Earticle

This study investigates the emergence of emotional dynamics and intrinsic motivation within multi-agent reinforcement learning (MARL) systems. Traditional MARL frameworks rely solely on extrinsic task rewards, which often limit exploration and adaptability. To address this, we propose an Affective-Motivated MARL (AM-MARL) framework where agents integrate curiosity-based intrinsic rewards and emotionmodulated affective feedback alongside extrinsic reinforcement. Agents operate in a continuous multiagent environment, learning through Q-learning, Actor- Critic, or Advantage Actor-Critic (A2C) methods depending on their action space. The intrinsic reward is defined as the state-prediction error between observed and expected future states, while the affective reward arises from temporal changes in emotional state and the social influence among peers. Experimental results show that incorporating intrinsic and affective rewards enhances exploration coverage, stabilizes emotional trajectories, and improves coordination efficiency compared to extrinsic-only baselines. These findings suggest that emotional feedback, when coupled with curiosity-driven intrinsic signals, fosters more humanlike adaptability, cooperative intelligence, and stable affect regulation in MARL environments.

This study investigates the effects of a stepwise AI literacy education program on hallucination understanding (HU), information verification (IV), and AI confidence (AIC) among 85 university students in agriculture-related fields. The program was delivered in two phases across a six-month interval: Phase 1 (foundational AI literacy) and Phase 2 (applied AI-assisted app development). Using SPSS 26.0, descriptive statistics, t-tests, correlations, and regression analyses were conducted. Results showed significant improvements in HU and IV, while a slight decrease in AIC during the applied stage reflected positive cognitive recalibration rather than diminished competence. HU and IV also exhibited strong positive associations, underscoring the value of scaffolded instructional design. These findings demonstrate that stepwise AI literacy education effectively enhances students’ metacognitive alignment and supports responsible engagement with generative AI.

Existing simulators for performance analysis of resource management techniques in edge computing have a limitation: they lack horizontal management features such as inter-server clustering and container registry placement. To address this issue, this paper proposes EdgeNet, a new simulator specialized for modeling of network overhead and server clustering algorithm in edge computing environments. EdgeNet provides a Python library that can be used to develop leader election algorithms for clustered server groups, facilitating research into horizontal resource management approaches that were previously difficult to study.

Phishing attacks increasingly exploit mobile platforms and target users communicating in lowresource or code-mixed languages, posing challenges to traditional centralized detection systems. This study proposes hybrid knowledge distillation and federated learning framework for real-time, on-device phishing detection. The approach integrates a fine-tuned XLM-RoBERTa "teacher" model with a compact MobileBERT "student" model, distilled to achieve near teacher-level performance while enabling efficient offline inference. The distilled MobileBERT model, converted to ONNX for platform portability, achieved a fivefold reduction in size while preserving 98% of the original model’s accuracy. We conducted zero-shot evaluations on Korean, Spanish, and Turkish datasets to guarantee crosslinguistic robustness, and consistently obtained good accuracy and recall rates. Moreover, privacy-preserving updates were made possible using a federated learning simulation, which permits decentralized model enhancements without data exchange. The suggested architecture offers a cost-effective, expandable, and privacy-conscious approach to phishing detection in scenarios with limited connection and linguistic diversity.

This paper evaluates SnapCheck.io, a smart workflow platform aimed at improving the efficiency of project management in remote work environments. Through the use of AI-driven task scheduling, smart time tracking, and real-time analytics, SnapCheck.io enhances resource utilization and lowers the waiting time for task dependency. A mixed-method study at various organizations showed an improvement in performance by 18%, which was thus corroborated by the increased rate of task completion and quicker workflow execution. The results reveal that the use of intelligent automation in SnapCheck.io not only helps in streamlining the project coordination process but also promotes the ongoing productivity and collaboration of distributed teams.

Image-based sign language recognition frequently exhibits high computational costs and is susceptible to background noise. In this study, we introduce an efficient skeleton-based framework that utilizes 3D landmarks and transformers. A pivotal component of this approach is the creation of similar landmarks method, an ensemble technique that extracts and averages landmarks from multiple augmented video views. This approach enhanced the system's resilience to noise and missing coordinates. The model was evaluated using quadruple crossvalidation on a 30-word dataset combining AI-Hub (studio) and KSL (wild) data. The results demonstrated that the technique applying ensembles produced superior results in comparison to the technique not applied.

