Computation, Volume 13, Issue 12 (December 2025) – 25 articles

The rising deployment of artificial intelligence in public services is constrained by computational costs and limited domain-specific data, particularly in multilingual contexts. This study proposes a generalizable Agentic AI pipeline for developing question–answer chatbot systems using small language models (SLMs), demonstrated through a case study on the Thai Student Loan Fund (TSLF). The pipeline integrates four stages: OCR-based document digitization using Typhoon2-3B, agentic question–answer dataset construction via a clean–check–plan–generate (CCPG) workflow, parameter-efficient fine-tuning with QLoRA on Typhoon2-1B and Typhoon2-3B models, and retrieval-augmented generation (RAG) for source-grounded responses. Evaluation using BERTScore and CondBERT confirmed high semantic consistency (F_BERT = 0.9807) and stylistic reliability (F_BERT = 0.9839) of the generated QA corpus. Fine-tuning improved the 1B model’s domain alignment (F_BERT: 0.8593 → 0.8641), while RAG integration further enhanced factual grounding (F_BERT = 0.8707) and citation transparency. Cross-validation with GPT-5 and Gemini 2.5 Pro demonstrated dataset transferability and reliability. The results establish that Agentic AI combined with SLMs offers a cost-effective, interpretable, and scalable framework for automating bilingual advisory services in resource-constrained government and educational institutions. Full article

Ship maneuverability models are typically defined by three degrees of freedom: surge, sway, and yaw. However, patrol vessels operating in riverine environments often exhibit significant roll motion during course changes, necessitating the inclusion of this dynamic. This study develops interpretable machine learning models capable of predicting vessel behavior in four degrees of freedom (4-DoF): surge, sway, yaw, and roll. A dataset of 125 h of simulated maneuvers was employed, including 29 h of out-of-distribution (OOD) conditions to test model generalization. Four models were implemented and compared over a 15-step prediction horizon: linear regression, third-order polynomial regression, a state-space model obtained via the N4SID algorithm, and an AutoRegressive model with eXogenous inputs (ARX). Results demonstrate that all models captured the essential vessel dynamics, with the state-space model achieving the best overall performance (e.g., NMSE = 0.0246 for surge velocity on test data and 0.0499 under OOD conditions). Variable-wise, surge and sway showed the lowest errors, roll rate remained stable, and yaw rate was the most sensitive to distribution shifts. Model-wise, the ARX model achieved the lowest NMSE for surge prediction (0.0149), while regression-based models provided interpretable yet less accurate alternatives. Multi-horizon evaluation (1-, 5-, 15-, and 30-step) under OOD conditions confirmed a consistent monotonic degradation across models. These findings validate the feasibility of using interpretable machine learning models for predictive control, autonomous navigation, and combat scenario simulation in riverine operations. Full article

Overlapping communities are key characteristics of the structure and function analysis of complex networks. Shared or overlapping nodes within overlapping communities can either form subcommunities or act as intersections between larger communities. Nodes at the intersections that do not form subcommunities can be identified as overlapping nodes or as part of an internal structure of nested communities. To identify overlapping nodes, we apply a threshold rule based on the number of nodes in the nested structure. As the threshold value increases, the number of selected overlapping nodes decreases. This approach allows us to analyse the roles of nodes considered overlapping according to selection criteria, for example, to reduce the effect of noise. We illustrate our method by using three small and two larger real-world network structures. In larger networks, minor disturbances can produce a multitude of slightly different solutions, but the core communities remain robust, allowing other variations to be treated as noise. While this study employs our own method for community detection, other approaches can also be applied. Exploring the properties of shared nodes in overlapping communities of complex networks is a novel area of research with diverse applications in social network analysis, cybersecurity, and other fields in network science. Full article

