Keywords: Agent-Based Modeling, AI and Applications, Application of Artificial Intelligence
Abstract: Cyber-physical multi-agent systems (CPMAS) integrate various autonomous systems, and even human participants, through both cyber-layer communication and physical-layer collaboration, enabling them to operate collectively in complex, open environments. Traditionally, these systems have relied on model-based control methods to ensure robust coordination and reliable performance. However, recent advances in artificial intelligence (AI) offer promising opportunities to enhance autonomy, adaptability, and scalability when faced with various constraints at both the cyber and physical layers—such as malicious cyberattacks and internal system faults. This talk illustrates the technological evolution from model-based approaches to AI-empowered frameworks for CPMAS, highlighting how data-driven techniques—such as reinforcement learning, deep neural networks, and bio-inspired AI—can address limitations in traditional design and control frame. We will explore the challenges of seamlessly integrating model-driven and AI-driven strategies, including stability, safety, and security. By combining advanced control methodologies with emerging AI capabilities, we present a series of experimental case studies that verify our proposed techniques with potential real-world applications in intelligent manufacturing, autonomous vehicles, and smart infrastructure. Ultimately, this talk provides insights into the future of CPMAS, setting the stage for innovative research and practical deployments.

Keywords: Deep Learning, AI and Applications
Abstract: Learned Perceptual Image Patch Similarity (LPIPS) is an image similarity metric that is typically used to evaluate the effectiveness of reconstruction attacks in privacy leakage of federated learning (FL). Our study shows that such image similarity metric is insufficient to assess the degree of privacy leakage. To tackle this issue, a new evaluation method capable of revealing preserved privacy in privacy-preserving FL (PPFL) is explored in this paper. Through extensive empirical experiments, which employ a differential privacy (DP)-based PPFL (DP-PPFL) and a state-of-the-art image reconstruction adversary (at the time of writing this paper), high and stable perceptual metric values were observed across all ranges of a privacy budget in DP. These LPIPS values suggest the privacy of images can be preserved well in terms of perceptual quality preservation, regardless of how the DP-PPFL is configured. However, there are multiple discrepancies between the classification accuracy of original and reconstructed images. This finding suggests that there is indeed a privacy leakage that the LPIPS failed to capture; thus, LPIPS should not be used as a standard for evaluating privacy leakage. Alongside the discovery, this paper also proposes a so-called ``quadrant chart,'' to quantify the assessment of privacy leakage in PPFL.

Keywords: Deep Learning, AI and Applications
Abstract: Time-series anomaly detection is essential for ensuring the reliability of large-scale data stream applications. However, existing deep learning approaches primarily rely on reconstruction-based methods, making them highly sensitive to the quality of training data. To mitigate this limitation, we introduce an anomaly injection strategy to enhance model robustness under unsupervised settings. Additionally, most deep learning models are inherently constrained by their structural limitations, leading to weaknesses in capturing specific temporal characteristics. To address this, we propose PRAD-Net (Periodic Reorganization Anomaly Detection Network), which incorporates a Periodic Reorganization (PR) module to decompose time series into long-period and short-period features. These features are then processed separately by two specialized modules: the Long-Period Temporal Attention (LPTA) Module, which enhances long-term dependency modeling, and the Multi-Scale Focal (MSF) Module, which focuses on local temporal variations. By strategically assigning periodic features to appropriate modules, PRAD-Net effectively exploits their complementary strengths while mitigating their respective weaknesses. Extensive experiments on five real-world datasets demonstrate that PRAD-Net significantly outperforms state-of-the-art baselines, highlighting its strong capability for unsupervised anomaly detection.

Keywords: Deep Learning, AI and Applications, Application of Artificial Intelligence
Abstract: Large language models (LLMs) have demonstrated the ability to perform in-context learning (ICL) with only a few demostrations, achieving remarkable performance across various downstream tasks. The selection of demonstrations plays a critical role in shaping the performance of ICL due to its high sensitivity. However, previous researchers have primarily focused on either the tricks for selecting demonstrations or the sequencing of demonstrations, thereby neglecting the critical role of retrieval models in ICL. The rigid structures and parameters employed in these studies often fail to align with the specific requirements of downstream tasks. To address this problem, we propose CLUG: contrastive learning unified retrieval with graph-ranked demonstrations for enhanced in-context learning, integrating demonstrations retrieval selection and demonstrations order to establish semantically coherent sequences of demonstrations, thereby ensuring enhanced semantic alignment and consistency. Experimental results across multiple datasets demonstrate consistent improvements with our method. Additional analyses further validate and explain the effectiveness and generalizability of our approach.

Keywords: Deep Learning, AI and Applications, Expert and Knowledge-Based Systems
Abstract: Language models have been proven to handle time series data after tokenization and show generalization performance on unseen forecasting tasks. However, existing techniques for tokenizing time series data struggle to eliminate redundant information and noise, which can lead to signal aliasing and cumulative quantization errors, making it difficult to further improve prediction performance. In this paper, we propose an Adaptive Component Extraction and Tokenization (ACET) framework, which includes two key novelties to address these challenges: the Dynamic Component Extraction Module (DCEM) and the Time Series Tokenization Module (TSTM). The DCEM dynamically isolates the principal components from the interference in the original signal, eliminating the need for manual parameter tuning. This not only enhances the accuracy of signal tokenization but also mitigates the adverse effects of high-frequency noise. Then, the TSTM tokenizes continuous time series data while preserving long-term trend features, ensuring that critical information is retained for subsequent forecasting. Extensive cross-domain experiments on various real-world datasets demonstrate that, in zero-shot forecasting scenarios, ACET achieves improvements of 19.27% in WQL and 6.63% in MASE, compared to baseline methods.

Keywords: Deep Learning, AI and Applications, Machine Vision
Abstract: As an open research topic in the field of deep learning, learning with noisy labels has attracted much attention and grown rapidly over the past ten years. Learning with label noise is crucial for driver distraction behavior recognition as real-world video data often contains mislabeled samples, impacting model reliability and performance. However, label noise learning is barely explored in the driver activity recognition field. In this paper, we propose the first label noise learning approach for the driver activity recognition task. Based on the cluster assumption, we initially enable the model to learn clustering-friendly low-dimensional representations from given videos and assign the resultant embeddings into clusters. We subsequently perform co-refinement within each cluster to smooth the classifier outputs. Furthermore, we propose a flexible sample selection strategy that combines two selection criteria without relying on any hyper-parameters to filter clean samples from the training dataset. We also incorporate a self-adaptive parameter into the sample selection process to enforce balancing across classes. A comprehensive variety of experiments on the public Drive&Act dataset for all granularity levels demonstrates the superior performance of our method in comparison with other label-denoising methods derived from the image classification field. The source code is avialable at https://github.com/ilonafan/DAR-noisy-labels.

Keywords: Image Processing and Pattern Recognition, Machine Learning, Deep Learning
Abstract: Human Action Recognition is a fundamental task in computer vision, yet remains challenging due to factors such as background clutter, occlusion, illumination changes,and viewpoint variation. Existing skeleton-based methods often rely on joint coordinates or depth data, while neglecting the geometric relationships between body parts. In this paper, we propose LieGNN, a geometry-aware framework that integrates Lie group representations with spatio-temporal graph-based learning for robust action recognition. The framework comprises three main modules: a high-precision 3D pose estimation module that combines multi-resolution feature fusion with camera-aware depth modeling; a Lie group trajectory modeling module, which encodes pose sequences as curves on the Lie manifold and computes alignment errors via class-specific standard curves; and a graph-based classification module that incorporates both pose features and alignment errors to enhance discriminative learning. Extensive experiments on three public datasets demonstrate that LieGNN outperforms several methods and exhibits strong generalization across diverse action categories. These results highlight the effectiveness of incorporating geometric priors into graph-based motion modeling.

Keywords: Machine Vision, Deep Learning
Abstract: The proliferation of 3D colored mesh models lead to growing scholarly interest in assessing their visual quality. Existing 3D mesh quality assessment methods primarily rely on single-modal features derived from either 2D projections or the three-dimensional structure of the model. 2D projections contain rich semantic and texture information, but they cannot show enough quality loss caused by structured distortion. However, the 3D structure of the model can accurately reflect geometric distortions but does not fully utilize color information, making it difficult to comprehensively represent the perception of human visual system of complex distortions. Therefore, we propose a multi-modal learning for no-reference 3D colored mesh models quality assessment method (MM-MQA), First, we obtain projection images of the 3D colored mesh model from different viewpoints and split the model into equally sized patches. We then employ MeshNet and ResNet to encode the structural features of the model patches and the texture features of the projection images. Finally, double cross-attention is employed to achieve multi-modal fusion to perceive the overall quality of the mesh model. Experiments demonstrate that our method outperforms existing approaches in both CMDM and TMQ datasets, validating the effectiveness of multi-modal learning in mesh quality assessment tasks.

Keywords: Application of Artificial Intelligence, Deep Learning, AI and Applications
Abstract: In the context of an increasingly complex and congested space debris environment, the development of high-precision long-term orbit prediction models has become a core technology for space situational awareness and space traffic management. Traditional orbit prediction methods often struggle to achieve an ideal balance between accuracy and computational efficiency. While deep learning-based orbit prediction approaches have demonstrated promising potential, they commonly face challenges related to high model complexity. This paper analyzes the long-term orbital prediction problem from the perspective of time series forecasting. Specifically, a dataset framework is first constructed, based on real high-precision orbital data, which incorporates temporal periodic features. Then, a novel model combining the Kolmogorov-Arnold Network (KAN) structure with a Gated Recurrent Unit (GRU) is proposed, named KiGRU, with the goal of enhancing the performance of orbital prediction. Experimental results show that the proposed KiGRU model outperforms existing methods in terms of prediction accuracy and achieves a good balance between model complexity and performance, making it more suitable for practical applications. Furthermore, this study demonstrates the effectiveness of integrating the KAN structure into traditional deep learning models to enhance their performance.

