Search Results (81)

This paper proposes an intelligent power adaptation framework for Free-Space Optics (FSO) communication systems operating under different weather conditions exploiting a deep learning (DL) analysis of received eye diagram images. The system incorporates two Convolutional Neural Network (CNN) architectures, LeNet and Wide Residual Network (Wide ResNet) algorithms to perform regression tasks that predict received signal quality metrics such as the Quality Factor (Q-factor) and Bit Error Rate (BER) from the received eye diagram. These models are evaluated using Mean Squared Error (MSE) and the coefficient of determination (${\mathrm{R}}^2$ score) to assess prediction accuracy. Additionally, a custom CNN-based classifier is trained to determine whether the BER reading from the eye diagram exceeds a critical threshold of ${10}^{-4}$; this classifier achieves an overall accuracy of 99%, correctly detecting 194/195 “acceptable” and 4/5 “unacceptable” instances. Based on the predicted signal quality, the framework activates a dual-amplifier configuration comprising a pre-channel amplifier with a maximum gain of 25 dB and a post-channel amplifier with a maximum gain of 10 dB. The total gain of the amplifiers is adjusted to support the operation of the FSO system under all-weather conditions. The FSO system uses a 15 dBm laser source at 1550 nm. The DL models are tested on both internal and external datasets to validate their generalization capability. The results show that the regression models achieve strong predictive performance, and the classifier reliably detects degraded signal conditions, enabling the real-time gain control of the amplifiers to achieve the quality of transmission. The proposed solution supports robust FSO communication under challenging atmospheric conditions including dry snow, making it suitable for deployment in regions like Northern Europe, Canada, and Northern Japan. Full article

We propose a high-speed all-optical logic gate design based on silicon racetrack and ring resonators patterned on a silica substrate. The architecture features racetrack resonators at both the input and output, with a central ring resonator enabling the required phase-sensitive interference for logic processing. Logic operations are achieved through the interplay of constructive and destructive interference induced by phase-shifted input beams. Using the finite-difference time-domain (FDTD) method in Lumerical software, we simulate and demonstrate seven fundamental Boolean logic functions, namely XOR, AND, OR, NOT, NOR, NAND, and XNOR, at an operating wavelength of 1.33 µm. The system supports a data rate of 47.94 Gb/s, suitable for ultrafast optical computing. The performance is quantitatively evaluated using the contrast ratio (CR) as the reference metric, with more than acceptable values of 13.09 dB (XOR), 13.84 dB (AND), 13.14 dB (OR), 13.80 dB (NOT), 14.53 dB (NOR), 13.80 dB (NAND), and 14.67 dB (XNOR), confirming strong logic level discrimination. Comparative analysis with existing optical gate designs underscores the advantages of our compact silicon-on-silica structure in terms of speed, CR performance, and integration potential. This study validates the effectiveness of racetrack–ring configurations for next-generation all-optical logic circuits. Full article

Alzheimer’s disease (AD) is a progressive, non-curable neurodegenerative disorder that poses persistent challenges for early diagnosis due to its gradual onset and the difficulty in distinguishing pathological changes from normal aging. Neuroimaging, particularly MRI and PET, plays a key role in detection; however, limitations in data availability and the complexity of early structural biomarkers constrain traditional diagnostic approaches. This review investigates the use of generative models, specifically Generative Adversarial Networks (GANs) and Diffusion Models, as emerging tools to address these challenges. These models are capable of generating high-fidelity synthetic brain images, augmenting datasets, and enhancing machine learning performance in classification tasks. The review synthesizes findings across multiple studies, revealing that GAN-based models achieved diagnostic accuracies up to 99.70%, with image quality metrics such as SSIM reaching 0.943 and PSNR up to 33.35 dB. Diffusion Models, though relatively new, demonstrated strong performance with up to 92.3% accuracy and FID scores as low as 11.43. Integrating generative models with convolutional neural networks (CNNs) and multimodal inputs further improved diagnostic reliability. Despite these advancements, challenges remain, including high computational demands, limited interpretability, and ethical concerns regarding synthetic data. This review offers a comprehensive perspective to inform future AI-driven research in early AD detection. Full article