In this paper, a fast distributed algorithm is proposed for solving the winners and price determination problems in a uniform price auction in which each bidder bids for multiple units out of a lot of k identical items with a per-unit price. In a conventional setting, all bidders disclose their bidding information to an auctioneer and let the auctioneer allocate the items and determine the uniform price, i.e., the least winning price. In our setting, all bidders do not need to disclose their bidding information to the auctioneer. The bidders and the auctioneer collaboratively compute by the distributed algorithm to determine in a small number of steps the units allocated and the uniform price. The number of steps is independent of the number of bidders. At the end of the computing process, each bidder can only know the units allocated to him/her and the uniform price. The auctioneer can only know the units being allocated to the bidders and the uniform price. Therefore, neither the bidders nor the auctioneer are able to know the per-unit bidding prices of the bidders except the uniform price. Moreover, the auctioneer is not able to know the bidding units of the losing bidders. Bidders’ per-unit bidding prices are protected, and the bidding units of the losing bidders are protected. Bidding information privacy is preserved. Full article

Skeleton-based action recognition has achieved remarkable advances with graph convolutional networks (GCNs). However, most existing models process spatial and temporal information within a single coupled stream, which often obscures the distinct patterns of joint configuration and motion dynamics. This paper introduces the Dual-Path Cross-Attention Graph Convolutional Network (DPCA-GCN), an architecture that explicitly separates spatial and temporal modeling into two specialized pathways while maintaining rich bidirectional interaction between them. The spatial branch integrates graph convolution and spatial transformers to capture intra-frame joint relationships, whereas the temporal branch combines temporal convolution and temporal transformers to model inter-frame dependencies. A bidirectional cross-attention mechanism facilitates explicit information exchange between both paths, and an adaptive gating module balances their respective contributions according to the action context. Unlike traditional approaches that process spatial–temporal information sequentially, our dual-path design enables specialized processing while maintaining cross-modal coherence through memory-efficient chunked attention mechanisms. Extensive experiments on the NTU RGB+D 60 and NTU RGB+D 120 datasets demonstrate that DPCA-GCN achieves competitive joint-only accuracies of 88.72%/94.31% and 82.85%/83.65%, respectively, with exceptional top-5 scores of 96.97%/99.14% and 95.59%/95.96%, while maintaining significantly lower computational complexity compared to multi-modal approaches. Full article

The rapid expansion of medical data, characterized by its complex high-dimensional attributes, presents numerous promising opportunities and substantial challenges in healthcare analytics. Adopting effective feature selection techniques is essential to take advantage of the potential of such data. This research presents a modified algorithm called (mDA), which is the hybrid algorithm between the Evolutionary Population Dynamics and the Dragonfly Algorithm. This method combines Evolutionary Population Dynamics’s strength with the Dragonfly Algorithm’s flexible capabilities, offering a robust evolutionary machine learning approach specifically designed for medical data analysis. By integrating the dynamic population modeling of Evolutionary Population Dynamics with the adaptive search techniques of Dragonfly Algorithm, the proposed mDA significantly improves accuracy, reduces the number of features, and obtains the minimum average of the fitness scores. Comparative experiments conducted on seven diverse medical datasets against other established algorithms confirm the superior performance of the proposed mDA, establishing it as a valuable approach in examining complex medical data. Full article

To address the challenges of interdependent design parameters and reliance on empirical trial-and-error in ultrasonic cell levitation culture devices, this study proposes a top-down design framework integrating multi-physics modeling with complex network analysis. First, acoustic field simulations optimize transducer arrangement and define the cell manipulation field, establishing the Top-level Basic Structure (TBS). A skeleton model of the acoustofluidic coupled field is constructed based on the TBS. Core parameters are then determined by refining the TBS through multi-physics analysis. Second, a 24-node design change propagation network is constructed. Leveraging the TBS model coupled with multi-physics fields, a directed network model analyzes parameter interactions. The HITS algorithm is applied to prioritize the design sequence based on authority and hub scores, resolving parameter conflicts. Experimental validation demonstrates a device acoustic pressure of 1.3 × 10⁴ Pa, stable cell levitation within the focused acoustic field, and a 40% reduction in design cycle time compared to traditional methods. This framework systematically sequences parameters, effectively determines the design order, enhances design efficiency, and significantly reduces dependence on empirical trial-and-error. It provides a novel approach for developing high-throughput organoid culture equipment. Full article

