Early and accurate detection of brain tumors remains a major challenge in medical imaging due to limited dataset size, trained deep learning models, patient variability, and the complexity of manual interpretation; traditional approaches sometimes rely on lower-level feature extraction, which may not apply well to medical images. To address these challenges, we propose a unique explainable deep learning model that integrates Tiny-ConvNeXt and DenseNet169 using multi-stage brain tumor classification. Two clinically approved, open-access Magnetic Resonance Imaging (MRI) datasets were combined to generate a bigger, more customized dataset, and substantial data augmentation techniques were used to improve model generalization. According to experimental results, our proposed model outperforms current state-of-the-art methods with a classification accuracy of 99.74%. Additionally, statistical significance tests confirmed the incremental contribution of each model component, and ablation studies showed the robustness of the conclusions. Additionally, explainable artificial intelligence techniques like Local Interpretable Model-agnostic Explanations and Gradient-weighted Class Activation Mapping++ were employed to enhance interpretability, enabling visual explanations of tumor localization. These results indicate that the suggested model provides a transparent and reliable framework for brain tumor detection, with potential applications in practical clinical decision support systems. Received: 18 May 2025 | Revised: 29 April 2026 | Accepted: 7 July 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in the American Society of Clinical Oncology at https://www.cancer.org/cancer/types/brain-spinal-cord-tumors-adults/key-statistics.html, in Kaggle at https://www.kaggle.com/datasets/ahmedhamada0/brain-tumor-detection, and in Kaggle at https://www.kaggle.com/datasets/navoneel/brain-mri-images-for-brain-tumor-detection. Author Contribution Statement Md Sadi Al Huda: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration. Kazi Tanvir: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Visualization. Zubaida Akhter: Software, Validation, Formal analysis, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization. Md. Shahidul Khan Pappo: Validation, Writing - review & editing, Visualization. Md. Asraf Ali: Resources, Writing - review & editing, Supervision, Project administration. Nasim Ahmed: Resources, Writing - review & editing, Supervision, Project administration.
The use of artificial intelligence (AI) in cybersecurity serious games and training platforms is an innovative approach, which enhances the effectiveness of security training by improving the development of practical skills. Numerous studies have focused on the potential of AI-enhanced serious games, highlighting the flexibility and adaptability of this form of education in the cybersecurity domain. The overall aim of this study is to map out the AI technology landscape relevant to cybersecurity games and assess relevant theory and pedagogical frameworks. First, a systematic review of publications from 2019 to 2025 is performed to identify the primary AI techniques and technologies adopted in cybersecurity and to find those relevant to serious games and training platforms. Then, the factors that contribute to their success and the barriers that affect their implementation in game-based environments are analyzed, together with the theoretical frameworks and pedagogical models that support the design of AI-enhanced training platforms. As an illustrative example of these results, an AI-based reverse Turing game is presented. The article will be of interest to those who design cybersecurity games and also to educators who are trying to create training formats that are effective in responding to the fast-changing cybersecurity environment. Received: 12 August 2025 | Revised: 25 May 2026 | Accepted: 26 June 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Bilgin Metin: Investigation, Resources, Writing – review & editing, Supervision, Project administration. Tuncay Avci: Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Atakan Yeşilkayali: Investigation, Resources, Data curation, Writing–original draft, Writing – review & editing, Visualization. Hikmet Sami Karaca: Investigation, Resources, Data curation. Ali Nehir Dündar: Investigation, Resources, Data curation. Martin Wynn: Writing – review & editing, Supervision.
Generative artificial intelligence (GenAI), particularly since the introduction of ChatGPT, has not only led to a quantitative increase in publications in management and business research but also created a qualitative transformation in research themes and collaborative structures. However, the current literature is insufficient to comprehensively present the epistemic and thematic consequences of this transformation in the field of management. This study aims to fill this gap by examining 316 articles indexed in the Social Sciences Citation Index database within the Web of Science Core Collection and published between 2023 and 2025 using bibliometric analysis. The analyses show that normative and governance-oriented themes such as ethics, artificial intelligence (AI) literacy, and agentic AI have become increasingly visible in the management literature in the post-ChatGPT era; rather than displacing performance- and productivity-centered approaches, these normative themes emerged alongside them within an increasingly diversified keyword and thematic structure. The findings reveal that GenAI research in management has evolved from being merely a technological innovation to a research axis that reshapes discussions on organizational responsibility, human–machine interaction, and governance. In this respect, the study goes beyond mapping the current state of the literature and offers theoretical and methodological implications on how GenAI should be positioned in management research. Received: 18 May 2026 | Revised: 24 June 2026 | Accepted: 15 July 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Irem Tanyıldızı Baydili: Conceptualization, Methodology, Formal analysis, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Burak Tasci: Software, Writing–original draft, Writing – review & editing, Visualization. Sengul Dogan: Validation, Investigation, Writing – original draft, Writing – review & editing, Visualization. Turker Tuncer: Validation, Investigation, Writing – original draft, Writing – review & editing, Visualization.