Driven by climate change and stringent ecological conservation policies, arid and semi-arid pastoral areas face acute grassland degradation and forage–livestock imbalances. In Sunan County (Gansu Province, China), herders have increasingly turned to off-site grazing—leasing crop fields in adjacent oases during autumn and winter—to alleviate local grassland pressure and adapt their livelihoods. However, the interplay between the evolving land use system (L) and this emergent borrowed pasture system (B) remains under-explored. This study introduces a coupled analytical framework linking L and B. We employ multi-temporal remote sensing imagery (2018–2023) and official statistical data to derive land use dynamic degree (LUDD) metrics and 14 indicators for the borrowed pasture system. Through entropy weighting and a coupling coordination degree model (CCDM), we quantify subsystem performance, interaction intensity, and coordination over time. The results show that 2017 was a turning point in grassland–bare land dynamics: grassland trends shifted from positive to negative, whereas bare land trends turned from negative to positive; strong coupling but low early coordination (C > 0.95; D < 0.54) were present due to institutional lags, infrastructural gaps, and rising rental costs; resilient grassroots networks bolstered coordination during COVID-19 (D ≈ 0.78 in 2023); and institutional voids limited scalability, highlighting the need for integrated subsidy, insurance, and management frameworks. In addition, among those interviewed, 75% (15/20) observed significant grassland degradation before adopting off-site grazing, and 40% (8/20) perceived improvements afterward, indicating its potential role in ecological regulation under climate stress. By fusing remote sensing quantification with local stakeholder insights, this study advances social–ecological coupling theory and offers actionable guidance for optimizing cross-regional forage allocation and adaptive governance in arid pastoral zones. Full article

Many academic introductions suffer from inconsistencies and a lack of comprehensive structure, often failing to effectively outline the core elements of the research. This not only impacts the clarity and readability of the article but also hinders the communication of its significance and objectives to the intended audience. This study aims to automate the CaRS (Creating a Research Space) model using machine learning and natural language processing techniques. We conducted a series of experiments using a custom-developed corpus of 50 biology research article introductions, annotated with rhetorical moves and steps. The dataset was used to evaluate the performance of four classification algorithms: Prototypical Network (PN), Support Vector Machines (SVM), Naïve Bayes (NB), and Random Forest (RF); in combination with six embedding models: Word2Vec, GloVe, BERT, GPT-2, Llama-3.2-3B, and TEv3-small. Multiple experiments were carried out to assess performance at both the move and step levels using 5-fold cross-validation. Evaluation metrics included accuracy and weighted F1-score, with comprehensive results provided. Results show that the SVM classifier, when paired with Llama-3.2-3B embeddings, consistently achieved the highest performance across multiple tasks when trained on preprocessed dataset, with 79% accuracy and weighted F1-score on rhetorical moves and strong results on M2 steps (75% accuracy and weighted F1-score). While other combinations showed promise, particularly NB and RF with newer embeddings, none matched the consistency of the SVM–Llama pairing. Compared to existing benchmarks, our model achieves similar or better performance; however, direct comparison is limited due to differences in datasets and experimental setups. Despite the unavailability of the benchmark dataset, our findings indicate that SVM is an effective choice for rhetorical classification, even in few-shot learning scenarios. Full article

(1) Background: The highly pathogenic avian influenza virus H5N1 clade 2.3.4.4b is an emerging threat that poses a great risk to the poultry industry. A few human cases have been linked to the infection with this clade in many parts of the world, including the USA. Unfortunately, there are no specific vaccines or antiviral drugs that could help prevent and treat the infection caused by this virus in birds. Our major objective is to identify/repurpose some (novel/known) antiviral compounds that may inhibit viral replication by targeting some key viral proteins. (2) Methods: We used state-of-the-art machine learning tools such as molecular docking and MD-simulation methods from Biovia Discovery Studio (v24.1.0.321712). The key target proteins such as hemagglutinin (HA), neuraminidase (NA), Matrix-2 protein (M2), and the cap-binding domain of PB2 (PB2/CBD) homology models were validated through structural assessment via DOPE scores, Ramachandran plots, and Verify-3D metrics, ensuring reliable structural representations, confirming their reliability for subsequent in silico approaches. These approaches include molecular docking followed by molecular dynamics simulation for 50 nanoseconds (ns), highlighting the structural stability and compactness of the docked complexes. (3) Results: Molecular docking revealed strong binding affinities for both sofosbuvir and GS441524, particularly with the NA and PB2/CBD protein targets. Among them, GS441524 exhibited superior interaction scores and a greater number of hydrogen bonds with key functional residues of NA and PB2/CBD. The MM-GBSA binding free energy calculations further supported these findings, as GS441524 displayed more favorable binding energies compared to several known standard inhibitors, including F0045S for HA, Zanamivir for NA, Rimantadine and Amantadine for M2, and PB2-39 for PB2/CBD. Additionally, 50 ns molecular dynamics simulations highlighted the structural stability and compactness of the GS441524-PB2/CBD complex, further supporting its potential as a promising antiviral candidate. Furthermore, hydrogen bond monitor analysis over the 50 ns simulation confirmed persistent and specific interactions between the ligand and proteins, suggesting that GS441524 may effectively inhibit the NA, and PB2/CBD might potentially disrupt PB2-mediated RNA synthesis. (4) Conclusions: Our findings are consistent with previous evidence supporting the antiviral activity of certain nucleoside analog inhibitors, including GS441524, against various coronaviruses. These results further support the potential repurposing of GS441524 as a promising therapeutic candidate against H5N1 avian influenza clade 2.3.4.4b. However, further functional studies are required to validate these in silico predictions and support the inhibitory action of GS441524 against the targeted proteins of H5N1, specifically clade 2.3.4.4b. Full article