Fraud in financial services—especially account opening fraud—poses major operational and reputational risks. Static rules struggle to adapt to evolving tactics, missing novel patterns and generating excessive false positives. Machine learning promises adaptive detection, but deployment faces severe class imbalance: in the NeurIPS 2022 BAF Base benchmark used here, fraud prevalence is 1.10%. Standard metrics (accuracy, f1_weighted) can look strong while doing little for the minority class. We compare Logistic Regression, SVM (RBF), Random Forest, LightGBM, and a GRU model on N = 1,000,000 accounts under a unified preprocessing pipeline. All models are trained to minimize their loss function, while configurations are selected on a stratified development set using validation-weighted F1-score f1_weighted. For the four classical models, class weighting in the loss (class_weight $\in \{\mathrm{None},‘\mathrm{balanced}’\}$) is treated as a hyperparameter and tuned. Similarly, the GRU is trained with a fixed class-weighted CrossEntropy loss that up-weights fraud cases. This ensures that both model families leverage weighted training objectives, while their final hyperparameters are consistently selected by the f1_weighted metric. Despite similar AUCs and aligned feature importance across families, the classical models converge to high-precision, low-recall solutions (1–6% fraud recall), whereas the GRU recovers 78% recall at 5% precision (AUC $=0.8800$). Under extreme imbalance, objective choice and operating point matter at least as much as architecture. Full article

Flow quality at the engine face, especially total pressure recovery and swirl, is central to the performance and stability of external compression supersonic inlets. Steady-state RANS-based numerical computations are performed to quantify bleed/swirl trade-offs in a single-ramp intake. The CFD simulations were performed first without a bleed system over M_∞ = 1.4–1.9 to locate the practical onset of a bleed requirement. The deterioration in pressure recovery and swirl beyond M_∞ ≈ 1.6, which is consistent with a pre-shock strength near the turbulent separation threshold, motivated the use of a bleed system. The comparisons with and without the bleed system were performed next at M_∞ = 1.6, 1.8, and 1.9 across the operation map parameterized by the flow ratio. The CFD simulations were performed using ANSYS Fluent, with a pressure-based coupled solver with a realizable k-ε turbulence model and enhanced wall treatment. The results provide engine-face distortion metrics using a standardized ring to sector swirl ratio alongside pressure recovery. The results show that bleed removes low-momentum near-wall fluid and stabilizes the terminal–shock interaction, raising pressure recovery and lowering peak swirl and swirl intensity across the map, while extending the stable operating range to a lower flow ratio at a fixed M_∞. The analysis delivers a design-oriented linkage between shock/boundary-layer interaction control and swirl: when bleed is applied at and above M_∞ = 1.6, the separation footprints shrink and the organized swirl sectors weaken, yielding improved operability with modest bleed fractions. Full article

Droplet-based microfluidics offers a promising approach for enhancing heat transfer in microchannels, which is critical for the thermal management of microsystems. This study presents a two-dimensional numerical investigation of flow and heat transfer characteristics of liquid–liquid two-phase droplet flow in a rectangular flow-focusing microchannel. The phase-field method was employed to capture the interface dynamics between the dispersed (water) and continuous (oil) phases. The effects of total velocity and droplet size on pressure drop and heat transfer performance are systematically analyzed. The results indicate that the heat transfer of two-phase droplet flow was significantly enhanced compared to single-phase oil flow, with its maximum heat transfer coefficient being approximately three times that of single-phase oil flow. The average heat transfer coefficient increases with total velocity and exhibits a non-monotonic dependence on droplet size. These findings provide valuable insights into the design and optimization of rectangular flow-focusing droplet-based microfluidic cooling systems. Full article

Brain–computer interfaces (BCIs) have emerged as an option that allows better communication between humans and some technological devices. This article presents a BCI based on the steady-state visual evoked potentials (SSVEP) paradigm and low-cost hardware to control a virtual prototype of a robotic hand. A LED-based device is proposed as a visual stimulator, and the Open BCI Ultracortex Biosensing Headset is used to acquire the electroencephalographic (EEG) signals for the BCI. The processing and classification of the obtained signals are described. Classifiers based on artificial neural networks (ANNs) and support vector machines (SVMs) are compared, demonstrating that the classifiers based on SVM have superior performance to those based on ANN. The classified EEG signals are used to implement different movements in a virtual prosthetic hand using a co-simulation approach, showing the feasibility of BCI being implemented in the control of robotic hands. Full article

