Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder marked by persistent challenges in social interaction, communication, and behavioral patterns. The growing evidence shows that individuals with ASD have atypical brain connectivity, which makes electroencephalography (EEG) a valuable tool to study related neural activity. The non-invasive nature of EEG, its low cost, and its high temporal resolution have made it a focus for the early detection of ASD. However, most studies use an offline analysis, which limits their use in clinical practice. This paper introduces a real-time diagnostic platform that combines Internet of Things (IoT) technology and EEG to support the early identification of ASD using deep learning techniques.The framework integrates real-time EEG acquisition, signal preprocessing, deep learning-based classification models, and a Flask-based clinical interface. The clinical interface allows healthcare professionals to record EEG signals, perform real-time predictions, and generate diagnostic reports, whereas a secure guardian portal enables controlled access for patient monitoring. We evaluated time-series-based deep learning models, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), Gated Recurrent Units (GRU), Bidirectional Gated Recurrent Units (Bi-GRU), and hybrid architectures, using various data-splitting strategies and optimization algorithms. The CNN–BiGRU model trained with the Adam optimizer achieved the highest performance, with an F1-score of 96.40% and a Receiver Operating Characteristic–Area Under the Curve (ROC–AUC) of 99.37%. These results show that deep learning models can effectively identify EEG patterns associated with ASD and enable reliable classification. The web-based platform also supports real-time monitoring and decision support, increasing its value in clinical settings.
As Information Technology (IT) and Operational Technology (OT) networks converge within critical infrastructure, cybersecurity risk must be quantified under uncertainty and incomplete observability, conditions under which conventional, static scoring methods perform poorly. This paper presents the Multi-Stage Composite Fuzzy and Neuro-Adaptive (MCFNA) framework for quantifying operational cyber-risk in Industrial Control System (ICS) environments. MCFNA composes three interpretable Mamdani fuzzy inference stages that resolve, in turn, threat capability, attack likelihood, and overall operational risk, and augments them with an Adaptive Neuro-Fuzzy Inference System (ANFIS) layer that calibrates the fuzzy model to observed telemetry. The coupled learning procedure is proven to converge to a first-order stationary point. The framework is evaluated on the Secure Water Treatment (SWaT) benchmark under a leakage-controlled feature pipeline and an attack-scenario-wise train/test split that prevents dependent time windows from crossing the split boundary. Across regression and classification measures, the adaptive framework improves risk-estimation accuracy over linear-regression, neural-network, and ensemble baselines, and a sensitivity analysis identifies asset vulnerability as the dominant risk driver. These results indicate that MCFNA can provide interpretable, numerically stable, and actionable risk estimates to help prioritise Security Operations Center remediation in converged IT/OT infrastructure.
Early and reliable detection of strawberry foliar diseases is essential for reducing yield loss, limiting unnecessary pesticide use, and supporting timely site-specific crop management. This paper presents FRAISE-Net, a one-stage Strawberry Disease Inspection Network for automated localization and recognition of strawberry disease symptoms. The proposed architecture consists of an enhanced disease-aware backbone, a PAN-FPN-style neck, and a three-scale detection head. It integrates three complementary modules: Scale-Aware Channel Attention (SACA) to emphasize lesion-relevant channel responses, Context Aggregation Module (CAM) to capture broader disease context, and Adaptive Cross-Scale Fusion (ACSF) to combine fine spatial details with high-level semantic features across detection scales. Experiments were conducted on a 7-class strawberry disease detection dataset constructed from KUStraw and SDDD after removing the healthy class. The final experimental split contained 4766 annotated images, including 3810 images for training, 474 images for validation, and 482 images for testing. The FRAISE-Net algorithm was implemented on a real robotic arm for real-time disease detection in a controlled greenhouse laboratory setup. Under a unified 100-epoch training protocol with an input size of 320 × 320, FRAISE-Net was compared with YOLO-v9, YOLO-v10, YOLO-v11, YOLO-v12, YOLO-v26, and RT-DETR. On the held-out test set, FRAISE-Net achieved 95.61% precision, 87.88% recall, 91.58% F1-score, 94.48% mAP50, 77.58% mAP50:95, and 35.98 FPS on an RTX 2070 GPU. Five-fold cross-validation showed low fold-to-fold variability across partitions of the current combined dataset, with mean F1-score of 92.15±0.49%, mean mAP50 of 95.91±0.32%, and mean mAP50:95 of 80.09±0.73%. Ablation analysis confirmed the complementary contributions of SACA, CAM, and ACSF, with the complete FRAISE-Net achieving the strongest overall performance. The robotic experiments support the feasibility of the approach under the controlled, greenhouse-like laboratory conditions evaluated here; however, its generalizability to different sites, camera systems, seasons, disease distributions, and open-field settings remains to be established. These results indicate strong performance on the evaluated dataset, strong false-positive control, and practical inference speed. The code will be available at GitHub: FRAISE-Net.