Keywords: Application of Artificial Intelligence
Abstract: Software testing is a critical component in software development that is closely related to software quality. Traditional test generation methods face challenges such as producing test cases that are difficult to read and maintain synchronously. Meanwhile, with the advancement of large language models (LLMs) in code generation, the quality of LLM-generated code is increasingly comparable to human-written code. Therefore, this paper proposes a test code generation framework using Model-driven Multiple Results Filtering and Multi-Round Generation strategy (Model-MRFaG). To better adapt to test generation tasks for mainstream programming languages, we built an Alpaca-format Test-Code DataSet for Finetuning Baseline Models (TCDSF) containing six programming languages: Python, Java, JavaScript, C++, C#, and Go, and used this dataset to fine-tune a Baseline Model to obtain the TestCoder model. Subsequently, we developed the Model-MRFaG framework based on the TestCoder model to further improve the accuracy of test code generation. Through comparative experiments evaluating both general test sets and test code generation capabilities, TestCoder outperforms the Baseline Model in both general test sets and test code generation accuracy. Furthermore, the Model-MRFaG framework can further improve the accuracy of test code generation, providing a new solution approach for the intelligent development of software testing.

Keywords: Application of Artificial Intelligence, AI and Applications
Abstract: 支持人工智能 （AI） 的推荐系统 越来越多地用于电子商务的推广 特定产品和服务。此类系统如何影响 消费者自我认同的行为 然而，目前尚不清楚。本研究探讨了 AI 身份对消费者采用行为的影响 使用 AI 身份的支持 AI 的推荐系统 理论和行为推理理论。相关的 研究了文献，从而得出了 用于理解 这种情况。然后可以测试这样的模型，并且 使用实证数据进行验证。本研究有助于 消费者行为研究的 用于更好地探索 AI 影响的概念模型 使用 AI 时消费者行为的标识 电子商务中的推荐系统。

Keywords: Application of Artificial Intelligence, AI and Applications
Abstract: 在本文中，我们提出了一个知识引导样本 选择提示生成 （KGSS） 方法以减少 通过关注逻辑相关性来模拟幻觉 知识要点。我们认为，特定于任务的上下文 应该从知识关键点中得出，而不是 仅依赖语义相似性，特别是在 高考等复杂场景。KGSS系统 首先构建知识关键点的知识图谱 在主题内。然后，它采用两阶段样本 整合知识关键点的选型策略 过滤和知识级特征提取。 最初，系统会过滤掉不相关的知识密钥 根据输入的相关性进行点，确保 候选样本在上下文中是适当的。然后，一个 进行多级样品选择过程以平衡 知识关键点的相似性和多样性。 在Gaokao-Bench数据集上的实验结果，推导 从中国高考题，展示 KGSS 在以下方面明显优于现有方法 提高各种模型的准

Keywords: Application of Artificial Intelligence, Machine Learning, Representation Learning
Abstract: In recent years, single-label intent recognition has faced significant challenges in handling short and semantically sparse user inputs. Traditional methods often treat intent labels as flat categories, neglecting their inherent hierarchical relationships and limiting model performance. To address these issues, we propose HiIntent , a novel zero-shot intent detection framework that integrates hierarchical semantic modeling with a collaborative generation-discriminative mechanism. HiIntent first constructs a structured label hierarchy through a two-stage process: a large language model (LLM) generates semantic abstractions of intent labels, which are then evaluated and refined by a discriminative module to ensure coherence and correctness. This is followed by a similarity-driven convergence strategy that enhances intra-class consistency and inter-class separability using multi-metric similarity calculations. Finally, a contrastive prompt construction method leverages the learned label hierarchy to generate enriched semantic descriptions for each intent, improving representation learning and facilitating accurate classification even in zero-shot scenarios. Extensive experiments on both general-purpose (CLINC-150) and domain-specific (RFMR) datasets demonstrate that HiIntent consistently outperforms existing approaches across multiple evaluation metrics. Ablation studies and hyperparameter analyses further validate the effectiveness of each component in the proposed framework.

Keywords: Application of Artificial Intelligence, AI and Applications, Deep Learning
Abstract: Unmanned aerial vehicles (UAVs) are increasingly relied upon in a variety of critical applications, including infrastructure inspection, search-and-rescue, and traffic monitoring. While modern UAVs are typically equipped with global positioning system (GPS), inertial measurement unit (IMU) modules, and often include safeguards against adverse environmental conditions, they remain susceptible to sensor malfunctions and signal disruptions. These challenges have led to the need for robust, GPS-free solutions capable of maintaining accurate trajectory prediction and state identification. This work proposes a real-time multi-task learning framework for UAVs that employs Transformer-based neural networks to perform simultaneous trajectory prediction and state identification. The proposed system enables GPS-free UAV operations with the employment of a feature-based visual odometry algorithm that is implemented and fused with telemetry data to achieve accurate localization. The proposed system is implemented (hardware and software modules) in a functional prototype and validated through extensive outdoor experiments using a custom-built dataset, demonstrating strong performance and improved prediction accuracy in GPS-denied environments. The results demonstrate the framework’s ability to enhance the autonomy and reliability of UAV systems in challenging operational scenarios.

Keywords: Application of Artificial Intelligence, AI and Applications, Deep Learning
Abstract: The Segment Anything Model 2 (SAM 2) demonstrates strong capabilities for video object segmentation (VOS) . We present a SAM 2-based framework for self-checkout product counting, where box prompts are generated by a detector selecting optimal tracking initiation frames. Our key contribution is Selective-SAM, a training-free enhancement that improves robustness in complex scenes via a selective memory bank. This mechanism selectively preserves high-quality and diverse features while filtering poor segmentation priors. For final counting, we introduce a novel Mask Overlap Degree metric to analyze object trajectories. By segmenting trajectories based on the mask overlap degrees, we accurately determine the product count. Experiments across various retail scenarios show improvements of 1.95% IDF1 and 3.86% MOTA.

Keywords: Big Data Computing,, Knowledge Acquisition, Deep Learning
Abstract: Transformer-based sequential attention models have achieved excellent results in capturing dynamic changes in user interest. However, many studies have shown that these models lead to a convergence of the semantic representations of item embeddings to be the same, which leads to a loss of differentiated information and ultimately the accuracy of the model cannot be further improved. In order to tackle that, we propose a novel method called Split-Band Mixture of Experts with Implicit Distillation for Sequential Recommendation (SMoERec). The model fuses features from both the time and frequency domains for sequential recommendation, specifically, we use the self-attention mechanism to extract the time domain information, and use mixture of experts (MoE) to fine-grain features in different frequency bands in the frequency domain, and different experts focus on specific frequency bands to effectively improve the quality of features in the frequency domain. Finally, the features in the two spatial domains are adaptively fused using implicit distillation. We validate the effectiveness of our approach compared to baseline methods on five real datasets.

Keywords: Application of Artificial Intelligence, AI and Applications, Machine Learning
Abstract: This paper presents a machine learning-based approach for classifying the operational states in the biogas Dry Reforming (DR) reactor, focusing on catalyst activation, reaction, and irregularity detection. A key challenge in DR processes is the formation of coke, which can lead to reactor clogging. To address this, we propose incorporating the virtual drifts number, which are changes in the input data distribution, as an additional feature to enhance model performance. Different drift detection algorithms and classifiers were evaluated on a dataset comprising nine distinct DR reactions. Experimental results demonstrate that integrating virtual drift counts improves the average accuracy from 84.74% to 88.01% and the average F1 scores from 81.25% to 85.39% across all models, with RF achieving the highest performance (accuracy from 88.40% to 92.35% and F1 score from 85.95% to 91.59%). Our results highlight the potential of drift-aware features for real-time monitoring and fault detection in DR systems, offering a scalable solution to optimize reactor operations.

Keywords: Distributed Intelligent Systems
Abstract: In the Internet of Things (IoT) scenarios, edge computing assists in completing machine learning on resource-constrained terminal devices. As one of the most significant operations, large-scale matrix multiplication remains a huge efficiency bottleneck. Existing distributed computation approaches typically decompose the matrix computation into subtasks of the same scale, overlooking edge computing environments with unbalanced computing resources. In this paper, we propose a secure distributed matrix multiplication outsourcing scheme in edge computing with unbalanced resources. Specifically, the large-scale matrix multiplication is decomposed into several subtasks of varying scales according to unbalanced edge computing resources, which achieves better distributed computational performance. A novel matrix blinding method is presented by employing perturbation matrices and permutation matrices to guarantee input and output privacy. Experimental results demonstrate that our scheme improves the efficiency by 51.13% to 98.79% compared to traditional matrix multiplication without outsourcing. Additionally, our scheme outperforms state-of-the-art outsourcing schemes, with an average improvement of 9.36% in edge computing with balanced resources and 14.83% in unbalanced environments.

Keywords: Distributed Intelligent Systems
Abstract: The non-independent and identically distributed (non-IID) data distribution in federated learning (FL) poses significant challenges to the training performance of FL models. To tackle this problem, we propose an novel framework called Federated Learning with Manifold Mixup of Prototype (FedMP) which takes advantage of both prototype learning and manifold mixup. Prototype learning ensures the effectiveness of the training results but also, in conjunction with manifold mixup, further enhances the generalization ability of the FL model. We also design a mixup matrix to regulate the training process and propose a strategy to redistribute the matrix weights. Additionally, to enhance the reliability of prototypes in the latent space, we introduce a conditional invertible generative network (cINN) to reconstruct the obtained prototypes. These reconstructed prototypes, in turn, participates in the training process to improve the model performance. We conduct experiments on two real-word datasets for the validation of the proposed FedMP framework. Experimental results show that FedMP outperforms other baselines on both training performance and communication costs.