This study introduces a visualization framework designed to make GPT-driven voice command systems in virtual reality (VR) more transparent and predictable. Conventional interfaces usually show only the final action result, leaving users uncertain about how their spoken instructions were understood or processed. Our approach instead visualizes each stage of interpretation—from speech recognition and semantic understanding to planning and execution—so that users can preview and verify the system’s reasoning in real time. The framework integrates GPT-based language interpretation with a Unity visualization engine and live feedback loops. A prototype implementation was tested through scenario-based experiments, showing that process visualization helps users understand system behavior, recover from errors more efficiently, and develop stronger trust compared with traditional result-focused interfaces.

Modern cloud applications often operate as unmodified third-party services or legacy code, where direct instrumentation for observability is infeasible. This paper investigates whether a minimal black-box observability pipeline can still provide actionable insights in such contexts. We compose a standard stack—Envoy proxy, OpenTelemetry Collector, Prometheus, Jaeger, and Grafana—to collect metrics and traces without modifying application code, and we apply it to two representative workloads: a lightweight demo service and the OWASP Juice Shop. Our evaluation shows that proxy-only instrumentation captures meaningful demand and latency signals, that exported spans faithfully reflect traffic bursts visible in proxy metrics, and that distributed traces reveal endto- end error paths (e.g., 404 failures). These findings indicate that a carefully orchestrated open-source stack can approximate the diagnostic value of white-box instrumentation. The contributions of this work are a reproducible pipeline design and an empirical assessment demonstrating how such a configuration can reduce mean-time-to-detect (MTTD) failures and support service-level objective monitoring under synthetic load scenarios.

Urban subsurface deterioration, including sinkholes, demands accurate and timely detection of underground cavities. Ground penetrating radar (GPR) offers non destructive subsurface imaging, but the large scale manual interpretation of noisy radargrams is impractical, which motivates automated detection with deep learning based object detectors. We apply YOLOv12, an attention enhanced one stage detector, to GPR based cavity detection and benchmark it against YOLOv8 under identical preprocessing and augmentation pipelines. On a public five-class dataset from The Open AI Dataset Project (AI-Hub, S. Korea), YOLOv12 achieves a mean average precision at intersection over union 0.50 of 0.940 and a mean average precision averaged over intersection over union values from 0.50 to 0.95 of 0.651, while sustaining at least 30 frames per second on a single graphics processing unit and outperforming YOLOv8 by up to 2.2 percentage points. These results show that attention based multi scale feature fusion in YOLOv12 substantially improves cavity detection performance, particularly for small and low contrast hyperbolic targets, and supports practical ground penetrating radar based monitoring of urban road infrastructure.

In this paper, a machine learning-based technique is presented for detecting compromised IoT devices in edge computing networks. The model profiles device behavior using parameters such as CPU usage, network traffic, and power data, detecting anomalies that suggest an attack may be in progress. The lightweight framework can achieve high detection accuracy at low computational cost and is capable of processing in realtime.

The more intense use of renewable energy brings about variability overwhelming grid stability. Five state-of-the-art machine learning models, namely, Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost are used in this paper to make predictions about grid stability on the basis of hourly data on electricity consumption and production. The binary stability label was created using a production-consumption imbalance and the models have been evaluated in terms of accuracy, precision, recall, F1-score and ROC-AUC. As it can be seen, the best performing model was CatBoost with its highest accuracy of 98.88 and ROC-AUC of 0.999. The findings point to the opportunities of AI-based forecasting to enhance the incorporation of the renewable energy and the overall stability of the grids.

Asthma is an important global health problem affecting nearly 300 million individuals and responsible for 250,000 deaths each year. Asthma is defined by obstruction to the airways, and difficult testing of lung function via spirometry or body plethysmography require complete cooperation by patients in all populations, including the elderly and those who are otherwise ill. Knees even more problematic is that smoking tobacco or respiratory changes in patients with asthma are only largely detectable when it is too late, meaning their respiratory performance has been compromised.

This research analyzes methods to enhance Urban Sound Classification performance by converting audio into Melspectrogram and MFCC images. Using the UrbanSound8K dataset, we compared single-representation and feature-level fusion strategies across CNN architectures. Results show that ResNet achieved the highest accuracy of 0.9594. However, a DenseNet-based fusion model proved more efficient, reaching a competitive accuracy of 0.9456 with fewer resources, demonstrating the potential for practical models that balance performance and efficiency.

This study proposes a hybrid model for accurate Korean named entity recognition and sentiment analysis within the tourism domain. To address the limitations of existing research due to the lack of domain-specific language models, we collaborated with industry partners to collect and construct tourism-specialized textual datasets. The resulting AI-Hub tourism corpus, developed through this data acquisition process, was utilized for model development and performance evaluation. Our approach combines the pre-trained KoELECTRA model with a Bidirectional Gated Recurrent Unit (BiGRU) architecture. KoELECTRA leverages the strengths of Transformer-based contextual representations, while BiGRU enhances contextual coherence by processing bidirectional sequential information. Evaluation results confirm that the proposed hybrid model demonstrates effective applicability for natural language processing tasks in the Korean tourism domain.