Automated cleaning of floating garbage is a critical component of AI-empowered environmental monitoring for sustainable Smart Cities. Intelligent mobile sensing platforms, such as Unmanned Surface Vehicles (USVs), serve as key IoT nodes for aquatic ecosystem maintenance. However, existing detectors often misidentify natural debris as garbage under complex water conditions, a challenge known as visual mimicry. This study proposes ThinkRiver, a Coarse-to-Fine Semantic Verification Framework designed to enhance the sensing reliability of IoT devices by integrating a high-recall YOLOv8s proposer with a lightweight multimodal system, Qwen3-VL. By conducting semantic verification through Explicit Attribute Reasoning, the system enables more intelligent decision-making within the sensing network. Experiments on an Out-of-Distribution Challenge Set show that ThinkRiver reduces false positives from 261 to 82 (a 68.6% decrease) and improves Precision by 23.76%. These results indicate that semantic-reasoning-enhanced sensors can significantly strengthen the reliability of detection systems in dynamic and visually ambiguous environments, supporting the deployment of scalable AI-IoT solutions.
Autonomous and intelligent vehicles increasingly rely on perception pipelines that must operate reliably under uncertainty and adversarial manipulation. Traffic Sign Recognition (TSR) is a safety critical perception system because it translates structured road semantics that directly affect planning and control. Despite high accuracy in favorable settings, modern TSR systems that are driven by deep neural networks remain vulnerable to adversarial perturbations, including digital manipulations, physical patches, shadows, lighting effects, and sensor-level attacks that can cause unsafe misclassification. This survey combines adversarial threats and defenses for TSR through a cognitive computing lens. It treats TSR not only as an image classification problem but also as an intelligent perception task that requires understanding traffic sign meaning, handling uncertainty, using domain rules, and supporting safety-critical driving decisions. We provide a TSR specific taxonomy of adversarial attacks spanning attack medium (digital/physical), attacker objective, model access, and attack phase. We also consolidated a view of TSR datasets, recognition pipelines, and evaluation protocols and analysis of robustness metrics beyond attack success rate, including detection and system oriented measures used in recent literature. The survey further examines defense strategies across preprocessing, model training, inference-time detection, verification, and structure-aware reasoning, including approaches that exploit the geometric, color, and semantic regularities of traffic signs. Overall, the survey aims to support the development of trustworthy TSR systems for intelligent vehicles operating in complex, real-world traffic environments.
Industrial grasping requires accurate pose estimation and identity-aware selection, yet most deep-learning grasp detectors are object-agnostic and computationally heavy, so most existing approaches limit goal-directed manipulation and deployment on lightweight embedded systems. This paper presents a robot grasping system that combines RGB-based grasp detection and depth-based 3D localization with low-cost robot control. We use YOLOv11 with Oriented Bounding Boxes (YOLOv11-OBB) to simultaneously predict object pose and classification from RGB images. These detections are combined with depth data from an Intel RealSense D435 RGB-D camera to compute a 3D grasping pose. A 4-DOF robot arm controlled via a PLC performs pick-and-place operations based on the estimated poses. The paper evaluates two scenarios: a grasping-only model trained on a combination of Cornell and custom real-world datasets, and a grasping and classification model that allows for the selective manipulation of multiple object types. Experimental results show that the grasp-only model achieves 99.5% mAP@0.5, 94.0% mAP@0.5:0.95, and 99.4% precision at an IoU threshold of 0.6, while maintaining an inference time of 29 ms under the tested hardware setting. Compared with several representative grasp detection methods, the proposed approach achieves competitive accuracy and real-time performance. The grasp+classification model achieves over 97% grasp success across various object types with only 619 training images, indicating good performance under the tested experimental conditions despite the limited dataset size.