Keywords: Distributed Intelligent Systems
Abstract: In multi-user edge intelligence system, achieving efficient and stable DNN inference under heterogeneous task priorities is a critical challenge. This paper presents a priority-aware edge-device collaborative inference scheme that models the entire inference workflow while explicitly incorporating user-level priority constraints. The optimization objective is twofold: to maximize, in priority order, the number of users with stable local queue under the strict constraint of server queue stability, and to minimize the total end-to-end latency across all users. A key component is the Server Queuing Latency Estimation (SQLE) algorithm, which decomposes the latency contributions of user priority interactions and iteratively estimates task queuing times. Compared with classical models such as M/D/1, SQLE achieves significantly higher accuracy under dynamic workloads. Based on the predicted latency, we further develop a two-stage offloading decision algorithm: Maximum User Prioritized Selection for Local Queue Stability (MUPS) determines the maximal subset of users whose local queues can be stabilized, and Fine-grained Offloading Point for Latency Optimization (FOPL) refines offloading points to minimize global latency. Experiments on a heterogeneous edge-device system show that SQLE consistently achieves over 90% estimation accuracy with low error variance across varying system scales, significantly outperforming classical queuing models. Under different load, MUPS and FOPL supports more stable users and reduces average end-to-end latency, demonstrating its robustness and superiority over state-of-the-art methods.

Keywords: Distributed Intelligent Systems, Adaptive Systems
Abstract: We proposes Attention Gumbel-Softmax Multi-Agent Deep Deterministic Policy Gradient (AGS-MADDPG), a multi-agent reinforcement learning algorithm for collaborative pathfinding under partial observability. Traditional methods face challenges including non-differentiable policy gradients and inefficient exploration in discrete action spaces. We address these by integrating the MADDPG framework with Gumbel-Softmax for differentiable policy optimization and adaptive exploration via temperature annealing. A multi-head attention mechanism in the critic network explicitly models inter-agent dependencies by dynamically weighting teammates’ action features, enhancing Q-value estimation precision and training stability. Evaluations in Pogema environments with random obstacles and complex layouts demonstrate improved convergence speed, pathfinding success rates, and average path length compared to baseline methods. The results validate the effectiveness of combining differentiable discrete policies with explicit attention-based coordination for multi-agent collaboration in partially observable settings.

Keywords: Distributed Intelligent Systems, Service Systems and Organizations
Abstract: Cross machine traffic caused by distributed microservice deployment in edge computing significantly affects service performance. And the dynamics of the edge computing, such as fluctuating user request patterns and different network delay make static scheduling stategies challenging. To address these two issue, we first propose a Cross-Machine Traffic-Aware Scheduling Algorithm (CTSA), which models the microservice deployment process as a Markov Decision Process and utilizes a Dueling DQN-based approach to minimize cross-machine traffic while balancing node resource usage. Furthermore, we propose a Dynamic QoS-Aware Scheduling Framework (DQSF) that adapts deployment decisions in real time based on system monitoring to address the challenge of dynamics. Experimental evaluations using a real-world microservice application show that our approach significantly reduces request response time up to 30.2%, improves throughput to 36.7% and ensure Quality of Service (QoS) under dynamic edge computing continuum.

Keywords: Distributed Intelligent Systems, System Modeling and Control
Abstract: Driven by the rapid progress of digital technology, manufacturing cells of the smart shop floor have achieved the ability to collect data, compute, reason, and make autonomous decisions. By utilizing such cell intelligence, production schedules can be flexibly adjusted and production performance can be improved. Against this backdrop, this study proposes a novel distributed scheduling method based on data-augmented QMIX for smart shop floors to fully exploit the potential of cell intelligence. In the proposed method, QMIX is adopted to construct the scheduling agents for manufacturing cells and collaboratively optimize the decisions of agents. Furthermore, a data augmentation mechanism based on denoising diffusion probabilistic model is developed to improve the training efficiency of the QMIX algorithm. Finally, experiments conducted on the semiconductor production shop floor MiniFab verified the performance of the proposed method.

Keywords: Distributed Intelligent Systems, System Modeling and Control, Quality and Reliability Engineering
Abstract: 信息技术的发展，分布式系统越来越受到重视。数据 一致性已成为分布式 系统，因为它直接影响系统的可靠性和 用户体验。数据不一致可能源于 事务一致性、实时一致性或人工 篡改数据。本文的主要重点是 验证和检测数据一致性的可靠性 分布式系统。为了解决这个问题，我们建议 时间 Petri 网模型表征数据变化 系统。为了更好地观察系统行为，我们提出了一个 将时间Petri网转换为定时的方法 胞 自动 机。该转换方法用于验证和 检测分布式系统中的数据一致性。最后，到 证明我们方法的可靠性，我们使用 以购物系统为例，进行验证 在 UPPAAL 工具的支持下进行测试，提供 对系统设计和实施的有力支持。

Keywords: Cyber-physical systems, Distributed Intelligent Systems, System Modeling and Control
Abstract: We propose BST-ID, a variable-length binary spatiotemporal identifier tailored for hierarchical and bandwidth-constrained systems such as LoRa-based drone networks. Unlike traditional spatial indices that treat time as an auxiliary attribute, BST-ID encodes four-dimensional coordinates (x, y, h, t) with independently adjustable resolution, enabling compact and semantically adaptive representation. The identifier includes a binary header and bit-encoded coordinates, enabling prefix filtering and low-overhead communication with anisotropic control. The format is suitable for edge devices operating in multi-hop LPWAN environments. We implemented BST-ID and evaluated it against GeoHash and Google S2, demonstrating compactness (13 bytes average), encoding speed, and flexible prefix queries. Furthermore, we discuss its potential for decentralized data exchange, including blockchain integration and Lightning Network micropayments, where each ID acts as a hash-stable, time-scoped transaction reference. BST-ID thus offers a scalable, query-efficient, and token-aware foundation for distributed sensing, routing, and autonomous information sharing.

Keywords: Autonomous Vehicle, Trust in Autonomous Systems, Decision Support Systems
Abstract: Semantic segmentation is a crucial task in autonomous driving, enabling pixel-level understanding of complex road environments. DeepLabv3+ has emerged as a robust architecture for semantic segmentation, owing to its powerful Atrous Spatial Pyramid Pooling (ASPP) and decoder modules. However, the performance of DeepLabv3+ heavily depends on the choice of backbone network, which influences feature extraction, segmentation accuracy, computational efficiency, and inference speed. This paper investigates the impact of various backbones, including ResNet, Xception, Inception, and MobileNet, on the performance of DeepLabv3+ for semantic segmentation of driving scenes using the CamVid dataset. The study systematically compares these backbones, addressing their trade-offs between real-time applicability and segmenta- tion precision. Results reveal that InceptionResNetV2 achieves the highest performance with an accuracy of (90.83%) and mIoU of (83.58%) but at the cost of higher computational demands, while MobileNetV2 offers a lightweight alternative with competitive performance (mIoU 81.33%). The analysis highlights the strengths and limitations of each backbone, offering valuable insights into optimizing segmentation models for autonomous driving scenarios. Future research should explore hybrid backbones to balance accuracy, efficiency, and explainability for diverse driving environments.

Keywords: Robotic Systems, Autonomous Vehicle, Conflict Resolution
Abstract: This paper presents the design, implementation, and evaluation of a vision-based Detect and Avoid (DAA) system for small uncrewed aircraft systems (UAS) using monocular vision and deep learning. The proposed system integrates YOLOv8 for object detection, BoT-SORT for tracking, and a rule-based avoidance strategy based on the right-hand traffic convention. The system is deployed on RyzeTech Tello EDU drones with offboard image processing and tested in indoor flight scenarios across multiple approach speeds. Experimental results demonstrate a detection accuracy of 91.7% and an 80% success rate for avoidance maneuvers in non-cooperative environments. Detection and avoidance timing varied by speed, with greater delays at lower speeds and more consistent performance at higher speeds. The system successfully tracked intruders between 6.55 and 9.25 meters, confirming its effectiveness in short-range encounters. Limitations include a narrow field of view, latency due to ground-based processing, and reduced performance in lateral approaches. Future improvements will target fisheye-compatible detection, onboard inference acceleration, and adaptive avoidance logic. These findings support the feasibility of lightweight, vision-only DAA systems for infrastructure-less, low-altitude airspace.

Keywords: Cooperative Systems and Control, Decision Support Systems, Service Systems and Organizations
Abstract: The unmanned aerial vehicle (UAV) resupply mode offers a viable alternative to parcel delivery by replenishing ground vehicles at intermediate open-space supply points. Usually, orders are treated as equally important with no guarantees of on-time delivery. In this paper, we propose a priority based delivery framework to efficiently manage real-time delivery demands with varying priorities. Specifically, this framework integrates a fleet of UAVs for resupply and trucks for final delivery. By considering the inherent priority of orders and their elapsed waiting time, a dynamic priority discipline is introduced and customized to different order contexts. This approach ensures that urgent orders are prioritized to ensure timely fulfillment within their deadlines, while low-prioritized orders can still be completed without excessive waiting. Simulations using a real-world city map of Helsinki, along with mobility features of both trucks and UAVs, show significant performance gains over baseline schemes in terms of on-time delivery rate, average delivery time, and waiting fairness among all orders.

Keywords: System Modeling and Control, Decision Support Systems, Modeling of Autonomous Systems
Abstract: Many practical problems such as engineering design may be induced to constrained optimization problems. To solve those problems, even though penalty function methods are widely applied for handling constraints, the difficulty of adjusting a parameter remains a significant challenge. This study proposes a reinforcement learning-based penalty function approach. Based on it, we suggest a self-adaptive fitness in metaheuristics, for example genetic algorithms and particle swarm optimization, and also aim to improve their performance as a global optimization tool for the problems with constraints. The method contributes to effectively search for the design space that satisfies constraints, which may yield the generation of much better optimal solution. Through several benchmark test problems, this study demonstrates that the proposed method can provide relatively good performance in comparison with conventional constraint handling techniques.