The precise poultry chicken price forecasting is necessary to maintain food security and market stability as well as protein demand mitigation in many countries. Existing machine learning-based forecasting models have problems with feature selection, hyperparameter tuning, linear assumptions, lower accuracy values, and insufficient data integration. Additionally, they did not offer highly accurate and month-based poultry chicken price forecasts. A deep learning (DL)-based framework for 30-day broiler chicken price forecasting is presented in this work. This work makes use of daily data from multiple sources, such as retail prices, feed costs, and macroeconomic variables. The proposed method combines ensemble feature selection, resilient feature engineering, and sophisticated data preparation, and it is optimized using walk-forward cross-validation and Bayesian optimization using Optuna. This work proposed a hybrid DL ensemble technique for monthly poultry price forecasting that combines Long Short-Term Memory (LSTM) networks with the Gated Recurrent Unit method. The proposed model outperforms previous studies by obtaining at least 2% lower MAPE and 10% higher R 2 value, with a MAPE score of 3.69% and R-squared value of 0.832. The feature contribution is explained in this work for forecasting using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) methods. Received: 8 August 2025 | Revised: 13 April 2026 | Accepted: 25 June 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/mhchowdhury37/poultry-chicken-dataset-new. Author Contribution Statement Mohammad Yasin Arafat: Methodology, Software, Validation, Formal analysis, Investigation, Resources. Mahfuzulhoq Chowdhury: Conceptualization, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.
Leukemia, particularly acute lymphoblastic leukemia (ALL), is one of the most prevalent and fatal blood cancers, with more than 460,000 new cases and over 300,000 related deaths reported globally in 2021. Despite advances in deep learning, conventional convolutional neural network models applied to blood cell cancer classification often deliver suboptimal performance, with reported accuracies frequently below 90%. These limitations are due to difficulty in capturing subtle morphological variations, high similarity among malignant subtypes, and problems like class imbalance and noise in peripheral blood smear images. To overcome these difficulties, this study proposes a VGG16-based multiscale feature fusion (MFF) model that fuses the shallow and deep feature maps from different convolution blocks of the VGG16 backbone. Before classification, these feature maps are spatially aligned and concatenated into a single multiscale representation at a consistent resolution. The model was tested on the Blood Cell Cancer (ALL) dataset, which includes 3242 peripheral blood smear images that are classified into four categories: benign, malignant early Pre-B, malignant Pre-B, and malignant Pro-B. The experimental results showed that the proposed VGG16-MFF obtained 99.69% accuracy, precision, recall, and F1-score, which significantly outperformed baseline architectures such as ResNet50 (50.00%), ResNet101 (76.85%), VGG16 (88.27%), and VGG19 (87.34%). These results support the effectiveness of MFF in improving classification robustness, and future work may consider its extension to larger and more varied medical imaging datasets. Received: 30 August 2025 | Revised: 14 April 2026 | Accepted: 25 June 2026 Conflicts of Interest The author declares that he has no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/mohammadamireshraghi/blood-cell-cancer-all-4class. Author Contribution Statement Simeon Yuda Prasetyo: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.