Deforestation and forest fires are escalating global threats that require timely, scalable, and cost-effective monitoring systems. While UAV and ground-based solutions offer fine-grained data, they are often constrained by limited spatial coverage, high operational costs, and logistical challenges. In contrast, satellite imagery provides broad, repeatable, and economically feasible coverage. This study presents a deep learning framework that combines the DeepLabV3+ architecture with an EfficientNet-B08 backbone to address both deforestation and wildfire detection using satellite imagery. The system utilizes advanced multi-scale feature extraction and Group Normalization to enable robust semantic segmentation under challenging atmospheric conditions and complex forest structures. It is evaluated on two benchmark datasets. In the Amazon forest segmentation dataset, the model achieves a validation Intersection over Union (IoU) of 0.9100 and a pixel accuracy of 0.9605, demonstrating strong performance in delineating forest boundaries. In FireDataset_20m, which presents a severe class imbalance between fire and non-fire pixels, the framework achieves 99.95% accuracy, 93.16% precision, and 91.47% recall. A qualitative analysis confirms the model’s ability to accurately localize fire hotspots and deforested areas. These results highlight the model’s dual-purpose utility for high-resolution, multi-temporal environmental monitoring. Its balanced performance across metrics and adaptability to complex terrain conditions make it a promising tool for supporting forest conservation, early fire detection, and evidence-based policy interventions. Full article

Background/Objectives: Oral cancer, particularly oral squamous cell carcinoma (OSCC), ranks as the sixth most prevalent cancer globally, with rates of occurrence on the rise. The diagnosis of OSCC primarily depends on histopathological images (HIs), but this method can be time-intensive, expensive, and reliant on specialized expertise. Manual diagnosis often leads to inaccuracies and inconsistencies, highlighting the urgent need for automated and dependable diagnostic solutions to enhance early detection and treatment success. Methods: This research introduces a deep learning (DL) approach utilizing EfficientNet-B3, complemented by learning rate tuning (LRT), to identify OSCC from histopathological images. The model is designed to automatically modify the learning rate based on the accuracy and loss during training, which improves its overall performance. Results: When evaluated using the oral tumor dataset from the multi-cancer dataset, the model demonstrated impressive results, achieving an accuracy of 99.84% and a specificity of 99.92%, along with other strong performance metrics. Conclusions: These findings indicate its potential to simplify the diagnostic process, lower costs, and enhance patient outcomes in clinical settings. Full article

Background/Objectives: Deep learning (DL) has shown strong potential in analyzing food images, but few studies have directly predicted mass, energy, and macronutrient content from images. In addition to the importance of high-quality data, differences in country-specific food composition databases (FCDBs) can hinder model generalization. Methods: We assessed the performance of several standard DL models using four ground truth datasets derived from Nutrition5k—the largest image–nutrition dataset with ~5000 complex US cafeteria dishes. In light of developing an Italian dietary assessment tool, these datasets varied by FCDB alignment (Italian vs. US) and data curation (ingredient–mass correction and frame filtering on the test set). We evaluated combinations of four feature extractors [ResNet-50 (R50), ResNet-101 (R101), InceptionV3 (IncV3), and Vision Transformer-B-16 (ViT-B-16)] with two regression networks (2+1 and 2+2), using IncV3_2+2 as the benchmark. Descriptive statistics (percentages of agreement, unweighted Cohen’s kappa, and Bland–Altman plots) and standard regression metrics were used to compare predicted and ground truth nutritional composition. Dishes mispredicted by ≥7 algorithms were analyzed separately. Results: R50, R101, and ViT-B-16 consistently outperformed the benchmark across all datasets. Specifically, when replacing it with these top algorithms, reductions in median Mean Absolute Percentage Errors were 6.2% for mass, 6.4% for energy, 12.3% for fat, and 33.1% and 40.2% for protein and carbohydrates. Ingredient–mass correction substantially improved prediction metrics (6–42% when considering the top algorithms), while frame filtering had a more limited effect (<3%). Performance was consistently poor across most models for complex salads, chicken-based or eggs-based dishes, and Western-inspired breakfasts. Conclusions: The R101 and ViT-B-16 architectures will be prioritized in future analyses, where ingredient–mass correction and automated frame filtering methods will be considered. Full article