Keywords: Cooperative Systems and Control, Intelligent Transportation Systems, System Modeling and Control
Abstract: With the rapid development of low-altitude logistics technology, the collaboration between vehicles and drones has emerged as a significant research focus in the field of emergency logistics. Traditional single-mode transportation methods often suffer from low efficiency due to road damage and traffic congestion during natural disasters and other emergencies. To address this challenge, this paper focuses on a region-based multimodal vehicle-drone collaborative delivery problem in emergency scenarios. A mixed-integer linear programming (MILP) model is proposed to maximize customer demand satisfaction and delivery balance under dynamic environmental conditions. Different disaster regions are classified (such as vehicle-drone handover zones, drone-only or truck-only zones), and an integrated scheduling and routing optimization approach is designed by considering transportation capacity, endurance limitations, and weather conditions. Several datasets are constructed and solved using the gurobi optimizer. Experimental results show that the proposed approach can significantly enhance the completion rate and balance of emergency deliveries in complex environments, providing theoretical support and practical guidance for emergency logistics planning.

Keywords: Trust in Autonomous Systems, Infrastructure Systems and Services, Smart Sensor Networks
Abstract: Modern Industrial Control Systems (ICS) have become prime targets for cyberattacks due to the increasing connectivity and complexity of operational infrastructures. These attacks can have severe consequences, making intrusion detection crucial. Recent deep learning approaches have steadily improved intrusion detection accuracy. However, they often fail to model the relationships between multiple sensors and actuators, lack the integration of physical priors, and demonstrate limited adaptability to varying conditions. In this paper, we propose GECAT (Graph-Enhanced Causality-Aware Transformer), a novel intrusion detection framework tailored for ICS data streams. Our approach combines graph-based sensor relationship modeling, causal inference principles, and a specially designed temporal transformer backbone to robustly detect and classify complex attacks.We integrate physical domain knowledge with data-driven graph learning modules to create an adaptive representation that adapts to diverse scenarios and enforces realistic process constraints. Experimental results on SWaT and WADI demonstrate that GECAT significantly outperforms a comprehensive set of state-of-the-art baselines across multiple metrics, indicating its effectiveness in defending ICS against sophisticated adversaries. Additionally, extensive ablation studies confirm the distinct contributions of each core component in our design.

Keywords: Decision Support Systems, Discrete Event Systems, Manufacturing Automation and Systems
Abstract: Discrete-event simulation (DES) is widely used in industrial decision-making to evaluate system performance before implementing changes. In complex environments like helicopter assembly lines, DES provides accurate key performance indicators (KPIs) such as Work-in-Progress (WIP), investment, and the rate of respect of customer due dates (customer satisfaction). However, each simulation run typically takes several hours, and users must test multiple configurations to find an acceptable solution. This makes DES impractical for daily operations or real-time planning. We developed four surrogate models—Random Forest, XGBoost, Transformers, and CNN-MLP that approximate simulation outputs in milliseconds. Despite their speed, these models lose accuracy when production conditions change. We propose an online active learning approach that updates surrogate models with selected new simulations to maintain accuracy over time.

Keywords: Adaptive Systems, Control of Uncertain Systems, Decision Support Systems
Abstract: This paper explores the application of the Adaptive Online Non-Singleton (ADONiS) framework for automated decision making for neonatal oxygen support. Maintaining optimal oxygen saturation (SpO2) levels in preterm infants is critical for preventing severe complications such as chronic lung disease and retinopathy of prematurity. Current clinical practice relies on manual adjustment of oxygen support by bedside caregivers, a process complicated by sensor uncertainty caused by contextual factors, such as challenging placement of sensors on, and high mobility of babies. Studies indicate this manual approach results in infants spending only 30–40% of time within target SpO2 ranges, highlighting the potential for automated systems. Crucially however, such systems must combine the handling sensor uncertainty handling with clinical interpretability. The ADONiS framework was designed specifically to address systems where input noise is a challenge while established and ideally immutable rule sets and associated well-defined linguistics terms and fuzzy sets, are available which describe the desired and verified behaviour of a given system. In this paper we explore the applicability of ADONiS to the setting of neonatal oxygen support. We collaborated with a neonatology expert at Queen’s Medical Centre (QMC), one of the largest hospitals in Europe, to define the system’s membership functions and rule base, ensuring clinical interpretability remained central to the design. Additionally, we collected a real-world dataset from six neonates at QMC to validate our approach. Using these expertderived structures, we apply the ADONiS framework to model oxygen support, systematically evaluating its performance on both synthetic scenarios and the collected real patient data. The resulting oxygen support suggestions were then reviewed by a neonatologist, whose assessments on ADONiS’s usage in compare to the traditional singleton approaches. Quantitive metrics were not feasible as in neonatal care optimal oxygen adjustment inherently lacks a definitive ground truth which leaves clinical judgment to be the most appropriate evaluation metric. This work represents an initial, exploratory application of the ADONiS framework to

Keywords: Image Processing and Pattern Recognition
Abstract: The advent of 3D Gaussian Splatting (3DGS) has achieved a major breakthrough in the field of 3D reconstruction and realized real-time high-quality novel view synthesis. However, 3DGS relies on the color backpropagation gradients, which are greatly affected by the illumination intensity, to optimize the position parameters of gaussian spheres such as the average center coordinates and opacity, resulting in issues like blurring and artifacts when handling scenes with uneven illumination. This is particularly important in 3D reconstruction of autonomous driving scenarios, especially in night scenes with streetlights and vehicle lights. To address this issue, we propose the POGS framework, which integrates depth and edge constraints to optimize the position parameters of gaussian spheres and reduce the interference of illumination intensity. Our approach first incorporates a pre-trained monocular depth estimation network to generate depth maps, which are used to constrain the positions of gaussian spheres. In addition, we introduce an Edge Optimization Loss and incorporate the Segment Anything Model (SAM) to generate contour maps for constraining the weights of gaussian spheres, enabling the model to pay more attention to the gaussian spheres of the objects composing the scene. Our method significantly enhances the reconstruction performance of 3DGS on datasets with low texture, high depth of field, and low illumination, improving the SSIM by 3.1%, and outperforms the state-of-the-art method RawNeRF (2% SSIM) in NeRF while maintaining real-time rendering speed.

Keywords: Image Processing and Pattern Recognition, AI and Applications, Deep Learning
Abstract: Real-Time DEtection TRansformer (RT-DETR) has attracted considerable attention for its innovative real-time and end-to-end object detection. However, it suffers from insufficient spatial and channel information in shallow features as well as information loss during the network process. To alleviate these problems, this paper proposes two novel modules: an Atrous Attention Module and a Cross-Feature Interaction Module. The Atrous Attention Module captures contextual information by operating on feature maps at different scales, enhancing them in both spatial and channel dimensions and improving their information richness. The Cross-Feature Interaction Module, inspired by the Cross-Attention mechanism of Transformer, facilitates better interactions between high-level and low-level features. It mitigates information loss by capturing channel similarities and differences across feature maps at different scales. To demonstrate our model's effectiveness, we conducted extensive experiments on MS COCO 2017 and Pascal VOC 2012 datasets. The results show that our model achieves significant performance improvements without notable reduction in efficiency.

Keywords: Image Processing and Pattern Recognition, Application of Artificial Intelligence, Deep Learning
Abstract: Early and accurate tumor classification in ultrasound images plays a pivotal role in improving cancer diagnosis and patient outcomes. Existing computer-aided diagnostic (CAD) algorithms often rely on cropping-based single feedforward pathways, which can result in the loss of crucial contextual information around the tumor. The surrounding ultrasound data, including relative intensity, plays a significant role in tumor diagnosis, and incorrect cropping or positioning may lead to unreliable results. To overcome these limitations, we propose a novel Brain-inspired Dual-stream Network (BidsNet), aiming to emulate the functional mechanisms of the dorsal and ventral streams in human visual processing. BidsNet processes the entire ultrasound image as input, preventing errors or loss of contextual details from cropping. The dorsal stream in BidsNet specializes in extracting spatial features, such as shape and texture, while the ventral stream focuses on object recognition and classification. A cross-stream communication mechanism is introduced to facilitate dynamic information sharing between the streams: spatial attention generated in the dorsal stream informs the ventral stream to improve feature localization, while channel attention derived from the ventral stream refines spatial feature representation in the dorsal stream. This collaborative interplay boosts both the interpretability and performance of the network. Extensive experiments on multiple ultrasound datasets demonstrate that BidsNet delivers superior accuracy and interpretability, validating the effectiveness of its dual-stream design and cross-stream communication mechanism.

Keywords: Image Processing and Pattern Recognition, Deep Learning, Neural Networks and their Applications
Abstract: The automated analysis of chest X-rays using deep learning technologies has emerged as a critical trend in healthcare, offering the potential to enhance diagnostic efficiency and reduce clinician workload. However, the prohibitive cost of annotation and privacy concerns restrict the expansion of chest X-ray datasets, hindering the development of chest X-ray analysis models. One solution is to introduce generative models that synthesize new samples to expand datasets. Recent advances in diffusion models have demonstrated impressive capabilities, yet prevalent diffusion-based data augmentation techniques typically depend on alterations in color and shape, which are not always appropriate for medical imaging tasks. Consequently, we propose a diffusion-based data augmentation method that generates noise incorporating only the essential information, derived from the region of interest and guided by textual prompts. This noise is then combined with noise that encapsulates global information through concatenation, resulting in a synthesized version of chest X-rays that satisfy the fine-grained characteristics of medical images. We fine-tune the diffusion model using a publicly available dataset and evaluate our method on typical chest X-ray classification tasks. The results show that our method outperforms common data augmentation methods on multiple tasks.