Communication is a fundamental human right. However, millions of deaf individuals worldwide continue to face significant barriers in their daily interactions due to the lack of accessible communication tools. This paper presents an IoT-enabled system, implemented as a wearable glove-based architecture, designed to improve communication between the deaf and hearing communities by translating Greek Sign Language (GSL) into text. The proposed system integrates wearable sensor technology with pattern-matching algorithms to accurately capture and interpret the Greek finger alphabet and gesture-based signs. Based on experimental evaluations, the system achieved an average recognition accuracy of 95.63%, while sensor faults were identified as the primary cause of misclassification when hand positions were altered. Through careful parameter tuning, such as threshold calibration, and algorithmic optimization for comparing sensor values against predefined thresholds, the translator demonstrated accurate and reliable performance in converting sign language gestures into written text, marking an important advancement in inclusive communication technologies. Despite notable progress in IoT-based wearable systems, a significant gap remains in achieving real-time, accurate, and language-specific translation of sign languages. Most existing approaches focus on generic gestures or widely used languages such as ASL, leaving Greek Sign Language (GSL) largely unexplored. Furthermore, limitations in expressivity, contextual understanding, and communication stability hinder real-world deployment. The proposed SpeakWithSigns+ system addresses these challenges by introducing an IoT-based wearable architecture that captures synchronized multisensor data, transmits it to a cloud server, and performs real-time pattern recognition for precise translation of GSL gestures. This development contributes to the advancement of inclusive, adaptive, and human-centered interaction technologies for the deaf and hard-of-hearing community.
Sign Language Recognition (SLR) enables communication between deaf and hearing individuals. Usually, SLR systems often depend on single-modality inputs, viz., sensor data, RGB videos, etc., which do not address issues such as signer variability and background, affecting the robustness and generalization of these systems. To address these challenges, we propose a novel framework, viz., Visual, Pose, and Motion-based Sign Language Recognition (ViPoMo-SLR), which incorporates RGB video frames, optical flow, and skeletal data—three complementary modalities. RGB frames are extracted to represent spatial features, while optical flow captures motion dynamics, and skeletal data provides hand and body joint positions. We select the crucial frames independently from both the RGB and skeleton streams to focus on the most informative segments. The extracted features are subsequently fused using a fusion technique and processed through a Bidirectional Long Short-Term Memory (BiLSTM) network to model temporal dynamics. Our empirical analysis shows that ViPoMo-SLR significantly outperforms unimodal methods, offering enhanced accuracy, real-time performance, and a scalable solution for robust SLR.
In vessel traffic management systems, an Automatic Identification System (AIS) is usually used to track vessel positions and control the flow. One disadvantage of AIS is that the ship’s crew can deactivate the transceiver installed on the ship. To support the AIS, other systems need to be implemented, including image-based vessel reidentification systems. In this paper, we propose a part pooling with a mixture of attention networks (PP-MAN) model for image-based vessel reidentification problems. The proposed model is formed by attaching mixtures of attention networks to a Swin Transformer V2 backbone with a part-pooling mechanism. Three different attention mechanisms were used to form the MAN module, including Multi-Head Attention, Gather Excite, and Global Context. To test the performance of our proposed model, we used three reidentification datasets, including Warship, VesselReID-1248, and ShipReID-2400. Experiments on those three datasets show that our proposed model achieves state-of-the-art performance across all datasets, with an mAP of 89.1% and rank-1 of 97.2% on the Warship dataset, an mAP of 61.3% and rank-1 of 72.3% on the VesselReID-1248 dataset, and an mAP of 42.7% and rank-1 of 50.3% on the ShipReID-2400 dataset. Further GradCAM and t-SNE analysis show that our proposed model extracted features from some vessel parts and can distinguish between vessel IDs.