Artificial intelligence (AI) is increasingly being used to assist physicians with their medical diagnosis and patient care and improve communication between doctors and their patients. This paper demonstrates how machine learning (ML), a subset of AI, can pre-screen electroencephalograms (EEGs) to assess the overall quality of the EEG and identify artifacts or abnormalities. We developed the “Brain Panel,” an automated ML tool that supplements clinical neurophysiologists’ quality review screening. The Brain Panel generates results prior to visual inspection or quantitative EEG (QEEG) analysis, highlighting potential technical issues or clinical concerns through novel metrics and interpretation methods. We subjected 100 Brain Panel reports and 100 corresponding physician reports from the same EEGs to independent AI evaluations using Grok and Claude. Results show that the Brain Panel provides a sensitive, neurologically informed method to estimate artifacts, assess EEG quality, detect drowsiness, and indicate the likelihood of brain abnormalities. By identifying optimal combinations of Brain Panel metrics, we created detection algorithms that achieved sensitivities of 89–95% for clinically relevant findings. This approach demonstrates that using AI to analyze and interpret automated Brain Panel reports produces a system with clear clinical value for detecting technical and clinical issues in EEGs, enhancing neurologist productivity without replacing expert judgment. Received: 3 July 2025 | Revised: 25 March 2026 | Accepted: 19 May 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in the Online Data Resource https://www.dropbox.com/scl/fo/9lvvnoui19c16oehliarf/AJumW34SIPIwtjmNkRPumCU?rlkey=sw8lxip2dpnlo57eb5ag0w1c3&e=1&dl=0. Author Contribution Statement Thomas Collura: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Agostino Rosace: Methodology, Software, Formal analysis. Robert Turner: Conceptualization, Validation, Investigation, Resources, Supervision. David Ims: Conceptualization, Validation, Investigation, Resources. Bill Brubaker: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Project administration.
The rapid growth of Indonesian digital news content necessitates automated sentiment analysis systems capable of handling formal journalistic discourse, which differs substantially from the social media text used to train most existing sentiment classifiers. This study investigates domain adaptation for Indonesian news sentiment analysis by fine-tuning IndoBERTa, a RoBERTa-based model for Indonesian language processing. Using a structured data mining workflow inspired by the Cross-Industry Standard Process for Data Mining, we collected and manually annotated 1300 news articles from two major Indonesian news portals (Kompas and Detik) into positive, negative, and neutral categories. Zero-shot evaluation using a social-media-trained sentiment model yielded 14% accuracy, with over 89% of samples predicted as neutral, indicating a strong domain-induced classification bias. After fine-tuning, IndoBERTa achieved 98% accuracy, a Cohen’s kappa score of 0.98, and balanced F1-scores across all sentiment classes. While keyword-guided data collection and class balancing likely inflate absolute performance, the results clearly demonstrate the effectiveness of domain-specific adaptation for Indonesian news sentiment classification. The fine-tuned model is deployed in a real-time analysis pipeline and publicly released to support further research in Indonesian Natural Language Processing (NLP). Received: 31 August 2025 | Revised: 9 February 2026 | Accepted: 7 May 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. Author Contribution Statement Desi Masdin Dama: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Visualization. Tati Mardiana: Conceptualization, Methodology, Validation, Writing – review & editing, Supervision. Riki Supriyadi: Conceptualization, Methodology, Validation, Supervision. Zico Pratama Putra: Conceptualization, Methodology, Validation, Writing – review & editing, Supervision. Dicky Octaviano: Validation, Investigation, Writing – review & editing. Achmad Bayhaqy: Software, Resources, Data curation.
This study explored the adaptive remodeling of the foot’s neuro-biomechanical system induced by dance training. A multimodal function-and-control framework was established, integrating kinematic and kinetic analyses, electromyographic synergy modeling, and finite element simulation to reveal adaptive changes from multiple perspectives. The results showed that Latin dancers exhibited greater ankle and metatarsophalangeal flexion angles and joint moments but lower angular velocities during landing, indicating a more efficient energy absorption and release mechanism. Their muscular synergy patterns were activated earlier and in a more compact manner. Finite element analysis revealed higher stress concentrations in the first and second metatarsals and phalanges, suggesting adaptive changes in load transmission. These findings demonstrate that dance training can induce a feedforward-driven, multilevel coupling mechanism that enhances coordination among neural control, mechanical output, and structural remodeling, thereby improving impact absorption and postural stability. This study reveals the hierarchical plasticity of the neuro-biomechanical system during complex skill learning and provides new perspectives for the development of precision training, injury prevention, and rehabilitation strategies. Received: 17 November 2025 | Revised: 4 February 2026 | Accepted: 2 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Figshare at https://doi.org/10.6084/m9.figshare.31333936. Author Contribution Statement Xiangli Gao: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Datao Xu: Methodology, Resources. Minjun Liang: Methodology, Resources. Zanni Zhang: Investigation, Data curation. Tianle Jie: Investigation, Data curation. Huiyu Zhou: Methodology. Jingyi Ye: Methodology. Zsolt Radak: Supervision. Yaodong Gu: Supervision, Project administration, Funding acquisition.