Keywords: Image Processing and Pattern Recognition, Deep Learning, Neural Networks and their Applications
Abstract: Artificial histopathological image classification is emerging as a promising approach for assisting early cancer diagnosis and informing treatment planning. Although convolutional neural network (CNN) based methods have achieved significant success, they still struggle to capture the rich multi-scale and high-frequency details inherent in pathological tissues. To address the above issue, we propose WaveConvX, a novel histopathological image classification model that integrates multi-level wavelet decomposition with the ConvNeXt backbone. Our approach applies a two-stage discrete wavelet transform (DWT) to intermediate feature maps, enabling the decomposition of features into multiple frequency sub-bands. High-frequency components are enhanced using Adaptive Power Gabor Convolution (APGConv), while mid-frequency ranges are refined through tailored attention mechanisms. These processed sub-bands are then fused via inverse wavelet transforms, producing feature maps enriched with both global context and local morphological cues essential for accurate diagnosis. Comprehensive experiments are conducted on three public datasets (BreakHis for breast cancer, KBSMC for gastric cancer, and LC25000 for lung and colon cancer). The results demonstrate that WaveConvX consistently outperforms ten state-of-the-art benchmarks, achieving superior accuracy, F1 scores, and robustness across multiple types and magnifications of cancer. Our work demonstrates the significant potential of wavelet-enhanced CNNs in histopathological image analysis.

Keywords: Image Processing and Pattern Recognition, Machine Vision, Deep Learning
Abstract: Positive Unlabeled (PU) learning, as a form of weakly supervised learning, is characterized by only having a fraction of positive samples labeled, while all the others remain unlabeled. The objective of PU learning is to train a binary classifier that can effectively distinguish positive and negative samples with the incomplete labeling. To achieve a more realistic distribution of samples, we propose a novel approach leveraging the semantic information of the latent classes to expand positive samples via CLIP. Especially, we design and propose a kind of adapter training method, which enables the proposed framework to adapt the medical images well. We validate the effectiveness of the proposed ELSPU method. Extensive experiments demonstrate that ELSPU significantly outperforms baseline methods, with an average accuracy improvement of 1.96% on common benchmarks, highlighting its competitive performance.

Keywords: Machine Vision, Image Processing and Pattern Recognition
Abstract: Real-time surgical video tracking faces critical challenges in highly dynamic endoscopic environments, including rapid instrument-tissue interactions, computational inefficiency, and annotation scarcity. This paper proposes S-Track, a novel SAM-based framework integrating a memory-augmented spatiotemporal fusion module, a dual-path motion prediction mechanism, and a dynamic attention-guided computation strategy. These innovations synergistically address the accuracy-efficiency trade-off, achieving state-of-the-art performance while reducing GPU memory usage by 27% and accelerating inference speed to 27 FPS. Furthermore, we introduce URS-Lithotripsy, the first comprehensive dataset for multi-target surgical tracking, comprising 20 hours of high-resolution videos with 1.2 million expert-annotated frames. Cross-domain validation on EndoVis 2018 demonstrates robust generalization, and ablation studies confirm the necessity of each module. The efficiency, precision, and adaptability to sparse annotations highlight the potential of S-Track for real-time clinical deployment.

Keywords: Machine Vision, Image Processing and Pattern Recognition
Abstract: Multiple-Object Tracking (MOT) has been an attracting area in these years with excellent progresses. However, there are still complicated uncertainties caused by movement of pedestrians due to environmental disturbance and mental state. To handle this, we characterize pedestrian moving patterns through a dual-phase paradigm, and Probabilistic Graphical Model (PGM) is introduced into our work to build Hidden Markov Model (HMM) to better understand the uncertainties. Moreover, the weighted entropy mechanism is utilized in feature fusion for balanced importance between appearance and motion information. Finally, our method achieves state-of-the-art results on MOT17 and MOT20 datasets, especially in the category of graphical methods.

Keywords: Medical Informatics
Abstract: 医学图像分割在临床诊断中起着至关重要的作用。然而，分割的准确性仍然受到目标边界模糊、形状复杂、数据标注困难等因素的挑战。尽管基于深度学习的方法已经取得了重大进展，但这些方法在处理肿瘤边界模糊、多尺度分布和难以捕获小病灶等问题方面仍面临挑战。为了应对这些挑战，我们创新性地提出了一种基于 SAM 的脑肿瘤分割模型——SAM-BrainFocus。该模型基于原始 SAM 方法构建，引入了脑肿瘤特征提取器、肿瘤提示生成器和肿瘤边界细化器，显著提高了分割精度和稳健性。它在多尺度特征融合和复杂目标适应性方面取得了突破性进展。实验结果表明，SAM-BrainFocus 在脑肿瘤图像分割任务中优于现有的主流模型，实现了更高的准确性和可靠性。

Keywords: Medical Informatics
Abstract: Precise automated detection of parathyroid nodules is critical for enhancing the diagnosis and treatment of hyperparathyroidism (HPT). However, the ectopic nature, varied shapes, fuzzy boundaries, and significant patient variability of these nodules challenge traditional object detection methods. To tackle these issues, we present ASW-YOLO, an advanced detection framework built on YOLOv10, designed to address the limitations of medical imaging object detection. First, we introduce the Feature Fusion Assembly Block (FFAB), which integrates convolutional and Transformer-based modules to better capture the global context of parathyroid nodules. Second, we propose the Squeeze-Enhanced Axial Feature Pyramid Network(SEA-FPN), which uses adaptive weighting of high-level features to refine low-level feature selection and incorporates axial squeeze attention to balance detection accuracy and processing speed. Additionally, we adopt the WIoU v3 loss function to enhance bounding box accuracy by dynamically adjusting gradient gains, reducing the impact of unclear boundaries. Experiments on a specialized parathyroid ultrasound dataset show that ASW-YOLO significantly outperforms the baseline model, achieving a 95.3% mAP@50, with a 3.2% increase in recall and a 10.2% reduction in parameters. These results highlight the potential of ASW-YOLO for real-time, high precision computer-aided diagnosis in clinical practice.

Keywords: Medical Informatics, Brain-Computer Interfaces, Brain-based Information Communications
Abstract: A complete electroencephalogram (EEG) signal comprises multiple signals from various channels. Effectively aggregating these channel signals to extract spatial features is crucial for brain-computer interface applications. Traditional approaches have largely relied on methods such as channel selection and classic dimensionality reduction algorithms. However, with advancements in artificial intelligence, researchers have begun to explore deep learning-based feature extraction techniques. In recent years, numerous algorithms have been proposed for binary seizure detection, but comparatively less attention has been given to multi-class seizure classification. To address this gap, we present a novel EEG graph network, EEGGraphNet, specifically designed for classifying different seizure types. EEGGraphNet leverages a combination of a graph attention network and a gated recurrent unit to effectively capture both the spatial and temporal features of EEG data. Moreover, unlike many previous studies that focus on seizure-wise evaluation—an approach that may not adequately represent real-world scenarios—we conduct a comprehensive evaluation using both seizure-wise and patient-wise methodologies. Experimental results demonstrate that EEGGraphNet achieves a weighted F1 score of 98.15% for seizure-wise evaluation and 86.11% for patient-wise evaluation. Additionally, comparisons with state-of-the-art models reveal that EEGGraphNet matches or surpasses existing performance benchmarks.

Keywords: Medical Informatics, Cooperative Work in Design, Networking and Decision-Making
Abstract: To support the reform of integrated medical and elderly care and promote the development of the eldercare industry, this study integrates the Latent Dirichlet Allocation (LDA) into the Policy Modeling Consistency (PMC) framework, forming a three-dimensional evaluation system covering policy content, tool applicability, and network structure, thereby assessing central–local policy coordination. Based on 132 policy documents (2013–2024) from the central government and five provinces, the study extracts six key policy themes to serve as one of the variables within the PMC model and introduces Euclidean and cosine distances to measure implementation intensity and directional deviation. Results show that Shanghai demonstrates the highest level of coordination, with strong alignment and standardized execution. Guangxi and Inner Mongolia exhibit insufficient policy input and directional deviation, reflected in fiscal gaps, weak collaboration networks, and localized policy adjustments. Beijing and Qinghai require improvements in timeliness and policy network density. The study proposes a differentiated governance strategy, including targeted central oversight for high-deviation provinces, a policy diffusion network anchored in Shanghai, and enhanced policy adaptation mechanisms in ethnic regions.

Keywords: Medical Informatics, Human-Machine Interaction, Information Systems for Design and Marketing
Abstract: This paper presents a multitask learning approach based on long-short-term memory (LSTM) networks for the joint prediction of arboviral outbreaks and case counts of dengue, chikungunya, and Zika in Recife, Brazil. Leveraging historical public health data from DataSUS (2017–2023), the proposed model concurrently performs binary classification (outbreak detection) and regression (case forecasting) tasks. A sliding window strategy was adopted to construct temporal features using varying input lengths (60, 90, and 120 days), with hyperparameter optimization carried out using Keras Tuner. Model evaluation used time series cross-validation for robustness and a held-out test from 2023 for generalization assessment. The results show that longer windows improve dengue regression accuracy, while classification performance peaked at intermediate windows, suggesting an optimal trade-off between sequence length and generalization. The multitask architecture delivers competitive performance across diseases and tasks, demonstrating the feasibility and advantages of unified modeling strategies for scalable epidemic forecasting in data-limited public health scenarios.