Accurate object detection is fundamental to the safety and reliability of autonomous driving systems. Among anchor-free frameworks, FCOS has proven to be an effective baseline thanks to its simplicity and dense per-pixel prediction strategy. In FCOS, bounding-box regression estimates the offsets from each reference point to the four object sides, typically supervised by the Intersection over Union (IoU) loss, which jointly penalizes all four offsets. However, these offsets are not fully correlated, as each corresponds to a distinct region of the object. Moreover, IoU loss combines object size and shape into a single penalty term, making it difficult for the model to distinguish whether localization errors arise from scale mismatch or shape distortion. To address these limitations, we introduce Polar Loss, a novel IoU-free regression function that reformulates bounding-box prediction by estimating the positions of two diagonally opposite corners. This formulation explicitly decouples the learning of opposing offsets, allowing each side of the box to be optimized independently. Furthermore, Polar Loss expresses each corner in polar coordinates, where the radial distance represents object scale and the angular component captures shape variation. This enables the model to learn size and shape through distinct and interpretable terms. When integrated into the FCOS detector, Polar Loss improves accuracy across two autonomous-driving benchmarks, achieving +0.5% Average Precision on BDD100K and +1.1% on TJU-DHD without increasing inference cost. These gains reflect more accurate bounding-box regression, leading to enhanced perception reliability, which is particularly critical for safety in autonomous driving.
Stereo matching is a critical task for 3D perception in robotics and autonomous driving, offering a cost-effective and dense alternative to active sensors such as LiDAR. A key challenge lies in designing architectures that are both highly accurate and efficient enough for real-time deployment on resource-constrained hardware. In response to this challenge, we present EFSNet, a lightweight yet accurate stereo matching architecture that balances performance and efficiency. Building upon the efficient LightStereo framework, our architecture introduces three key contributions. First, we propose a hybrid 2D–3D aggregation module that synergistically combines the local disparity consistency captured by 3D convolutions with the global, all-to-all context modeled by 2D convolutions. Second, a dual-path refinement stage processes the fused features in parallel, with one path specialized for preserving fine details and another for adding robust estimation in low-texture areas. Finally, we introduce LinearLog loss, a training objective that effectively corrects both small and large errors by combining the properties of logarithmic and linear penalties. Our experiments show that EFSNet achieves a compelling trade-off between performance and speed. Our standard model achieves an EPE of 0.64 with an inference time under 20 ms on the SceneFlow dataset. Furthermore, our lightweight EFSNet-lite variant is the fastest among the compared methods, running at 16 ms while surpassing the accuracy of the previous fastest model.
The integration of 5G and AI technologies in the industrial internet supports collaborative production systems that orchestrate temporary compositions of industrial components. These compositions should be verified to ensure that their combined behavior satisfies expected functional requirements. However, the transfer of general large-language-model-based property-generation approaches to BPMN is non-trivial. BPMN models mix executable control-flow semantics with diagrammatic markup and encode business constraints through structural XML in an incomplete manner. They also require reasoning over gateways, loops, synchronization, and exception handling before temporal-logic properties can be formulated. To address these BPMN-specific difficulties, LLaMA-PG is proposed as an assistive framework for verification property generation from business workflows represented in BPMN. The framework comprises two task-oriented phases. In Phase 1, model-checking knowledge is injected into a base large language model through continued pre-training on a refined formal-verification corpus. This step enables the model to internalize the terminology and logical regularities required for BPMN verification. In Phase 2, the BPMN-to-property task is decomposed into four sequential steps, including attentional filtering, causal inference, property formulation, and symbolic translation. The model therefore generates properties through inspectable intermediate reasoning rather than direct black-box translation. In addition, a BPMN-oriented data preparation and quality-control pipeline is designed to reduce instances with structural inconsistency and semantic misalignment. Experimental results show that LLaMA-PG outperforms the compared methods in Syntactic Compliance Rate, Model-Checker Executability, Correctness, and Reference Coverage Recall, while maintaining a lower Redundancy Ratio in the generated property sets.
Service placement in fog computing has emerged as a fundamental research problem for enabling latency-sensitive and resource-constrained Internet of Things (IoT) applications. Although numerous optimization strategies have been proposed, existing studies remain fragmented, often addressing isolated objectives without systematically consolidating methodological trends, evaluation practices, and research gaps within a unified multi-objective perspective. This study presents a comprehensive Systematic Literature Review (SLR) that critically analyzes recent peer-reviewed research on service placement in fog computing, with a specific focus on multi-objective optimization approaches. Unlike prior surveys that emphasize architectural overviews or single-metric optimization, this review provides (i) a structured taxonomy of exact, heuristic, metaheuristic, game-theoretic, AI-driven, and hybrid approaches; (ii) a comparative analysis of optimization objectives including latency, energy consumption, cost, and resource utilization; and (iii) a synthesis of datasets, simulation tools, and validation frameworks used in empirical evaluation. The findings reveal a clear methodological shift from deterministic mathematical models toward adaptive and intelligence-driven frameworks integrating deep reinforcement learning, federated learning, and evolutionary optimization. Despite this progress, significant research gaps remain, particularly in integrating mobility-awareness, security and trust mechanisms, real-world validation, and lightweight AI models suitable for resource-constrained fog environments. By consolidating fragmented research streams into a unified analytical framework, this review provides a methodological foundation and research roadmap for developing intelligent, scalable, and empirically validated multi-objective service placement strategies in next-generation fog-based IoT systems.
