulation ( β = 0.279), and Motivation ( β = 0.178) as the strongest predictors of Neuro-AI Pedagogical Competence (R ² = 0.799). The findings suggest that higher scores on Neuro-AI Pedagogical Competence were associated with stronger ethical awareness, self-regulation, and motivation, rather than with frequency of AI use. The study contributes empirical evidence from Central Asian higher education and supports the design of level-differentiated AI-integrated curricula.

Surface electromyography (sEMG) signals are important for assessing muscle activity, neuromuscular behavior, and movement stability. sEMG signals are widely used in athlete performance monitoring and human–machine interface applications. However, existing methods have limitations in classification, accuracy and generalization across users. In this study, a real-world dataset was generated from 30 professional wrestlers using an 8-channel system based on 10 physical movements and technical elements. Nine time-domain and energy features, mean absolute value (MAV), integrated EMG (IEMG), root mean square (RMS), simple square integral (SSI), fourth power (4POW), wavelength (WL), difference absolute standard deviation (DASDV), variance (VAR), and average amplitude change (AAC), were systematically evaluated separately and in combination. Five classifiers were compared: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), k-Nearest Neighbor (KNN), and Neural Networks (NNs). The models were evaluated for accuracy, sensitivity, specificity, positive predictive value, and F1-score. The generalization ability was analyzed through cross-subject (24/6) and cross-session validation protocols. The nine feature combinations achieved the highest classification accuracy of 97.8% with the RF algorithm. The proposed approach can serve as a practical basis for real-time muscle activity monitoring, movement classification, and rehabilitation systems.

In many organizations, business intelligence systems support analytical reporting and operational decision making. As data volumes grow and analytical tasks become more complex, architectures based on centralized processing pipelines increasingly face limitations related to scalability and timely response. These challenges motivate the development of alternative architectural approaches capable of operating efficiently in data-intensive environments. This study presents a modular multi-agent business intelligence framework that distributes analytical tasks across autonomous agents and applies lightweight reinforcement learning at the decision-making stage. The analytical workflow is decomposed into agents responsible for data collection, preprocessing, analytical modeling, and decision execution. Decision adaptation relies on localized policy updates driven by operational feedback, which avoids complex learning coordination and helps preserve system stability and interpretability. The proposed framework is evaluated using real-world transactional data from an electronic commerce setting. Experimental results show that the approach consistently outperforms centralized analytical pipelines and non-agent machine learning baselines in terms of processing efficiency, classification accuracy, and balanced classification performance. Threshold-independent evaluation further confirms stronger discriminative behavior across varying decision thresholds. In addition, stability analysis across repeated experimental runs indicates reduced performance variance and more predictable system behavior. These findings suggest that the proposed multi-agent business intelligence framework provides a practical and scalable alternative to centralized analytical architectures for data-intensive decision-support environments, while maintaining the robustness and transparency required in enterprise systems. The evaluation is limited to a single dataset and a classification task, and results should be interpreted within this scope. Experiments on the Online Retail dataset (UCI Machine Learning Repository) show an average accuracy of 0.89 ± 0.012 (baseline: 0.74 ± 0.029) and decision latency of 94 ± 9 ms (baseline: 137 ± 16 ms) across 10 independent runs, indicating stable behavior under repeated execution.