The advent of sixth-generation (6G) communications envisions a paradigm of ubiquitous intelligence and seamless physical–digital fusion, demanding unprecedented performance from the optical transport infrastructure. Achieving terabit-per-second capacities, microsecond latency, and nanosecond synchronisation precision requires a convergent, flexible, open, and AI-native x-Haul architecture that integrates communication with distributed edge computing. This study conducts a systematic literature review of recent advances, challenges, and enabling optical technologies for intelligent and autonomous 6G networks. Using the PRISMA methodology, it analyses sources from IEEE, ACM, and major international conferences, complemented by standards from ITU-T, 3GPP, and O-RAN. The review examines key optical domains including Coherent PON (CPON), Spatial Division Multiplexing (SDM), Hollow-Core Fibre (HCF), Free-Space Optics (FSO), Photonic Integrated Circuits (PICs), and reconfigurable optical switching, together with intelligent management driven by SDN, NFV, and Artificial Intelligence/Machine Learning (AI/ML). The findings reveal that achieving 6G transport targets will require synergistic integration of multiple optical technologies, AI-based orchestration, and nanosecond-level synchronisation through Precision Time Protocol (PTP) over fibre. However, challenges persist regarding scalability, cost, energy efficiency, and global standardisation. Overcoming these barriers will demand strategic R&D investment, open and programmable architectures, early AI-native integration, and sustainability-oriented network design to make optical fibre a key enabler of the intelligent and autonomous 6G ecosystem. Full article

Non-covalent contacts play a significant role in binding between fragments in supramolecular assemblies. Understanding the non-covalent binding capabilities of closo-borate anions and their derivatives is a significant research challenge, due to their ability to interact with biomolecules. The present work was focused on the theoretical study of non-covalent complexes between glycine and closo-borate anions [B_nH_n−1X]^y− (X = H, NH₃, OH, SH, F; n = 10, 12; y = 1, 2). The main binding patterns between glycine and cluster systems were defined, and the effect of the exo-polyhedral substituent on the stability of non-covalent complexes was analysed. Complexes based on ammonium and hydroxy derivatives of closo-borate anions [B_nH_n−1X]^y− (X = NH₃, OH; n = 10, 12; y = 1, 2) were the most stable among all the derivatives considered. The findings of this work can be applied to the design of non-covalent complexes of closo-borate systems with biomolecules. Full article

Academic dropout is a persistent challenge in higher education, particularly in contexts with socio-economic disparities and diverse learning conditions. Traditional predictive models often fail to capture the complex, non-linear interactions underlying student trajectories due to their reliance on low-dimensional and linear representations. This study introduces a visual representation–based computational approach for a student dropout analysis, applied to a real institutional dataset from Colombia. The methodology transforms structured student records into enriched visual encodings that map variable magnitudes, correlations, and latent relationships into spatial and textural patterns. These image-based representations allow convolutional neural networks (CNNs) to exploit hierarchical feature extraction, uncovering hidden dependencies not accessible through conventional classifiers. Experimental results demonstrate that a Convolutional Neural Network (CNN) trained from scratch outperforms both baseline machine learning models and transfer learning architectures across all evaluation metrics. Beyond predictive accuracy, the approach enhances data expressiveness, interpretability, and generalization, offering a visual-analytical perspective for understanding dropout dynamics. The Colombian case study confirms the feasibility and potential of this strategy in real educational settings, supporting early identification of at-risk students and contributing to the development of robust, explainable models in educational data mining and learning analytics. Full article