Keywords: Medical Informatics, Intelligence Interaction, Human-Machine Cooperation and Systems
Abstract: Mental health disorders, including conditions like depression and suicidal ideation, have emerged as critical public health challenges. Leveraging the synergy between human expertise and intelligent systems, this study explores the potential of Bidirectional Encoder Representations from Transformers(BERT) variants for analysing mental health discourse. Using approximately 11,000 text samples collected from social media platforms, a meticulously pre-processed and annotated dataset was developed. Comparative evaluations were conducted using baseline models with balanced data across disorder categories. Results revealed that general BERT variants outperformed Clinical BERT, a medical-specific model. This study introduces a novel dataset and presents key findings that not only enhance machine learning model capabilities but also provide insights into behavioral patterns associated with mental health disorders. These contributions pave the way for developing targeted machine learning solutions tailored to mental health applications.

Keywords: Medical Informatics, Visual Analytics/Communication, Human-Machine Interaction
Abstract: The rapid advancement of artificial intelligence (AI) is revolutionizing radiology, particularly in the automation of medical report generation. AI-driven systems offer the potential to alleviate the increasing workload of radiologists while improving diagnostic accuracy, consistency, and overall workflow efficiency. Despite significant progress in multi-modal medical image captioning, existing approaches often suffer from high computational costs and limitations in modeling long-range dependencies between visual and textual features. To address these challenges, we propose MambaXray-CTL, a novel framework for medical report generation that integrates a Mamba-based vision backbone with a large language model (LLM) text decoder. By leveraging a multi-stage training strategy—including autoregressive pretraining, image-text contrastive learning, and supervised fine-tuning with contrastive regularization—our model achieves precise visual-textual alignment while maintaining computational efficiency. Experimental results on the IU X-ray and CheXpertPlus datasets demonstrate that MambaXray-CTL achieves performance comparable to or surpassing state-of-the-art methods in key metrics, particularly BLEU-4 and CIDEr, while significantly reducing inference cost compared to Vision Transformer (ViT)-based architectures. These findings highlight the promise of state space models and contrastive learning in building scalable and effective vision-language systems for real-world clinical deployment.

Keywords: Medical Informatics
Abstract: Despite the potential implications for injury prevention, research examining the specific mechanisms by which non-Newtonian fluids midsoles affect collateral ligament strain remains limited. This knowledge gap warrants further investigation, as understanding these interactions could inform both footwear design and injury prevention strategies for runners. This study quantitatively evaluates the mechanical interactions within the non-Newtonian midsole-knee joint coupling system through finite element analysis, providing a theoretical foundation for sports equipment optimization. Eighteen young college students voluntarily participated in this study. This study represents the first systematic evaluation of how NF midsole materials influence knee collateral ligament strain through an integrated approach combining finite element analysis and biomechanical testing. Our findings demonstrate that the NF midsole indirectly modulates collateral ligament strain distribution by significantly altering knee joint kinetic parameters (moments and angles). The findings suggest that strategic implementation of NF materials in footwear can positively influence knee joint loading patterns, offering promising directions for injury prevention technologies in athletic footwear.

Keywords: Application of Artificial Intelligence, Agent-Based Modeling, Neural Networks and their Applications
Abstract: With the proliferation of sixth-generation (6G) communication networks, mobile edge computing (MEC) has become essential for addressing diverse user demands. The increasing heterogeneity of mobile devices, along with the rise of delay-sensitive and environmentally friendly tasks, exacerbates latency and energy consumption challenges. This paper proposes GERL-OA, a task offloading and resource allocation algorithm based on graph attention auto-encoder(GATE) assisted deep reinforcement learning (DRL). The MEC environment is modeled as an undirected graph, with spatial features extracted via graph neural network (GNN). GATE is utilized for unsupervised feature extraction, and a DRL framework is applied to optimize offloading decisions and resource allocation, minimizing the weighted sum of delay and energy consumption. Simulation results validate the superior convergence speed and global optimization capability of GERL-OA. Furthermore, Extensive experimental results demonstrate the algorithm’s strong adaptability to different task types, confirming its effectiveness in balancing the requirements of delay-sensitive and environmentally friendly applications.

Keywords: Swarm Intelligence, Agent-Based Modeling, Optimization and Self-Organization Approaches
Abstract: We address a fundamental challenge in coordinating large-scale 3D drone swarms: how to achieve rapid collective response to environmental stimuli while ensuring group stability and safety. Existing swarm navigation modals often rely on sophisticated individual perception and communication capabilities, which can be computationally expensive and impractical for large swarms. In this paper, we propose the Non-reciprocal Collective Emergent Navigation model (NRCE), a decentralized approach designed for real-world drone flocking in complex environments. Unlike traditional models, our approach leverages localized non-reciprocal interactions, where boundary drones detect environmental stimuli and propagate this information throughout the swarm without directly controlling individual trajectories. Through extensive numerical simulations and physical experiments with up to 28 drones, we demonstrate how this model achieves coordinated collective motion while effectively balancing stability with responsiveness. Our findings reveal two notable insights: (1) intermediate cohesion levels (ωc) optimize collective response—a “Goldilocks zone” where individuals are neither too tightly coupled nor too independent, challenging the conventional wisdom that stronger cohesion always improves coordination; and (2) swarm queue configuration significantly affects optimal interaction parameters, with divergent trends observed between attraction- and repulsion-based coordination mechanisms as layer count increases. These discoveries provide critical design principles for cost-effective, high-density swarm systems while advancing the theoretical understanding of collective dynamics in both artificial and biological systems.

Keywords: Swarm Intelligence, Metaheuristic Algorithms, Agent-Based Modeling
Abstract: We propose a method using ant colony optimization (ACO) with load-balancing techniques for multi-agent path finding problems. In automated warehouses and manufacturing facilities, multiple automated guided vehicles (AGVs) must efficiently perform transportation tasks without colliding with obstacles or other agents. Although several metaheuristic approaches, including ACO-based methods, address this challenge, conventional methods do not account for scenarios with many agents, and/or the planning efficiency and quality significantly decrease as the number of agents increases. The proposed method is inspired by conventional ACO-based approaches but differs in three aspects. First, we carefully define the collisions by considering the agent size and travel time. Second, we propose a tailored collision avoidance method for numerous agents by determining whether they should wait for other agents or generate detours. Finally, a directional load-balancing method that considers the movement direction is proposed for more effective ACO path generation. Our experiments demonstrate that the proposed method generates superior coordinated paths without increasing CPU time, even in high-density environments.

Keywords: Agent-Based Modeling, Metaheuristic Algorithms, Heuristic Algorithms
Abstract: In this paper, a daily activity schedule for each resident is generated according to Survey on Time Use and Leisure Activities (STULA) in Japan. To construct agent-based model as a social simulation, it is important to develop activity schedule for each agent. Each agent should take their actions according to their roles in their community such as mother in a family, a worker in a com- pany, a player in a football team, and so on. Their activity schedule should be generated according to their roles in their community. To generate such daily schedule, we employ STULA survey in Japan that is conducted every 5 years by Statistics Bureau, Ministry of Internal Affairs and Communications, Japan. Since this survey is conducted by the government and all residents selected as respond- ents are required to reply to the survey mandatorily. We employed the results of this survey to generate daily activity schedule for each resident. We propose a daily activity schedule generation method to consider behavior continuity in each schedule. We compare the results of daily activity schedules generated by a method that are not consider the continuity of activities with the results of the proposed method. Comparing those results, the proposed method can generate more schedules with continuous activities.

Keywords: Intelligent Transportation Systems
Abstract: Urban traffic flow prediction is a critical task in Intelligent Transportation Systems, yet existing methods often struggle with capturing long-range spatial dependencies and preserving high-frequency signals. To address these challenges, this paper proposes LGM4TFP, a novel framework that incorporates a Graph Attention with High-frequency Enhancement (GAHE) module. GAHE integrates dynamic graph attention for spatial feature extraction and a spectral-domain high-frequency enhancement mechanism to alleviate the oversmoothing problem. Extensive experiments on four real-world datasets demonstrate that LGM4TFP consistently outperforms state-of-the-art models by 5–10% in RMSE, MAE, and MAPE. Ablation studies confirm the effectiveness of the proposed modules, and results show that the model maintains strong robustness across diverse prediction scenarios, highlighting its practical value for dynamic traffic management.

Keywords: Intelligent Transportation Systems
Abstract: The scope of this paper is to tackle a non-linear multi-objective optimization fixed-cost transportation problem (MO-FCTP) with multiple modes of transportation with given capacities, having two contradictory minimization objectives: the total transportation cost and, respectively, the total distribution time. The problem studied is solved using a modified NSGA-II, obtained by using a Prufer-based representation, designing an efficient decoding procedure, and integrating into its framework newly developed genetic operators tailored to accommodate multiple modes of transportation. We perform computational experiments for existing instances from the literature. The results obtained show that our Prufer-based NSGA-II outperforms the current state-of-the-art algorithm for solving the non-linear MO-FCTP with multiple modes of transportation with given capacities, providing improved Pareto fronts for all instances considered.

Keywords: Intelligent Transportation Systems, System Architecture, Decision Support Systems
Abstract: With the rise of smart mobility and shared e-mobility services, numerous advanced technologies have been applied to this field. Cloud-based traffic simulation solutions have flourished, offering increasingly realistic representations of the evolving mobility landscape. LLMs have emerged as pioneering tools, providing robust support for various applications, including intelligent decision-making, user interaction, and real-time traffic analysis. As user demand for e-mobility continues to grow, delivering comprehensive end-to-end solutions has become crucial. In this paper, we present a cloud-based, LLM-powered shared e-mobility platform, integrated with a mobile application for personalized route recommendations. The optimization module is evaluated based on travel time and cost across different traffic scenarios. Additionally, the LLM-powered RAG framework is evaluated at the schema level for different users, using various evaluation methods. Schema-level RAG with XiYanSQL achieves an average execution accuracy of 0.81 on system operator queries and 0.98 on user queries.