Generative artificial intelligence (GenAI) tools are redefining education, particularly in science disciplines such as physics. Even though these systems are widely used, there are still only a few studies that clearly summarize their use in teaching and their impact. In this PRISMA-guided systematic review, we analyzed 21 empirical studies (from 2023 to 2025) on the role of GenAI in physics education. We combined bibliometric mapping with content-based refinement and classified the studies by educational level, tools used, research methodology, and pedagogical role. Three main functions were identified: problem solver (57% of studies), student’s tutor (38%), and teacher’s assistant (33%). Problem-solving roles demonstrated variable accuracy, ranging from 100% to 0% across task types. Tutoring roles, though few, showed the strongest evidence of positive learning outcomes. Assistant roles mainly supported efficiency in grading and content generation, with less direct evidence of learning outcomes. The review contributes by providing this physics-specific PRISMA synthesis of GenAI in education and by identifying the roles GenAI tools assume in this context. This research is limited by the small, diverse set of included studies, which focus mainly on university contexts, and by its reliance on Scopus-indexed literature. Our findings suggest that GenAI tools can potentially improve conceptual understanding, formative assessment, and instructional efficiency. However, their effective integration requires human supervision, critical evaluation, and strategies to address issues of accuracy, equity, and multimodality. Received: 12 August 2025 | Revised: 29 December 2025 | Accepted: 28 January 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Maria Moundridou: Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Supervision. Nikolaos Moutis: Conceptualization, Methodology, Writing – review & editing, Supervision, Project administration. Konstantinos Charalampopoulos: Investigation, Writing – review & editing. Nikolaos Matzakos: Conceptualization, Methodology, Formal analysis, Data curation, Writing – original draft, Writing – review & editing, Visualization.
Cervical spine fractures can cause instability in the cervical spine and may result in spinal cord injuries. If not promptly detected and treated, these fractures may deteriorate over time. Hence, the diagnosis of cervical spine injuries must be conducted urgently to prevent further complications. Current deep learning models face limitations in accurately diagnosing these fractures due to issues such as insufficient attention to subtle fracture features and poor generalization across varying scales. This research proposes MSA–CSpineNet (Multi-Scale Spatial Attention Cervical Spine Network), a deep learning framework for accurate cervical spine fracture diagnosis from computed tomography scans. The pre-trained MobileNet was used for feature extraction, which was passed to the multi–scale and attention module for relevant feature selection. The results show an accuracy of 99.75%, sensitivity of 99.99%, specificity of 99.50%, and precision of 99.50%. Compared with the existing state-of-the-art approaches that used transfer learning and conventional convolutional neural network techniques, experimental results demonstrated that the proposed MSA-CSpineNet outperforms existing methods in image classification. The results of this research have the potential to greatly improve cervical spine fracture early diagnosis and treatment, which would benefit patients' outcomes. Gradient-weighted class activation mapping visualization demonstrates that the model develops spatially selective attention patterns, providing interpretability that supports clinical trust in model predictions.
Microscopic imaging is fundamental to modern medical diagnostics, offering detailed structural and morphological insights essential for studying cellular behavior and detecting diseases. However, automated analysis of such images remains challenging due to variability in cell shapes, overlapping structures, noise, and inconsistent staining. These factors complicate accurate interpretation and necessitate advanced computational techniques. A critical step in this process is precise segmentation of cellular structures, which significantly impacts downstream tasks like classification and diagnosis. Effective segmentation enhances visual clarity and improves the reliability of computer-assisted diagnostic systems by identifying well-defined regions of interest. To address these challenges, this paper proposes a novel segmentation framework, SUFGSO (Superpixel-based Fuzzy Galactic Swarm Optimization). The approach integrates type-II fuzzy logic to manage uncertainty and ambiguity, with galactic swarm optimization, a metaheuristic inspired by hierarchical swarm intelligence. Additionally, superpixel techniques are employed to group pixels into meaningful regions, reducing computational complexity and improving spatial coherence, especially in high-resolution images. The proposed method is evaluated using four established cluster validity indices to ensure a comprehensive assessment of segmentation performance. Experimental results demonstrate that SUFGSO achieves improved accuracy and efficiency, indicating its effectiveness as a practical tool for microscopic image analysis in medical applications. Received: 2 June 2025 | Revised: 12 March 2026 | Accepted: 15 May 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in the Center for Research in Biological Systems at https://www.crbs.ucsd.edu/crbs-projects/highlighted-research-projects?highlight=94. Author Contribution Statement Debasish Biswas: Methodology, Software, Formal analysis. Shouvik Chakraborty: Conceptualization, Writing – original draft, Supervision, Project administration. Chinmoy Ghorai: Validation, Writing – review & editing, Supervision, Resources.