Distributed cloud networks spanning multiple jurisdictions face significant challenges in anomaly detection due to privacy constraints, regulatory requirements, and communication limitations. This paper presents a mathematically rigorous framework for privacy-preserving federated learning on hierarchical graph neural networks, providing theoretical convergence guarantees and optimisation bounds for distributed anomaly detection. A novel layer-wise federated aggregation mechanism is introduced, featuring a proven convergence rate $O\left(1/\sqrt{T}\right)$. That preserves hierarchical structure during distributed training. The theoretical analysis establishes differential privacy guarantees of $\left(\epsilon =1.0, \delta ={10}^{-5}\right)$ through layer-specific noise calibration, achieving optimal privacy–utility tradeoffs. The proposed optimisation framework incorporates: (1) convergence-guaranteed layer-wise aggregation with bounded gradient norms, (2) privacy-preserving mechanisms with formal composition analysis under the Moments Accountant framework, (3) meta-learning-based personalisation with theoretical generalisation bounds, and (4) communication-efficient protocols with a proven 93% reduction in overhead. Rigorous evaluation on the FEDGEN testbed, spanning 2780 km across Nigeria and the Democratic Republic of Congo, demonstrates superior performance with hierarchical F1-scores exceeding 95% across all regions, while maintaining theoretical guarantees. The framework’s convergence analysis shows robustness under realistic constraints, including 67% client participation, 200 ms latency, and 20 Mbps bandwidth limitations. This work advances the theoretical foundations of federated graph learning while providing practical deployment guidelines for cross-jurisdictional cloud networks. Full article

Advancements in deep learning have revolutionized materials discovery by enabling predictive modeling of complex material properties. However, single-modal approaches often fail to capture the intricate interplay of compositional, structural, and morphological characteristics. This study introduces a novel multimodal deep learning framework for enhanced material property prediction, integrating textual (chemical compositions), tabular (structural descriptors), and image-based (2D crystal structure visualizations) modalities. Utilizing the Alexandriadatabase, we construct a comprehensive multimodal dataset of 10,000 materials with symmetry-resolved crystallographic data. Specialized neural architectures, such as FT-Transformer for tabular data, Hugging Face Electra-based model for text, and TIMM-based MetaFormer for images, generate modality-specific embeddings, fused through a hybrid strategy into a unified latent space. The framework predicts seven critical material properties, including electronic (band gap, density of states), thermodynamic (formation energy, energy above hull, total energy), magnetic (magnetic moment per volume), and volumetric (volume per atom) features, many governed by crystallographic symmetry. Experimental results demonstrated that multimodal fusion significantly outperforms unimodal baselines. Notably, the bimodal integration of image and text data showed significant gains, reducing the Mean Absolute Error for band gap by approximately 22.7% and for volume per atom by 22.4% compared to the average unimodal models. This combination also achieved a 28.4% reduction in Root Mean Squared Error for formation energy. The full trimodal model (tabular + images + text) yielded competitive, and in several cases the lowest, error metrics, particularly for band gap, magnetic moment per volume and density of states per atom, confirming the value of integrating all three modalities. This scalable, modular framework advances materials informatics, offering a powerful tool for data-driven materials discovery and design. Full article

Accessibility and observability are two properties of dynamic models that provide insights into the structural relationships between their input, output, and state variables. They are closely related to controllability and structural local identifiability, respectively. Observability and identifiability determine, respectively, the possibility of inferring the unmeasured state variables and parameters of a model from output measurements; accessibility and controllability describe the possibility of driving its state by changing its input. Analysing these structural properties in nonlinear models of ordinary differential equations can be challenging, particularly when dealing with large systems. Two main approaches are currently used for their study: one based on differential geometry, which uses symbolic computation, and another one based on sensitivity calculations that uses numerical integration. These approaches are implemented in two MATLAB (R2024b) software tools: the differential geometry approach in STRIKE-GOLDD, and the sensitivity-based method in StrucID. These toolboxes differ significantly in their features and capabilities. Until now, their performance had not been thoroughly compared. In this paper we present a comprehensive comparative study of them, elucidating their differences in applicability, computational efficiency, and robustness against computational issues. Our core finding is that StrucID has a substantially lower computational cost than STRIKE-GOLDD; however, it may occasionally yield inconsistent results due to numerical issues. Full article