This study evaluates the performance characteristics of five OCI compliant container runtimes—runc, crun, youki, gVisor (runsc), and Kata Containers—through CPU, memory, and database benchmarks executed on an identical host environment. Sysbench CPU tests show that runc, crun, youki, and gVisor deliver nearly identical performance, while Kata Containers exhibits significantly lower throughput due to its virtualization-based architecture. Memory and PostgreSQL pgbench results highlight clearer differences: native runtimes achieve the highest throughput and lowest latency, gVisor shows moderate degradation due to system call mediation in user space, and Kata Containers demonstrates the greatest performance loss because of guest-to-host transitions and virtio based processing. Overall, the findings provide practical guidance for selecting container runtimes based on workload requirements, showing that native runtimes are optimal for latency-sensitive database services, while gVisor and Kata Containers are suitable for environments prioritizing stronger isolation at the cost of reduced performance.

Predicting accurate human migration patterns is crucial for effective urban planning. However, accurate human migration patterns prediction remains a challenging task. Existing methods, such as Graph Neural Network approaches, often overlook dynamic temporal variations and directional dependencies in large-scale migration data. To overcome this challenge, we propose MiGA-Net (Migration Graph Attention Network), a graph Neural network-based framework enhanced with an attention mechanism to capture complex spatiotemporal dependencies and highlight significant migration flows on the domestic and international level. We utilize two different datasets of the Shinan-gun, South Korea, for international and domestic regions. Experimental results show that the proposed MiGA-Net achieved superior performance over both datasets. The model achieved 0.0027 MAE for domestic flow and 0.0155 for international flow, demonstrating the effectiveness of the proposed framework.

Advanced Persistent Threats (APTs) represent a major headache for Security Operations Centers (SOCs) of the 21st century. Apart from being able to withstand constant monitoring, detection, and response, APTs are also extremely sophisticated and stealthy in nature. Centralized Intrusion Detection Systems (IDS), which are of a traditional nature, are usually not capable of providing adaptive, privacy-preserving, and collaborative detection functionalities across distributed networks. In this paper, a Federated Intrusion Detection Framework (FIDF), which uses Federated Learning (FL) to allow multiple SOC nodes to jointly train a smart detection model without the need to share raw data, is introduced. Local IDS agents, a central SOC aggregator, and a secure threat intelligence exchange mechanism are components of the system. The experimental performance is successful in showing that the detection accuracy is improved, the false positives are reduced, and the response time is enhanced for APT defense. The presented framework is a step toward federated solutions for the creation of a cybersecurity ecosystem that is scalable, privacy-aware, and resilient and is suitable for national defense infrastructures.

Employee placement is one of the vital functions of HR, which aligns the employee's skill with the organization's requirements. Traditional methods of placement have many shortcomings: skill mismatching, bias, and underutilization or misutilization of resources. The study uses machine learning (ML) to these problems by evaluating three algorithms-AdaBoost, support vector machine (SVM), and CatBoost-with demographic and job-related data from Kaggle. Results indicate that the maximum accuracy AdaBoost reached was 86%, then SVM with 81.4%, followed by CatBoost at 79%. These findings point to the reliability of AdaBoost in structured data and emphasize the potential that ML has for improving HR efficiency, employee satisfaction, and retention.

Retention of customers is vital for the long-term profitability of a business in the travel and tourism sector. In this sector churn, which is the continuous loss of clients, is an overwhelming hurdle. The major reason for this is that it is more time and cost-efficient to retain existing customers who spend money on acquiring new ones. There is a lot of value in being able to effectively predict customer churn. This knowledge will enable businesses to take a more proactive stance and offer personalized services in an effort to save and enhance the loyalty of the customer. The business problem here is to analyze and predict patterns of customer churn using various advanced algorithms like Categorical Boosting (CatBoost), Decision Tree (DT) and Knearest neighbors (KNN) techniques. Feature engineering and normalization are some of the many data preprocessing steps taken prior to training the machine learning (ML) model. Popular metrics such as accuracy, misclassification rate, precision, sensitivity, specificity, and F1 scores are used to examine the performance of each ML model.