Recent breakthroughs in robotics and artificial intelligence have permitted more autonomous systems, although most current techniques are limited by inflexible control structures developed from classical automation. Such frameworks stand in stark contrast to biological systems, which develop intelligence via ongoing structural and functional reconfiguration. Neural morphogenesis takes a developmental approach to machine intelligence, seeing robotic cognition as a dynamic and adaptable process. In this paradigm, artificial agents gradually adapt their internal circuitry, behavioral tactics, and physical morphology as they interact with their surroundings. Learning and design therefore occur together rather than sequentially. The approach integrates embodied perception and safety-governed developmental adaptation within a closed-loop control framework. We introduce a three-layer developmental controller that couples online neural plasticity, morphogenetic regulation, and energy-aware policy adaptation under explicit rollback safety constraints. In over 105 simulation cycles involving populations of 50, 100, and 200 agents, neural morphogenetic architectures achieved energy efficiency gains of up to 47%, network modularity increases of around 40%, and reductions in informational entropy between 13 and 15% (p < 0.01) when compared to deep reinforcement learning and model predictive control baselines. These gains were accompanied by improved cooperative behavior, steady performance under stress, and greater resistance to environmental shocks. Preliminary bio-hybrid tests suggest a link between morphological flexibility and integrated information (Φ), emphasizing the impact of physical structure on cognitive capabilities. These findings establish brain morphogenesis as a strong basis for adaptable, resilient, and sustainable intelligent robotic systems. Received: 5 September 2025 | Revised: 16 December 2025 | Accepted: 26 February 2026 Conflicts of Interest The author declares that he has no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Zenodo at https://zenodo.org/records/18175173, reference number [45]. Author Contribution Statement Edwin Gerardo Acuña Acuña: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.
We investigate whether complexity-inspired network descriptors extracted from visibility-graph (VG) representations of price dynamics can improve short-horizon forecasting of the Standard & Poor’s 500 movements. We consider two aligned forecasting tasks at a 14-day horizon: (i) Up/Down directional classification and (ii) regression of the 14-day-ahead standardized forward return. Using daily close prices from 1981 to 2025, we construct sliding window graphs (50/75/100 trading days) and compute a broad set of interpretable spectral, topological, and mesoscopic features, including spectral radius, algebraic/natural connectivity, graph index complexity, clustering/transitivity, efficiencies, assortativity, and path-length statistics. Beyond the standard (“natural”) VG, we introduce a volatility-adaptive quantile VG (VAQ-VG) and two additional geometric/motif descriptors—visibility angle entropy and square-motif density—to form a multi-view feature bank. Features are lagged (1–7 days), standardized, and filtered via mutual information. Six learning algorithms (regularized linear models, random forest, k-nearest neighbors, Gaussian process, and stochastic gradient descent) are evaluated under a purged expanding cross-validation protocol with a 150-day embargo to eliminate sliding window leakage, with hyperparameters tuned by randomized search. Empirically, the directional task shows modest but consistent skill: top models reach the area under the receiver operating characteristic curve ≈ 0.70–0.72 with accuracy ≈ 0.65–0.66, remaining reliably above chance across thresholds. In contrast, point prediction of 14-day return magnitudes remains challenging, with the best R 2 below 0.20 and limited gains from more complex regressors. Overall, VG-derived features provide a compact, interpretable representation that supports leakage-robust directional forecasting, while VAQ-VG yields small, model-dependent shifts that suggest complementary structure across views rather than a decisive single best construction. Received: 31 August 2025 | Revised: 24 February 2026 | Accepted: 10 April 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in GitHub at https://github.com/Butman2099/Complex-systems-book and https://github.com/Butman2099/Complex-Network-Paper-for-Artificial-Intelligence-and-Applications-AIA-journal. Author Contribution Statement Andrii Bielinskyi: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Vladimir Soloviev: Conceptualization, Methodology, Writing – review & editing, Supervision, Project administration. Andriy Matviychuk: Conceptualization, Methodology, Validation, Investigation, Writing – original draft, Writing – review & editing. Vitalii Bezkorovainyi: Methodology, Software, Formal analysis, Investigation, Data curation, Writing – review & editing, Visualization.