This study examines the flow behavior of water-lubricated heavy oil transport utilizing the core annular flow (CAF) technique. The goal is to enhance efficiency and minimize risks in pipeline operations. The flow was numerically simulated in a horizontal pipe using a Large Eddy Simulation (LES) model within a commercial Computational Fluid Dynamics (CFD) framework. The Geo Reconstruct scheme is employed to accurately capture the oil–water interface, and both oil and water initialization methods were assessed against experimental data. Results show that the LES model accurately reproduces the main flow features observed experimentally, particularly for low-viscosity oil–water systems. This suggests that the model can be a reliable tool for predicting flow behaviour in similar fluid systems. Further validation with varying parameters could enhance its applicability across a broader range of conditions. In cases of heavy oil, the velocity profile remains nearly constant within the oil core, indicating rigid body-like motion surrounded by a turbulent water annulus. Turbulence intensity and oil volume fraction distributions were closely related, with higher turbulence in water and lower in oil. Although wall adhesion modelling limited fouling prediction, simulations confirmed that fouling can significantly increase pressure losses. This illustrates the value of considering both fluid dynamics and material interactions in such systems. Future studies could explore the impact of varying temperature and pressure conditions on fouling behaviour to further refine predictive models. Overall, the LES approach proved suitable for analysing turbulent CAF, offering insights for optimizing viscosity ratios, flow rates, and design parameters for safer and more efficient heavy oil transport. Full article

When dealing with mathematical or statistical models involving one or more explanatory (independent) variables, one often wants to determine the effects of such variables on a response (independent) variable. In the case of linear regression models, one such effects measure is the so-called standardized regression coefficients used in various statistical software packages and discussed in regression textbooks. However, since strong reservations have been expressed against the use of standardized regression coefficients because of various limitations, the objective or initial working hypothesis behind the present research was that some alternative measure without such limitations ought to be explored. Consequently, elasticity coefficients are formulated and proposed as both relative and absolute such measures. While the standardized regression coefficients lack any convenient interpretation, the proposed elasticity coefficients have the particularly desirable property of having a logical and intuitively appealing interpretation in terms of the relative change in the value of the response variable as a consequence of a relative change (of 1 percent) in the value of one or more of the explanatory variables. Those elasticity measures have the flexibility of being applicable to individual or to all explanatory variables and to individual or to all observations or data sets. A numerical example is used to illustrate the use of these new measures. Comparison between values of the standardized regression coefficients and those of the corresponding elasticity coefficients based on reported data from various sources is provided. Also, the form of the elasticity coefficients for a variety of different types of models is presented. Statistical inferences are also discussed. Full article