The rapid growth of smart grids has led to large-scale deployment of smart meters that continuously report fine-grained electricity consumption data. While homomorphic encryption enables privacy-preserving aggregation, existing schemes typically assume static participation and rely on per-user signature verification, resulting in scalability bottlenecks and poor robustness when smart meters join or leave dynamically. To address these limitations, this paper proposes DUDA, a dynamic and secure data aggregation scheme that integrates Paillier homomorphic encryption with ciphertext re-randomization and SM2 batch signature verification. DUDA supports dynamic user participation without rekeying, provides unlinkability across reporting rounds, and significantly reduces verification cost at edge servers. Security analysis demonstrates confidentiality, integrity, authentication, and robustness against replay and forgery attacks. Experimental results show that DUDA achieves efficient communication and computation compared to related work. These findings confirm that DUDA is lightweight, scalable, and suitable for real-world smart-grid deployments with frequent user churn.
Medical diagnosis is one of the most challenging cognitive tasks in healthcare, with diagnostic errors affecting millions of patients annually. Large Language Models (LLMs) have recently shown real potential in supporting clinical reasoning, however pre-trained models remain susceptible to cognitive biases, such as premature diagnostic closure, confirmation bias, and limited consideration of alternative hypotheses. These limitations are particularly pronounced in smaller models with fewer than 10 billion parameters. Most existing solutions heavily rely on prompting strategies or simply scaling model size, which does not fundamentally strengthen the underlying reasoning process. In this paper, we investigate whether diagnostic reasoning can be meaningfully improved in small language models through targeted reinforcement learning (RL) post-training. We adopt a two-stage training strategy: cold-start phase using supervised fine-tuning, followed by reinforcement learning with Group Relative Policy Optimization (GRPO). We combine this with a low-rank adaptation (LoRA) approach, which updates only a small fraction of model parameters for computational efficiency. We evaluate our approach on a 3B-parameter model (Qwen2.5-3B-Instruct) and an 8B-parameter model (Qwen3-8B-Base) using unstructured clinical notes from the MIMIC-IV dataset. The results demonstrate substantial improvements in diagnostic accuracy from 47.40% to 63.79% for the 3B model and from 49.54% to 66.56% for the 8B model, achieved with just 2554 training samples at a compute cost of approximately $3. These findings highlight that reinforcement learning-based post-training, combined with LoRA, offers a practical and cost-effective path to transforming small language models into capable diagnostic reasoners.
The Web of Things (WoT) interconnects a rapidly growing population of smart devices and sensors, enabling innovative applications while exposing an ever‑expanding attack surface. Reinforcement learning (RL) can adaptively detect and mitigate such attacks, yet conventional RL struggles to converge in WoT’s high‑dimensional state‑action spaces. We address this limitation by augmenting RL with the Harris Hawks Optimization (HHO) algorithm. HHO is a recent meta‑heuristic optimization method that balances global exploration with local exploitation and is well suited to large search spaces. We propose an HHO‑based meta‑learning framework that aims to identify hyper‑parameters and network architecture for a deep‑Q network (DQN) defender, maximizing average episodic reward in simulated WoT environments. Experiments on the CIC‑IoT‑2023 and Bot‑IoT datasets show that an HHO‑optimized DQN converges faster and achieves higher accuracy than all tested baselines—including vanilla, double and dueling DQNs, PPO, A3C and Transformer-based agents—illustrating the promise of bio-inspired optimization for adaptive WoT security.