A hybrid method is proposed for solving large-scale mixed-integer linear programming (MILP) problems that arise in the optimization of the generation capacity structure of electric power systems combining conventional and renewable energy sources. The novelty of the proposed approach lies in combining the decomposition of the original generation expansion planning problem into investment-level search and operational-level evaluation with evolutionary exploration of the discrete space of generation capacity decisions and exact parallel solution of operational MILP subproblems on high-performance computing (HPC) resources. This combination enables detailed operational evaluation without replacing the operational model with a surrogate approximation. Numerical experiments using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and the Solving Constraint Integer Programs (SCIP) solver showed that the global optimum was achieved in 98% of runs for a population size of 80 and that, compared with the single-threaded SCIP baseline, the 128-thread configuration reduced the average runtime for successful runs by 3.77×. At the same time, scaling with respect to computation time and memory usage exhibited nearly linear behavior. A comparison with the alternative Evolutionary Centers Algorithm demonstrated the superiority of CMA-ES in convergence reliability. The obtained results indicate that the proposed method is well-suited for deployment in HPC environments and can serve as an effective tool for strategic planning of electric power system development. Future research should focus on extending the applicability of the approach to multi-period planning problems and analyzing the impact of algorithm parameters on convergence and solution accuracy. Received: 30 August 2025 | Revised: 24 March 2026 | Accepted: 19 May 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.20214170. Author Contribution Statement Sergii Saukh: Conceptualization, Methodology, Validation, Formal analysis, Resources, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Taras Puchko: Methodology, Software, Investigation, Data curation, Writing – review & editing, Visualization.
The most important element of cultural heritage is the embroidery used in the ancient clothes. Embroidery used in the ancient clothes represents centuries of creative tradition, regional uniqueness, and the passing down of skills from one generation to the next. These elaborate textile practices, which have their roots in manual craftsmanship and oral instruction, are increasingly threatened by industrial production, globalization, and the slow decline of traditional artisans. By developing an automated identification structure based on MobileNetV2, an efficient yet highly effective convolutional neural network (CNN), this work presents an AI approach to support the preservation of regional embroidery. The model used in the paper was trained very carefully with a dataset of 5,200 high-resolution images that included a range of regional embroidery styles using pretrained ImageNet weights. With an identification accuracy of 97.3%, the suggested MobileNetV2 outperformed traditional CNN architectures like VGG16 and ResNet by over 2.3%. The results demonstrate how small AI models can make a significant contribution to the preservation of cultural heritage by providing a scalable method for storing and assessing textile arts for museums, scholars, and craftspeople. This methodology exhibits promise in fields such as digital archives, fashion technology, and cultural data analytics for heritage conservation. Received: 17 August 2025 | Revised: 15 December 2025 | Accepted: 3 March 2026 Conflicts of InterestThe authors declare that they have no conflicts of interest to this work. Data Availability StatementThe data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/ask1999/embroiderydataset. Author Contribution StatementAmir Sohail Khan: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Junjie Zhang: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.