The underwater acoustic (UWA) communication system faces challenges due to environmental factors, extensive multipath spread, and rapidly changing propagation conditions. Deep learning based solutions, especially for orthogonal frequency division multiplexing (OFDM) receivers, have been shown to improve performance. However, the UWA channel characteristics are highly dynamic and depend on the specific underwater conditions. Therefore, these models suffer from model mismatch when deployed in environments different from those used for training, leading to performance degradation and requiring costly, time-consuming retraining. To address these issues, we propose a transfer learning (TL)-based pre-trained model for OFDM based UWA communication. Rather than training separate models for each underwater channel, we aggregate received signals from five distinct WATERMARK channels, across varying signal to noise ratios (SNRs), into a unified dataset. This diverse training set enables the model to generalize across various underwater conditions, ensuring robust performance without extensive retraining. We evaluate the pre-trained model using real-world data from Qingdao Lake in Hangzhou, China, which serves as the target environment. Our experiments show that the model adapts well to these challenging environment, overcoming model mismatch and minimizing computational costs. The proposed TL-based OFDM receiver outperforms traditional methods in terms of bit error rate (BER) and other evaluation metrics. It demonstrates strong adaptability to varying channel conditions. This includes scenarios where training and testing occur on the same channel, under channel mismatch, and with or without fine-tuning on target data. At 10 dB SNR, it achieves an approximately 80% improvement in BER compared to other methods. Full article

The increasing need for sustainable energy management in smart buildings calls for cost-effective solutions that balance energy efficiency and occupant comfort. This article presents a Large Language Model (LLM)-based recommendation system capable of generating proactive, context-aware suggestions from dynamic building conditions. The system was trained on a combination of real-world data and Sinergym simulations, capturing inputs such as weather conditions, forecasts, energy usage, electricity prices, and detailed zone parameters. Five models were fine-tuned and evaluated: GPT-2-Small, GPT-2-Medium, DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, and GPT-4. To enhance evaluation precision, a novel metric, the Zone-Aware Semantic Reward (ZASR), was developed, combining Sentence-BERT with zone-level scoring and complemented by F1-Score metrics. While GPT-4 demonstrated strong performance with minimal data, its high inference cost limits scalability. In contrast, open-access models like DeepSeek-R1-Distill-Qwen-1.5B, DeepSeek-R1-Distill-Qwen-7B, and GPT-2-Medium required larger datasets but matched or exceeded GPT-4’s performance at significantly lower cost. The system demonstrated adaptability across diverse building types, supported by heterogeneous datasets and parameter normalization. Importantly, the system was also deployed in a real-world multi-zone residential building in Thessaloniki, Greece. During a two-week operational period under near-identical weather and occupancy conditions, the model-assisted recommendations contributed to an estimated 10% reduction in electricity consumption, showcasing the practical potential of LLM-based recommendations in live building environments. Full article

This study presents a novel investigation into immersive teleoperation systems using collaborative, device-agnostic interfaces for advancing smart haptics in industrial assistive applications. The research focuses on evaluating the quality of experience (QoE) of users interacting with a teleoperation system comprising a local robotic arm, a robot gripper, and heterogeneous remote tracking and haptic feedback devices. By employing a modular device-agnostic framework, the system supports flexible configurations, including one-user-one-equipment (1U-1E), one-user-multiple-equipment (1U-ME), and multiple-users-multiple-equipment (MU-ME) scenarios. The experimental set-up involves participants manipulating predefined objects and placing them into designated baskets by following specified 3D trajectories. Performance is measured using objective QoE metrics, including temporal efficiency (time required to complete the task) and spatial accuracy (trajectory similarity to the predefined path). In addition, subjective QoE metrics are assessed through detailed surveys, capturing user perceptions of presence, engagement, control, sensory integration, and cognitive load. To ensure flexibility and scalability, the system integrates various haptic configurations, including (1) a Touch kinaesthetic device for precision tracking and grounded haptic feedback, (2) a DualSense tactile joystick as both a tracker and mobile haptic device, (3) a bHaptics DK2 vibrotactile glove with a camera tracker, and (4) a SenseGlove Nova force-feedback glove with VIVE trackers. The modular approach enables comparative analysis of how different device configurations influence user performance and experience. The results indicate that the objective QoE metrics varied significantly across device configurations, with the Touch and SenseGlove Nova set-ups providing the highest trajectory similarity and temporal efficiency. Subjective assessments revealed a strong correlation between presence and sensory integration, with users reporting higher engagement and control in scenarios utilizing force feedback mechanisms. Cognitive load varied across the set-ups, with more complex configurations (e.g., 1U-ME) requiring longer adaptation periods. This study contributes to the field by demonstrating the feasibility of a device-agnostic teleoperation framework for immersive industrial applications. It underscores the critical interplay between objective task performance and subjective user experience, providing actionable insights into the design of next-generation teleoperation systems. Full article