Purpose: While hybrid quantum–classical neural networks (HNNs) are a promising avenue for quantum advantage, the critical influence of the classical backbone architecture on their performance remains poorly understood. This study investigates the role of lightweight convolutional neural network architectures, focusing on LCNet, in determining the stability, generalization, and effectiveness of hybrid models augmented with quantum layers for medical applications. The objective is to clarify the architectural compatibility between quantum and classical components and provide guidelines for backbone selection in hybrid designs. Methods: We constructed HNNs by integrating a four-qubit quantum circuit (with trainable rotations) into scaled versions of LCNet (050, 075, 100, 150, 200). These models were rigorously evaluated on CIFAR-10 and MedMNIST using stratified 5-fold cross-validation, assessing accuracy, AUC, and robustness metrics. Performance was assessed with accuracy, macro- and micro-averaged area under the ROC curve (AUC), per-class accuracy, and out-of-fold (OoF) predictions to ensure unbiased generalization. In addition, training dynamics, confusion matrices, and performance stability across folds were analyzed to capture both predictive accuracy and robustness. Results: The experiments revealed a strong dependence of hybrid network performance on both backbone architecture and model scale. Across all tests, LCNet-based hybrids achieved the most consistent benefits, particularly at compact and medium configurations. From LCNet050 to LCNet100, hybrid models maintained high macro-AUC values exceeding 0.95 and delivered higher mean accuracies with lower variance across folds, confirming enhanced stability and generalization through quantum integration. On the DermaMNIST dataset, these hybrids achieved accuracy gains of up to seven percentage points and improved AUC by more than three points, demonstrating their robustness in imbalanced medical settings. However, as backbone complexity increased (LCNet150 and LCNet200), the classical architectures regained superiority, indicating that the advantages of quantum layers diminish with scale. The mostconsistent gains were observed at smaller and medium LCNet scales, where hybridization improved accuracy and stability across folds. This divergence indicates that hybrid networks do not necessarily follow the “bigger is better” paradigm of classical deep learning. Per-class analysis further showed that hybrids improved recognition in challenging categories, narrowing the gap between easy and difficult classes. Conclusions: The study demonstrates that the performance and stability of hybrid quantum–classical neural networks are fundamentally determined by the characteristics of their classical backbones. Across extensive experiments on CIFAR-10 and DermaMNIST, LCNet-based hybrids consistently outperformed or matched their classical counterparts at smaller and medium scales, achieving higher accuracy and AUC along with notably reduced variability across folds. These improvements highlight the role of quantum layers as implicit regularizers that enhance learning stability and generalization—particularly in data-limited or imbalanced medical settings. However, the observed benefits diminished with increasing backbone complexity, as larger classical models regained superiority in both accuracy and convergence reliability. This indicates that hybrid architectures do not follow the conventional “larger-is-better” paradigm of classical deep learning. Overall, the results establish that architectural compatibility and model scale are decisive factors for effective quantum–classical integration. Lightweight backbones such as LCNet offer a robust foundation for realizing the advantages of hybridization in practical, resource-constrained medical applications, paving the way for future studies on scalable, hardware-efficient, and clinically reliable hybrid neural networks. Full article

Effective communication between deaf–mute and visually impaired individuals remains a challenge in the fields of human–computer interaction and accessibility technology. Current solutions mostly rely on single-modal recognition, which often leads to issues such as semantic ambiguity and loss of emotional information. To address these challenges, this study proposes a lightweight multimodal fusion framework that combines gestures and micro-expressions, which are then processed through a recognition network and a speech synthesis module. The core innovations of this research are as follows: (1) a lightweight YOLOv5s improvement structure that integrates residual modules and efficient downsampling modules, which reduces the model complexity and computational overhead while maintaining high accuracy; (2) a multimodal fusion method based on an attention mechanism, which adaptively and efficiently integrates complementary information from gestures and micro-expressions, significantly improving the semantic richness and accuracy of joint recognition; (3) an end-to-end real-time system that outputs the visual recognition results through a high-quality text-to-speech module, completing the closed-loop from “visual signal” to “speech feedback”. We conducted evaluations on the publicly available hand gesture dataset HaGRID and a curated micro-expression image dataset. The results show that, for the joint gesture and micro-expression tasks, our proposed multimodal recognition system achieves a multimodal joint recognition accuracy of 95.3%, representing a 4.5% improvement over the baseline model. The system was evaluated in a locally deployed environment, achieving a real-time processing speed of 22 FPS, with a speech output latency below 0.8 s. The mean opinion score (MOS) reached 4.5, demonstrating the effectiveness of the proposed approach in breaking communication barriers between the hearing-impaired and visually impaired populations. Full article

Computational analysis using PSpice has become an indispensable tool for evaluating power electronics circuits, as it allows accurate simulation of transient effects, ripple, and component dynamics, enabling reliable assessment of complex topologies before physical implementation. In single-phase systems, electrolytic components are commonly used due to their high energy density, which helps mitigate low-frequency ripple caused by power oscillations between the DC and AC sides. However, these components have a limited lifespan, which compromises the system’s long-term reliability. This work proposes and evaluates the Stacked Switched Capacitor (SSC) topology as a power decoupling technique, implemented within a 200 W Cuk converter. The proposed SSC design enables a substantial reduction in required capacitance, replacing a conventional 600 μF capacitor with only three 36 μF capacitors, while maintaining voltage stability and output power performance. Simulation results show a high efficiency of 94% and a DC-link energy of 0.992 J, confirming the SSC’s ability to effectively mitigate voltage ripple at twice the grid angular frequency (2ω, rad/s) without compromising system durability. Comparative analysis with conventional decoupling strategies demonstrates that the SSC topology offers a compact, efficient, and reliable alternative, reducing the number of required passive components and switching devices. These results provide a strong basis for further exploration of SSC-based designs in space- and cost-constrained single-phase DC-AC applications. Full article