Schizophrenia is a complex psychiatric disorder in which traditional diagnostic methods, relying on subjective clinical assessment, suffer from significant limitations in scalability and early intervention. Machine learning can bring a paradigm shift to objective, data-driven diagnosis, but a critical gap exists between its technical performance and real-world clinical translation. The review systematically analyzes and synthesizes findings from 30 seminal studies on ML applications in schizophrenia diagnosis within the period of 2018–2026, following a structured methodology to evaluate models, datasets, performance, and translational challenges. Our review indicates that, though high accuracies (82–96%) have been reported using conventional and deep learning models in controlled research settings, essential barriers critically restrain their clinical utility: heavy class imbalance, a lack of model interpretability, that is, the “black-box” problem, biased datasets, and high computational costs. These limitations diminish their diagnostic accuracy in the real world and clinician trust. Hybrid ML frameworks are available with integrated explainable AI for transparency, GANs for data augmentation, and federated learning for privacy-preserving collaboration in order to bridge these gaps. The road to equitable precision psychiatry will have to be charted by overcoming socio-technical barriers through interdisciplinary co-design and adherence to emerging global ethical AI standards, for example, IEEE P7000, developing lightweight, accessible tools. This review provides a strategic roadmap to transition ML from a research tool into a clinically viable, equitable, and trustworthy asset for global mental health, with the ultimate aim of reducing misdiagnosis and improving patient outcomes. Received: 22 November 2025 | Revised: 3 April 2026 | Accepted: 25 May 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Syed Mossabbir Hossain: Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Writing – original draft. Nitun Kumar Podder: Conceptualization, Software, Validation, Resources, Writing – review & editing, Supervision, Project administration. Md. Raihanul Haque: Formal analysis, Data curation, Visualization. Poly Akter: Formal analysis, Investigation. Tasfia Rahman Asma: Formal analysis, Investigation.
This study presents a theoretical and methodological framework for examining trust and decision-making in AI-augmented human–machine teaming within military contexts. The framework combines a supervised learning backbone (a pre-trained Random Forest classifier) with a deterministic rule-based override layer that encodes non-negotiable constraints aligned with key laws and principles of armed conflict, including distinction, proportionality, and military necessity, as well as related rules-of-engagement logic. Using a structured generator of combat-relevant targeting scenarios, the system produces recommendations that can be accepted, escalated, vetoed, or deferred, allowing shifts in cognitive authority and reliance to be observed and measured. An interactive, scenario-driven interface exposes calibrated confidence, salient feature cues, and explicit override traces to support verification and controlled reliance in uncertain situations. On a large synthetic scenario corpus, the model exhibits stable performance and well-calibrated probability estimates. At the same time, the guardrail layer systematically redirects borderline engage outputs toward safer outcomes and audit-ready escalation states. The artifact is positioned as a research instrument rather than an operational decision-making authority. It is designed to elicit and quantify trust calibration behaviors, including cautious skepticism, confident alignment, and deliberative hesitation, across varying levels of complexity and ambiguity. The design is released for replication and iterative refinement, supporting interdisciplinary evaluation of transparent, doctrine-compatible AI decision support and providing a practical basis for controlled user studies on trust, bias, and ethical judgment in military human–AI teaming. Received: 27 February 2025 | Revised: 9 December 2025 | Accepted: 25 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data available on request from the corresponding author upon reasonable request. Author Contribution Statement Janar Pekarev: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Priit Värno: Validation, Investigation, Writing – review & editing.
Federated learning (FL) is promising for intrusion detection for Industrial Internet of Things (IIoT) without the necessity of centralizing raw telemetry, but there exist two stumbling blocks: (i) limiting what can be inferred about the clients (sites/devices) from their updates and (ii) providing reliability under heterogeneous, non-independent and identically distributed (non-IID) data with faulty or Byzantine members. We outline a systems design that intertwines client-level differential privacy (DP) with robust aggregation and experiment on the Edge-IIoTset workload specification and deterministic synthetic experiments with the goal of generating fully reproducible figures. The server applies per-client L2 clipping, adds calibrated Gaussian noise, and tracks privacy with Re´nyi DP (RDP) under subsampling; aggregation uses coordinate-wis → enforcing the target budget. Sweeping privacy budgets 0.5 reduces macro-F1 0.93 → 0.78 (−19.2%), AUROC 0.96 𝜀 ∈ {0.5 e median, → , 1 0.86 (−10.4%), and worst-client F1 0.89 , 2, 5 β , -trimmed mean, or Krum, with an auditable stop-on- 10} yields clear privacy–utility frontiers: tightening from → 0.70 (−21.3%). Under 10% 𝜀 controller 𝜀 = 10 corrupted clients at 𝜀 = 2, coordinate-wise median improves macro-F1 over mean by +11.1% and worst-client F1 by +20.3%. Time per round scales with the number of selected clients m = qK: 0.60 → 2.60 s as K = 10 → 200 at q = 0.1. The recipe exposes deployable knobs (𝜀, C, q, aggregator), auditable privacy via stop-on-𝜀, and tail-aware reporting—charting a practical path to regulator-aligned, privacy-preserving FL for IIoT intrusion detection. Received: 11 December 2025 | Revised: 22 January 2026 | Accepted: 24 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/mohamedamineferrag/edgeiiotset-cyber-security-dataset-of-iot-iiot, reference number [26]. Author Contribution Statement Mahavir Teraiya: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization. Madhu Shukla: Conceptualization, Methodology, Resources, Writing – review & editing, Supervision, Project administration.