Large language models (LLMs) are trained on huge datasets, which allow them to answer questions from various domains. However, their expertise is confined to the data that they were trained on. In order to specialize LLMs in niche domains like healthcare, various training methods can be employed. Two of these commonly known approaches are retrieval-augmented Generation and model fine-tuning. Five models—Llama-3.1-8B, Gemma-2-9B, Mistral-7B-Instruct, Qwen2.5-7B, and Phi-3.5-Mini-Instruct—were fine-tuned on healthcare data. These models were trained using three distinct approaches: retrieval-augmented generation (RAG) alone, fine-tuning (FT) alone, and a combination of both (FT+RAG) on the MedQuAD dataset, which covers a wide range of medical topics including disease symptoms, treatments, medications, and more. Our findings revealed that RAG and FT+RAG consistently outperformed FT alone across most models, particularly LLAMA and PHI. LLAMA and PHI excelled across multiple metrics, with LLAMA showing superior overall performance and PHI demonstrating strong RAG/FT+RAG capabilities. QWEN lagged behind in most metrics, while GEMMA and MISTRAL showed mixed results. Full article

The growing demand for sustainable business practices has led to the development of corporate sustainability assessment tools, with environmental, social, and governance (ESG) indicators becoming central to non-financial performance evaluation. These metrics increasingly influence investment decisions and corporate strategies. However, questions remain about whether sustainability practices have a measurable impact on economic value creation and distribution. This study investigates the causal relationship between corporate sustainability measured by the ISE-B3 index and stakeholder-oriented economic performance, specifically focusing on Distributed Added Value (DAV) and its main components. The analysis uses financial data from Brazilian companies listed in the ISE-B3 portfolios for the years 2022, 2023, and 2024. To address potential endogeneity, this study employs a panel data econometric approach using Instrumental Variables with Two-Stage Least Squares (IV-2SLS) as the primary estimation strategy, complemented by fixed and random effects models for robustness checks. The results indicate no statistically significant causal relationship between the ISE-B3 index and DAV or its components. The coefficient of ISE-B3 on DAV is −0.0006 (p = 0.896) in the IV-2SLS estimation, with similar non-significant results for all components. The models exhibit strong temporal dependence, with lagged dependent variable coefficients ranging from 0.8295 to 1.3578, reflecting the persistence of financial dynamics. These findings suggest that, within the Brazilian context, participation in the ISE-B3 index does not directly influence how companies create or distribute financial value to stakeholders. This study contributes to the literature by providing robust econometric evidence on the economic effects of corporate sustainability, offering a stakeholder-oriented perspective beyond the traditional shareholder-centric view. Full article

The detection of weak moving targets in heterogeneous clutter backgrounds is a significant challenge in radar systems. In this paper, we propose a track-before-detect (TBD) method based on information geometry (IG) theory applied to range-azimuth measurements, which extends the IG detectors to multi-frame detection through inter-frame information integration. The approach capitalizes on the distinctive benefits of the information geometry detection framework in scenarios with strong clutter, while enhancing the integration of information across multiple frames within the TBD approach. Specifically, target and clutter trajectories in multi-frame range-azimuth measurements are modeled on the Hermitian positive definite (HPD) and power spectrum (PS) manifolds. A scoring function based on information geometry, which uses Kullback–Leibler (KL) divergence as a geometric metric, is then devised to assess these motion trajectories. Moreover, this study devises a solution framework employing dynamic programming (DP) with constraints on state transitions, culminating in an integrated merit function. This algorithm identifies target trajectories by maximizing the integrated merit function. Experimental validation using real-recorded sea clutter datasets showcases the effectiveness of the proposed algorithm, yielding a minimum 3 dB enhancement in signal-to-clutter ratio (SCR) compared to traditional approaches. Full article