Keywords: Intelligent Transportation Systems
Abstract: Driving behavior recognition is significant to autonomous driving. However, the vast model size and computational expense makes deploying state-of-the-art driving behavior recognition models on vehicle-edge devices difficult. In this paper, a novel lightweight hybrid Model, LVBRM is proposed for vehicle driving behavior recognition, which consists of a Transformer-based vehicle detection network and an improved SlowFast behavior recognition network. Firstly, a Global-Local Knowledge Distillation (GLKD) method is proposed to efficiently compress the Transformer-based vehicle detection network and effectively improve the detection accuracy. Secondly, a Frame-Difference and Attention-enhanced (FDA) strategy is designed for behavior recognition network to improve driving behavior recognition accuracy with slightly increasing of model parameter. Finally, extensive experiments on a public large-scale BDD dataset and our self-built VBR-D dataset demonstrate the superiority of our proposed LVBRM, the recognition accuracy gains over the state-of-the-art models while with less parameters.

Keywords: Intelligent Transportation Systems, Autonomous Vehicle
Abstract: Radar sensors are superior at measuring distances and velocities, even in adverse weather conditions. However, the noisiness of radar point clouds requires a filtering cascade to make these sensors usable for applications in automated driving. This filtering may cause hazardous situations if points on existing objects are removed by a filter. To serve as a perception component in fully automated driving systems, radar's error rate needs to be reduced significantly. To achieve this goal, we advance the view that points should be filtered adaptively according to their presumable relevance for the current driving task. In this work, we present a real-world study of our radar point criticality estimation algorithm. In addition, we introduce methods for acting on critical points. While the Posterior method recalls points in critical regions, we demonstrate how the creation of clusters and tracks can be facilitated for critical radar points. Our methods are evaluated on a novel dataset including 92 critical scenes with pedestrians in a parking garage. We demonstrate that all methods significantly increase detection rates at both the cluster and track levels and enable earlier detection. However, utilizing every radar point inevitably leads to several false positives. Future work should investigate how to invalidate these points by additional perception sensors.

Keywords: Intelligent Transportation Systems, Autonomous Vehicle, Trust in Autonomous Systems
Abstract: Despite advancements in perception and planning for autonomous vehicles (AVs), validating their performance remains a significant challenge. The deployment of planning algorithms in real-world environments is often ineffective due to discrepancies between simulations and real traffic conditions. Evaluating AVs planning algorithms in simulation typically involves replaying driving logs from recorded real-world traffic.However, entities replayed from offline data are not reactive, lack the ability to respond to arbitrary AV behavior, and cannot behave in an adversarial manner to test certain properties of the driving policy. Therefore, simulation with realistic and potentially adversarial entities represents a critical task for AV planning software validation. In this work, we aim to review current research efforts in the field of traffic simulation, focusing on the application of advanced techniques for modeling realistic and adversarial behaviors of traffic entities. The objective of this work is to categorize existing approaches based on the proposed classes of traffic entity behavior and scenario behavior control. Moreover, we collect traffic datasets and examine existing traffic simulations with respect to their employed default traffic entities. Finally, we identify challenges and open questions that hold potential for future research.

Keywords: Intelligent Transportation Systems, Consumer and Industrial Applications, Autonomous Vehicle
Abstract: Traffic warning systems have attracted attention for reducing traffic accidents caused by motorcycles. The authors previously developed a highly portable system that uses an in-vehicle web camera and a smartphone to detect motorcycles approaching at high speed based on rear-camera input images and notifies the driver of danger. In this study, we propose a system that combines deep-learning-based object detection and heuristics to detect encroaching vehicles from images taken by a front-facing in-vehicle camera. First, we consider the shape of the bounding box of the detected vehicle to identify the encroaching vehicle. Since the encroaching vehicle is turned sideways when entering the main road, we identify side-facing vehicles by calculating the aspect ratio of the bounding box. For motorcycles, the criterion for distinguishing between parked motorcycles and encroaching motorcycles is the presence or absence of a rider. To distinguish between parked vehicles and moving vehicles, the values of the background pixels around the detected vehicle are compared with the values of pixels around the corresponding vehicle in the previous frame and the difference is used to determine whether the vehicle is moving. The results show that the above heuristics improve the accuracy of detecting encroaching vehicles.

Keywords: Intelligent Transportation Systems, System Modeling and Control, Large-Scale System of Systems
Abstract: 人轨迹预测具有重要意义 在自动驾驶、智能安防、 和机器人导航。其准确性直接影响 这些系统的安全性和可靠性。传统型号 通常仅依赖行人等固定信息 速度、方向和距离。他们不能有效地 捕捉复杂的行人交互关系和 未能深度融合场景和社交功能， 限制了预测精度的提高。在这个 研究，动态场景轨迹融合自适应图 卷积轨迹预测模型（DSTFAGCN）是 提出。该模型创新地引入了动态 重量调节机制和分层图 结构，实现更精准的动态社交 交互建模。它还使用了时间场景功能 结合场景关注的传播机制 实现更有效时空场景的机制 特征融合。同时，基于 Transformer 的图 采用卷积网络（TGCN）显着 增强图卷积效果。实验 ETH-UCY数据集显示，与现有模型相比， DSTFAGN模型的ট

Keywords: AI and Applications, Application of Artificial Intelligence
Abstract: Interactive art appreciation is a method in which multiple participants engage in repeated discussions to interpret artworks. In this approach, a facilitator plays a crucial role by asking questions to the viewers. However, there are currently limited opportunities for facilitator training. To address this, this paper proposes a Mixed Reality (MR) AI system that utilizes MR headsets and a Large Language Model to train facilitators. The system features five virtual viewers, each with distinct personalities and whose facial expressions change through dialogue, allowing users to practice facilitation through interactive dialogue. Contents of what the virtual viewers say are generated by the Large Language Model, GPT-4o. Moreover, the users’ voice is transcribed using the speech recognition model, Whisper, and the virtual viewers’ statements generated by GPT-4o is converted into voice using the speech synthesis system, Azure Text-to-Speech. In addition, the impression data was collected from approximately 500 people, who were asked about their thoughts and impressions after viewing the artwork. When generating the virtual viewers’ statements with GPT-4o, this impression data is provided as input to ensure that the virtual viewers’ responses are as close as possible to what real people actually think and feel when viewing the artwork. Evaluation experiments with facilitator trainees were conducted, and many participants expressed interest in the system.

Keywords: Cloud, IoT, and Robotics Integration, Transfer Learning, Image Processing and Pattern Recognition
Abstract: Professional Burnout Syndrome (PBS) among healthcare professionals has been considered a threat to both staff well-being and patient safety, especially in a high-stress medical environment. While robotics and AI have been increasingly integrated into healthcare, their impact on PBS has not been explored in detail yet. Therefore, this paper proposes a real-time PBS detection framework for healthcare professionals using emotion recognition. This framework includes a deep learning-based emotion detection system for humanoid companion robots, which then correlates emotional trends with the Circumplex model to identify burnout risk. Our experimental evaluation results across five deep learning architectures, MobileNet, RegNetY, Swin Transformer, ConvNeXt V2, and EVA-02, show a highest accuracy of 74.05% on a public emotion dataset. These results demonstrate the feasibility of integrating such systems into healthcare workflows for early PBS warnings. Also, this work suggests a human-in-the-loop diagnostic model, where robotic emotion detection complements clinical expertise by providing proactive support, strengthening workforce resilience, and maintaining the quality of patient care.

Keywords: Computational Intelligence, Hybrid Models of Computational Intelligence, Machine Learning
Abstract: Research and development of autonomous driving technology are progressing rapidly. However many people still have concerns regarding the comfort and safety of autonomous vehicles. One key issue is the discrepancy between human driving behavior and the driving style adopted by current autonomous systems. To address this problem, we have proposed an autonomous cornering control system based on the Sense of Circular Vision (SoCV), which takes human visual perception during cornering into account. In our previous work, the proposed system demonstrated that its generated trajectories in simulation closely resembled those of expert drivers. However, the impact on subjective discomfort and ride quality had not yet been clarified. This paper evaluates these aspects using both subjective and objective measures through experiments conducted in a VR simulation environment and with real-world driving videos. Experimental results show that the proposed system improves subjective comfort and perceived safety, and promotes anticipatory gaze behavior compared to conventional control methods.

Keywords: Deep Learning, Application of Artificial Intelligence, AI and Applications
Abstract: Swallowing disorders have become a major problem among the aging population. A chin-up posture, along with curvature of the back, is considered unfavorable for swallowing and may lead to aspiration of saliva or food. To prevent this, a chin-down posture has been widely adopted; however, it has not been consistently defined. Therefore, we propose an innovative aspiration risk assessment system based on swallowing posture, using computer-based image analysis. The main contribution of this study is that the proposed system can assess aspiration risk with a high Area Under the Curve (AUC) of 0.799. This result was achieved solely by training a Light Gradient Boosting Machine (LightGBM) using total 113 features on lateral-view images of swallowing posture. In addition, we identified three novel predictive indexes: shoulder angle, the angle formed by tragus, acromion, and posterior and superior iliac spine (PSIS), and the angle formed by acromion, 4th lumbar vertebrae (L4), and PSIS. The proposed system was evaluated using a dataset of 68 participants over 65 years of age, using leave-one-out cross-validation (LOOCV). Its performance showed improvement compared to previous manual evaluations based on head angle, cervical angle, shoulder angle, pelvic angle, kyphosis index, and cervical range of motion. The experimental results indicated that cervical range of motion was not a particularly important factor. Moreover, it is expected that this assessment system will be applicable in the future. This will require only lateral-view photographs of swallowing posture taken with a smartphone.