This paper proposes a modular Learning Management System (LMS) to establish an integrated digital training hub for highly-skilled equipment maintenance personnel. While advanced technologies like Virtual Reality (VR) and Augmented Reality (AR) are being adopted, distributed educational resources and a lack of integrated platforms hinder systematic competency management. To address this, we designed and implemented a web-based LMS centered on technical personnel competency management, training performance analysis, and collaborative learning support. The system features a scalable architecture designed to serve as a future hub for an 'intelligent digital tutor ecosystem,' which will incorporate xAPI-based learning analytics and AR training linked with S1000D technical manuals, thereby contributing to the advancement of industrial technology education.

The rapid expansion of online apparel retail has increased the demand for accurate size recommendations that minimize returns and enhance customer satisfaction. This study presents a data-driven analysis of a bioelectrical impedance analysis (BIA)–based size recommendation system implemented on a live e-commerce platform. Using anonymized transaction and feedback data from Boastfit.com, the research compares behavioral and perceptual outcomes between BIA-based recommendations and conventional size guides. The BIA group recorded a return rate of 9.6 percent compared with 17.2 percent in the control group, an average satisfaction score above 8 on a ten-point scale, and a repurchase ratio of 79 percent. These results confirm that physiological data–driven personalization improves predictive accuracy, post-purchase satisfaction, and repurchase intention. The findings contribute to next-generation computing and fashion retail analytics by demonstrating how body-composition data can be integrated into intelligent recommendation systems to enhance user trust and sustainable engagement.

Biomedical image captioning has become a rapidly advancing research area aimed at supporting clinical workflows by automatically generating descriptive medical reports. However, existing models often suffer from hallucinations, where clinically incorrect findings are described, and semantic misalignment, where captions fail to reflect key visual cues. These issues largely arise from architectures trained on general-domain data, relying on a single encoder, or models lacking robust visual–textual grounding. To overcome these challenges, MedSwinGPT, a reward-guided dual-encoder prefix-fusion model is proposed. It integrates MedCLIP (medical domain encoder) and Swin Transformer (general visual encoder) through a single linear projection to capture complementary global and local visual information. The fused representation conditions BioGPT via prefix tokens, enabling domain-aware and semantically coherent caption generation. To strengthen visual–textual alignment, we jointly optimize Cross- Entropy (CE) and Contrastive Learning (CL) objectives, followed by Self-Critical Sequence Training (SCST) fine-tuning with a multiobjective reward combining BERTScore and contrastive similarity. Evaluated on the ROCO radiology dataset, our reward-guided MedSwinGPT surpasses existing baselines across standard metrics. Qualitative results further demonstrate improved clinical accuracy, semantic grounding, and reduced hallucinations, underscoring its potential for reliable biomedical caption generation.

Cone-beam computed tomography (CBCT) is essential in adaptive radiation therapy (ART), yet its clinical utility is hindered by high noise levels, artifacts, and degraded textures. This study introduces a deep learning framework based on a Conditional Denoising Diffusion Probabilistic Model (DDPM) to synthesize high-quality CT (sCT) images from CBCT scans. The model incorporates a specialized encoder and Fusion Block for better fusing input and label images and preserve fine anatomical details. Trained on paired CBCT and deformed CT(dCT) pelvic datasets, the proposed method significantly reduces noise and artifacts while enhancing anatomical fidelity. This approach promises to improve CBCT usability in clinical workflows and enhancing ART planning accuracy.

This Leishmaniasis is common skin lesion parasitic disease caused by Leishmania protozoan parasites on exposed body and its polymorphic nature complicates to diagnosis because the lesion may create confusion with other dermatoses likewise fungi, bacteria and non-infectious diseases. The molecular techniques, microscopy, culture, and rapid diagnostic test are conventional methods that are timeconsuming, expensive, susceptible to errors with limited resources in health care services. Early diagnosis with timely identification of multifaceted Leishmaniasis is aided to selection of therapy and provide comfort to patient to combating with it. The promising integration of artificial intelligence (AI) with medical diagnostics has efficacy in numerous fields of identification of diseases as in Dermatology research. The fast, efficient and automatic diagnosing of leishmaniasis with microscopic images of lesion's seamer with VGG-16 deep learning (DL) model is the approach to reach the objective of this designed research to identify the negative and positive results. The exceptional performance of designed VGG-16 is achieved with accuracy of 88.14%, precision 100%, sensitivity 77.42%, specificity 100%, F1-score 0.87%, and ROC curve 97%. The proposed modified VGG-16 model is more precise, swift, reliable, efficient, effectual, economical and user-friendly substitute to address all key factors than human resource to find the leishmaniasis affected that may support medical care services.