This study proposes a fast, hybrid image cryptosystem that pairs classical encryption with transform-domain processing for securing multimedia and cognitive-computing use cases. The pipeline combines the 2D nonuniform FFT (2D-NUFFT) for nonuniform spectral scrambling, 2D discrete wavelet transform (2D-DWT) for multiresolution decomposition, chaotic random phase mask (CRPM) for nonlinear diffusion, and triple DES (TDES) for block-cipher protection. The results of simulations conducted show lossless reconstruction (MSE ≈ 10−24; PSNR = ∞, SSIM = 1.000) and strong security: NPCR = 99.65 %, UACI = 33.52 %, entropy = 7.9963, near-zero correlation (0.0009), and a near-ideal avalanche (50.06 %). The overall computational cost is O(N2logN) (with the NUFFT as the main contributor), and a 512 × 512 frame completes in approximately 172 ms on a CPU-only setup. Taken together, these results indicate that the proposed framework provides a reliable, lightweight solution for real-time, high-security applications while maintaining exact decryption and robust resistance to common cryptanalytic attacks.
In Mobile Edge Computing (MEC), task scheduling is used to overcome the problems related to resource constraints in mobile devices. However, the conventional task allocation approaches failed to consider both the energy consumption and security requirements, that leads to higher energy consumption, increased latency and potential vulnerabilities. To overcome these issues, this research proposes a secure and energy-aware task scheduling technique by considering energy and security needs at host level while allocating tasks to Virtual Machines (VMs). The evaluation of energy cost at task allocation helps to enhance the energy efficiency. On the contrary, security is improved via the combination of authentication, confidentiality, and integrity mechanism that ensures secure task execution while maintaining low energy consumption. Thus, the proposed model i.e., multi-tier MEC architecture performs secure resource allocation and reduced energy utilization through effective VM selection and optimized scheduling. The results are evaluated based on energy consumption, average delay, and make span time. For instance, when the result is evaluated by considering energy consumption, the proposed approach consumed 7.3 J of energy when time slot is assigned as 10, while the existing Improved NSGA-II consumed higher energy of 8.2 J in the similar time slot.
Advancements in unmanned aerial systems (UAS) and artificial intelligence (AI) have emerged in recent years, which have accelerated research in a variety of fields, including human-drone interaction, autonomous navigation, security, object detection, urban air mobility, energy-efficient design, environmental monitoring, archaeological research, wildlife conservation, medical supply delivery, disaster response, and precision agriculture. These represent just a few areas where AI-driven drones are making an impact in society. The study discussed the main principles of autonomous navigation and its mathematical aspects and presented the state-of-the-art algorithms in unmanned aerial vehicles (UAVs). The study also demonstrated an optimization model for path loss considering terrain and environmental constraints and provided a comparative analysis of the state-of-the-art models. This study contributes to the ever-expanding field of drone technology by addressing critical research gaps and introducing novel approaches, but also offers practical recommendations for professionals and policymakers. The findings of this study pave the way for the integration of UAS into various applications, fostering the adoption of innovative solutions. Additionally, this research has the potential to serve as a foundational reference for future studies, encouraging collaboration and driving advancements in drone technology.
The academic community places significant emphasis on publishing research in SCI and SSCI journals, which are known for their credibility, high quality, and strong reputations. Most research requires significant investment in human resources and equipment, making the acquisition of research funding crucial. Understanding the motivation behind authorship and its association with citation patterns in SCI and SSCI journals represents a significant research concern in bibliometric studies. This study identifies the relationship among author number, research funding, and citation count using content analysis techniques, including the chi-square and analysis of variance tests. Investment in AI research, development, and applications is increasing; thus, this study examines 4488 articles published in the field of artificial intelligence (AI) from Springer in 2018. The empirical results indicate that (1) the average number of authors is highest in Q1 journals, with non-single-author papers being more common than single-author papers and concentrated in higher rankings; (2) papers with research funding are more common than those without; (3) papers with citations are more frequent than those without; (4) the ranking of papers with research funding and citations is higher than that of other papers without funding; and (5) the average citation count of papers with research funding leads in Q1 and is higher than in other rankings. This study is the first attempt at highlighting papers in the field of AI from Springer. The results and important findings provide useful references for researchers, reviewers, publishers, and interested parties with different purposes for academic and technical publications with sustained success. This study uniquely integrates four dimensions—author count, research funding, journal ranking, and citation count—to offer novel insights into academic publishing performance in the AI field.