The FitzHugh–Nagumo (FHN) equation in one dimension is solved in this paper using an improved physics-informed neural network (PINN) approach. Examining test problems with known analytical solutions and the explicit finite difference method (EFDM) allowed for the demonstration of the PINN’s effectiveness. Our study presents an improved PINN formulation tailored to the FitzHugh–Nagumo reaction–diffusion system. The proposed framework is efficiently designed, validated, and systematically optimized, demonstrating that a careful balance among model complexity, collocation density, and training strategy enables high accuracy within limited computational time. Despite the very strong agreement that both methods provide, we have demonstrated that the PINN results exhibit a closer agreement with the analytical solutions for Test Problem 1, whereas the EFDM yielded more accurate results for Test Problem 2. This study is crucial for evaluating the PINN’s performance in solving the FHN equation and its application to nonlinear processes like pulse propagation in optical fibers, drug delivery, neural behavior, geophysical fluid dynamics, and long-wave propagation in oceans, highlighting the potential of PINNs for complex systems. Numerical models for this class of nonlinear partial differential equations (PDEs) may be developed by existing and future model creators of a wide range of various nonlinear physical processes in the physical and engineering sectors using the concepts of the solution methods employed in this study. Full article

Right-angle dihedral structures produce strong, highly correlated returns that dominate radar cross-section (RCS) and image signatures. Conventional absorbers or random coding metasurfaces often lose effectiveness across wide frequency bands and angles, and cannot adequately suppress the corner-induced hot spots. We propose a wideband super-dispersion encoded surface (SDES) conformally applied to dihedral facets. The approach co-designs input-admittance for absorption with a deterministic super-dispersion phase sequence to redistribute energy spectrally and angularly, thereby decorrelating the returns. We implement SDES on a thin composite panel and evaluate it on canonical dihedral and dihedral–cylindrical hybrid configurations. Unlike diffuse or random coding schemes, SDES enforces broadband, angle-stable dispersion with a deterministic sequence that specifically addresses corner singularity scattering. We also introduce perceptual-hashing as an imaging-domain metric to link RCS control with observable radar-image changes. From 12–18 GHz, SDES reduces the average monostatic RCS by 9.6 dB on a right-angle dihedral. In dihedral–cylindrical hybrids, SDES removes the corner hot spots and drives the radar-image similarity index down to 0.31, confirming substantial alteration of scattering signatures. Full article

This study proposes an enhanced deep learning framework for accurate detection of P-peaks in noisy photoplethysmographic (PPG) signals, utilizing a hybrid architecture that integrates wavelet-based analysis with neural network components. The P-peak detection task is formulated as a binary classification problem, where the model learns to identify the presence of a peak at each time step within fixed-length input windows. A temporal attention mechanism is incorporated to dynamically focus on the most informative regions of the signal, improving both localization and robustness. The proposed architecture combines Discrete Wavelet Transform (DWT) for multiscale signal decomposition, Convolutional Neural Networks (CNNs) for morphological feature extraction, and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies. A temporal attention layer is introduced after the recurrent layers to enhance focus on time steps with the highest predictive value. An evaluation was conducted on 30 model variants, exploring different combinations of input types, decomposition levels, and activation functions. The best-performing model—Type30, which includes DWT (3 levels), CNN, LSTM, and attention—achieves an accuracy of 0.918, precision of 0.932, recall of 0.957, and F1-score of 0.923. The findings demonstrate that attention-enhanced hybrid architectures are particularly effective in handling signal variability and noise, making them highly suitable for real-world applications in wearable PPG monitoring, digital twins for Heart Rate Variability (HRV), and intelligent health systems. Full article