Heart disease is a major cause of mortality in the world that is in dire need of accurate, interpretable predictive measures that could be utilized to manage it proactively. The writer of this paper proposes an Explainable AI (XAI) Ensemble Machine Learning model to predict heart disease using an 1,025 patient record dataset. To achieve methodological rigor and generalization, 5- Fold Stratified Cross-Validation (CV) was used to evaluate all models, such as LightGBM and Random Forest. LightGBM model was stable and better in performance as it had Mean CV Accuracy of ±0.9620 ±0.0178. Integration of XAI (SHAP/LIME) is the means of creating clinical trust; analysis has confirmed maximum heart rate (thalach) and type of chest pain (cp) as medically significant characteristics. This framework supports the sustainable smart city healthcare through a highly transparent decision-support system, which manages the resources in optimizing scalable public health programs.

The need to detect lung cancer accurately and early has been brought out through the fact that lung cancer has remained one of the top causes of cancer-related deaths in the world. A CT scan is usually interpreted manually which makes this time consuming and subjective. This paper hypothesizes that an explainable implementation of automated deep learning can be used to classify lung cancer based on the transfer learning models (VGG16, VGG19, ResNet50, and EfficientNetB3) on the IQ-OTH/NCCD dataset. The dataset was stratified by means of the SMOTE and was split into 80 and 20 percent training and validation subsets, respectively. All the pretrained CNNs were fine-tuned with the Adam optimizer and categorical cross-entropy loss. VGG16 performed the best and had a validation accuracy of 98.64, precision and recall of 98, and ROCAUC of 99. Visualization of tumor regions was done using explainable AI techniques (Grad-CAM, LIME), which are interpretable and have diagnostic transparency. The suggested framework proves that transfer learning combined with XAI is more effective in terms of accuracy and reliability in diagnosing medical images and is one of the steps to clinically reliable smart healthcare systems.

Early diagnosis of crop diseases as- sists the farmers to increase their output and save on their earnings. In this research, the Corn Leaf Disease Dataset with four classes is used, namely, Cercospora leaf spot, Common rust, Northern Leaf Blight, and Healthy. An image transformer (ViT) model is used, and image patches are treated as sequences, which enables them to cap- ture fine and global details. Application of transfer learning on a trained ViT enhances the accuracy and lowers the training time. Accuracy, precision, recall and F1-score measurements indicate that ViT is similar in performance to CNN models, and it is therefore useful in the detection of plant diseases.

Synchronous ring all-reduce is widely adopted for multi-GPU training due to its simplicity and scalability. However, its convergence-time advantage collapses in heterogeneous or unstable environments where a single slow worker (straggler) throttles overall progress. We present SAWS (Straggler-Aware Weighted Synchronization), a lightweight technique that (i) detects transient stragglers via adaptive, profile-driven timeouts, (ii) isolates them from the synchronous fast path without job aborts, and (iii) merges their partial progress through weighted model averaging proportional to processed-data ratio. In experiments on ResNet-18/CIFAR-10 with injected 5× slowdowns, SAWS improves wall-clock training time by up to 3.2× over vanilla Horovod while matching final accuracy within <1% of fully synchronous baselines. Compared to a straggler-drop variant, SAWS achieves competitive time-to-accuracy and consistently higher final validation accuracy.

Artificial Intelligence (AI) continues to evolve as applied to all aspects of life, such as Finance, Healthcare, and Education. However, advances in AI still depend on the availability and equal access to data worldwide. The Foundation for AI Rights and Equity of Education & Healthcare (FAIREEH) is a non-profit organization that aims to make AI equitable and available to all humanity, while facilitating ethical and data-driven innovation. FAIREEH is based on the principle that all people have a right to AI knowledge, tools, and data. Recently, FAIREEH has created a synthetic data initiative to address the data imbalance in worldwide data distribution. The organization is developing easily accessible and privacy-preserving datasets made of synthetic data in various fields and for various applications, such as image classification, image detection, EEG impression data, etc. The initiative bridges the digital divide and empowers researchers and institutions from under-resourced environments to make meaningful contributions to global AI development.

The present article describes the Cognitive GRC Cloud (CGC), a hybrid AI-powered model that integrates psychological behavior aspects with governance, risk, and compliance (GRC) modules for the dynamic management of a project. Conventional GRC models are mainly concentrated on technical automation and, thus, ignore the fact that cognitive biases affect the decision-making process and the perception of risk. The framework introduced relies on AI analytics and the psychological feedback system to detect, measure, and neutralize these biases on the spot. Their experimental performance demonstrates that CGC achieves lower levels of project risk indices, schedule slippage, and compliance violations than baseline systems. The findings demonstrate that integrating cognitive modeling with AI-based governance is a viable approach to cloud-enabled project management that is more efficient and less prone to bias.