Targeted weed management is an important element of precision agriculture, and accurate weed identification is a foundation for precision weed control. Over the past decade, convolutional neural networks have demonstrated high accuracy and generalization in recognizing weeds in agricultural environments. Edge computing, via edge devices, is one secure method to effectively deploy artificial intelligence algorithms (AI) for weed control on agricultural platforms. This study presents a bibliometric and systematic review of AI algorithms and edge-based systems for precision weed control from 2015 to 2025. Using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, a systematic search was conducted in Scopus and Web of Science, resulting in the inclusion of 43 documents. Results show a significant surge in publications on AI-based weed control systems deployed on edge computing resources since 2019. The analysis reveals that RGB cameras are the preferred data acquisition method, while object detection models, specifically the YOLO family, are widely adopted for AI deployment. Pretraining or training optimizations are preferred over post-training model optimization to maximize inference and improve detection accuracy, while NVIDIA Jetson series edge devices are commonly used for deploying AI algorithms. Precision chemical control approaches dominate, but laser weeding technology is emerging as an alternative. Despite technological advances, challenges hinder commercial deployment, including reliance on manual data annotation and model vulnerability to environmental variability. Future research should focus on semi-supervised learning, synthetic data generation, and multimodal vision systems to improve robustness and reduce annotation dependence. Unified evaluation protocols are also needed to benchmark system performance. Received: 22 December 2025 | Revised: 24 February 2026 | Accepted: 23 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Adeayo Adewumi: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Dharmendra Saraswat: Conceptualization, Methodology, Validation, Resources, Writing – review & editing, Supervision, Project administration, Funding acquisition.
Wrist fractures are one of the most frequent fractures in children that should be diagnosed properly and immediately to avoid any further complications. Regular radiographic analysis is time-consuming, relies on qualified radiologists, and is subject to human error. There has been a lot of buzz about the potential of deep learning technologies to automate medical image analysis in recent times. The algorithms in the YOLO series are sophisticated and refined, representing some of the best work in the field. The GRAZPEDWRI-DX pediatric wrist dataset is used in this study to address the variation in the YOLOv11 detection model’s inference time, recall, precision, and mean average precision (mAP). A medium-scaled variant of YOLOv5, YOLOv7, YOLOv8, YOLOv9, and YOLOv10 models were compared with YOLOv11-m. According to the experimental results, YOLOv11-m achieves the highest mAP@50–95 (0.569) for fracture detection, whereas YOLOv11-x attains the highest mAP@50–95 (0.424) across all category classes. An ablation study of YOLOv11 architectural elements was done to determine how the model optimized for medical imaging through the attention mechanism. On the other hand, the performance in fracture detection and inference time for YOLOv11 is superior to that of previous YOLO models, while it has lower computational complexity. Additionally, a performance comparison among recent YOLO-based models and the RT-DETR and Faster R-CNN models reveals variations in detection precision, recall, and accuracy, providing insight into the strengths and trade-offs of each approach. YOLOv11 could therefore make recurrent contributions and be of great value to clinical decision-making. Received: 18 August 2025 | Revised: 2 February 2026 | Accepted: 25 March 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available at https://doi.org/10.1038/s41597-022-01328-z, reference number [15]. Author Contribution Statement Muhanad Abdul Elah Alkhalisy: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Visualization. Qusay Shihab Hamad: Validation, Investigation, Writing – review & editing, Visualization, Supervision, Project administration. Ali Retha Hasoon Khayeat: Validation, Investigation, Writing – review & editing. Shahrel Azmin Suandi: Writing – review & editing, Supervision.