Keywords: Cybernetics for Informatics, Neural Networks and their Applications, Transfer Learning
Abstract: The present work enhanced the classification accuracy of motor imagery (MI) based on EEG signals using a novel deep neural network (DNN) architecture that integrated InternImage and ST-pooling to capture spatiotemporal features. While this architecture outperformed existing methods in classification accuracy, the validation loss indicates overfitting during training. We introduce a bias-variance guided layer selection (BVG-LS) strategy into the fine-tuning process to address this issue. This approach adaptively adjusts the number and selection of layers updated during fine-tuning, replacing the conventional single-layer update with a more effective multi-layer configuration guided by BVG-LS. Experimental evaluation on the cross-individual validation task using the PhysioNet EEG Motor Movement/Imagery dataset showed that the accuracy of two-, three-, and four-class classification were improved to 89.06%, 81.13%, and 70.90%, respectively, surpassing the performance of existing fine-tuning techniques, including full fine-tuning.

Keywords: Biometric Systems and Bioinformatics, Computational Life Science, Cybernetics for Informatics
Abstract: In recent years, research and development of human-assist robots equipped with brain-computer interfaces (BCIs) have accelerated due to the increasing number of stroke cases worldwide, which are a major contributor to permanent disabilities and loss of independence. BCIs can be an effective neurorehabilitation tool, as they assist the impaired sensorimotor loop by providing compensatory somatosensory feedback during motor attempts. The human-assist robot must support the user’s movements based on their intended actions. However, determining whether the robot’s assist motion is as intended by the user in real time remains challenging. This paper proposes a multi-stage method that utilizes time-frequency and statistical analyses for electroencephalography (EEG) signals to detect whether the arm movements are as intended. Furthermore, it employs a binary classifier with threshold optimization—guided by the highest classification accuracy— to determine the most effective threshold for distinguishing between intentional and unintentional movements. The effectiveness of the proposed method was evaluated through experiments involving multiple arm movements that combine both simple and complex tasks. A disturbance mechanism device was built and employed during these experiments to generate the sense of motion mismatch by applying disturbance forces. The experimental results demonstrate the effectiveness of the proposed method in detecting motion mismatch and distinguishing between intentional and unintentional movements. The proposed method is applicable to human-assist robots to assess whether the robot’s assist movement matches the user’s intended motion.

Keywords: Biometric Systems and Bioinformatics
Abstract: Impatience is a common psychological state experienced in daily life, particularly when performing tasks under time pressure. In such states, individuals may experience impaired judgment, which can lead to human errors. Therefore, detecting impatience in advance could prevent decision-making mistakes and reduce the occurrence of human errors. This study aims to develop a method for detecting impatient states using non-contact measurements of eye gaze and pupil diameter. We investigated the feasibility of classifying a state of impatience using logistic regression analysis, based on features extracted from gaze and pupil data. Three levels of impatience were defined, and classification performance was evaluated using two of these levels. As a result, a classification accuracy exceeding 70% was achieved when distinguishing between the highest and lowest perceived impatient states. Furthermore, the feature selection process revealed that variables related to gaze velocity were frequently selected, suggesting their effectiveness in impatience detection.

Keywords: Machine Vision, Image Processing and Pattern Recognition, Biometric Systems and Bioinformatics
Abstract: In recent years, the shortage of nurses has become a global problem. Furthermore, in large-scale surgeries such as total knee arthroplasty, a large number of surgical instruments with similar shapes are prepared, which places a heavy burden on operating room nurses who manage them. To solve these problems, we have developed an automatic surgical instrument recognition system to support operating room nurses. We have also developed a system that applies projection mapping technology to highlight automatically recognized surgical instruments in the real world in real time, making it easy for nurses in the operating room to understand. In this study, we propose a method to correct the highlighting using ArUco markers. In experiments, we confirmed object detection using surgical instruments and demonstrated the effectiveness of this system.

Keywords: Assistive Technology
Abstract: Peripheral nerve injury often impairs dorsiflexion, leading to asymmetrical gait and increased metabolic cost. While passive ankle-foot orthoses (AFOs) remain common in clinical use, their static, non-adaptive nature limits rehabilitation. Existing active orthoses face challenges such as unreliable gait transition detection, suboptimal actuation, poor generalization, and lack of open-sourcing. This work presents a smart, wearable interface featuring a reproducible design and Bowden cable transmission. A lightweight Bayesian framework enables robust real-time gait transition detection within robot operating system (ROS). The system employs phase-dependent control via a finite state machine, modulating ankle impedance based on gait events. Validation on three unseen users walking naturally demonstrated up to 25% dorsiflexion assistance and phase recognition accuracies of 99.21% (seen interfaces) and 98.75% (unseen), highlighting the system’s adaptability.

Keywords: Assistive Technology
Abstract: Tactile graphics are essential for providing access to visual information for the 43 million people globally living with vision loss. Traditional methods for creating these graphics are labor-intensive and cannot meet growing demand. We introduce TactileNet, the first comprehensive dataset and AI-driven framework for generating embossing-ready 2D tactile templates using text-to-image Stable Diffusion models. We fine-tune Stable Diffusion models using Low-Rank Adaptation and DreamBooth to generate high-fidelity, guideline-compliant graphics with reduced computational cost. Quantitative evaluations with tactile experts show 92.86% adherence to accessibility standards. Our structural fidelity analysis revealed near-human design similarity, with a Structural Similarity Index (SSIM) of 0.538 between generated and expert-designed tactile images. Notably, our method better preserves object silhouettes than human designs (binary mask SSIM: 0.259 vs. 0.215), addressing a key limitation of manual abstraction. The framework scales to 32,000 images (7,050 high-quality) across 66 classes, with prompt editing enabling customizable outputs (e.g., adding or removing details). By automating the 2D template generation step compatible with standard embossing workflows—TactileNet accelerates production while preserving design flexibility. This work demonstrates how AI can augment (not replace) human expertise to bridge the accessibility gap in education and beyond. Code, data, and models can be found at our project page https://tactilenet.github.io/.

Keywords: Assistive Technology, Brain-Computer Interfaces, Human-Machine Interaction
Abstract: This work presents the application of an upper-limb rehabilitation protocol in chronique stroke using a Brain-Computer Interface (BCI) based on Motor Imagery (MI), Non-Invasive Brain Stimulation (NIBS) like transcranial Alternating Current Stimulation (tACS), and Functional Electrical Stimulation (FES). This protocol uses the concept of Alternating Treatment Design (ATD), in which a chronic post-stroke subject is submitted to these techniques for recovery of his upper-limb movements affected by the stroke. The rehabilitation progress was verified through metrics, such as Fugl Meyer Assessment Scale (FMS), Functional Independence Measure (FIM), Modified Ashworth Scale (MAS), surface Electromyography (sEMG) and Electroencephalography (EEG). Results from these metrics include a 2.4% increase in FMS and an 11% increase in the muscle contraction of his finger extensors, evaluated by sEMG. For the EEG analysis, there was an energy increase in mu and beta rhythms at the end of protocol.

Keywords: Assistive Technology, Cognitive Computing, Medical Informatics
Abstract: Recognition of Activities of Daily Living (ADLs) has been the subject of research for several years now, with the aim of providing effective solutions, particularly in terms of personal assistance and home care. This paper aims to present an original approach to detect signs of social isolation by monitoring key digital indicators such as mobility and nutritional activity. In order to collect data in a real-life environment, we have deployed a sensor-based monitoring system for three months in the homes of five elderly people (aged 60 to 87) living alone. We have compared the results of four algorithms, to identify six activities of daily living (preparing meals, eating, washing up, hygiene, sleep/relaxation and other activities). The logical approach, based on a 20- minute time window, achieved the best results with a detection rate of 88.14% for the activity "eating a meal" in five elderly participants, and with a micro-average F1-score of 0.863 using the public ARUBA-1 database. These results represent an intermediate step in the assessment of social isolation in the elderly through the analysis of their ADLs.

Keywords: Assistive Technology, Human Perception in Multimedia, Haptic Systems
Abstract: This study examines the asynchrony detection ability between auditory and vibrotactile stimuli to improve the ``VIBES'' singing timing support system for Deaf and Hard-of-Hearing (DHH) individuals. We conducted synchrony judgment tasks with 20 DHH participants using a single device to present auditory and vibrotactile stimuli. We tested two conditions: a regular rhythm (RR) condition with contextual beats (120 beats per minute) surrounding the target stimulus and a no-rhythm (NR) condition where we presented only the target stimulus. Our results showed a significant negative shift in the point of subjective simultaneity (PSS: the timing at which two different sensory stimuli are perceived as co-occurring) under the NR condition (-15 [ms]), indicating that presenting vibrotactile stimuli before auditory stimuli facilitates simultaneity perception. We observed more expansive temporal binding windows (TBW: the period during which stimuli from different senses are integrated and perceived as a single event) when auditory stimuli preceded vibrotactile stimuli. This suggests that unintended vibrotactile interference from auditory stimuli occurs when using a single device. We also noted individual differences in asymmetry between the TBW of RR and the NR conditions. These findings suggest DHH participants may detect asynchrony based on comparison with surrounding stimuli rather than direct comparison of stimulus pairs, highlighting the importance of accurate timing estimation for adequate singing support.

Keywords: Assistive Technology, Human-Computer Interaction, Human-Machine Interface
Abstract: Our research project found that individuals with visual impairments (VI) have an advantage in developing applications for smart speakers, which are devices equipped with voice assistants. This may increase their job opportunities. This study examines the usage frequency of voice assistants, smart speakers, and smart home devices in Japan. It explores the relationship between usage frequency and reluctance to speak to a voice assistant, compares survey results from 2019 and 2023, and contrasts usage patterns between the general public and VI. Additionally, the significance of voice assistants for VI will be discussed.

Keywords: Assistive Technology, Human-Computer Interaction, Virtual/Augmented/Mixed Reality
Abstract: Real-time captions presented through AR glasses hold great promise for enhancing accessibility for Deaf and Hard-of-Hearing (DHH) individuals. However, conventional approaches often overlook the diversity of user preferences and perceptual characteristics. In this study, we explore how spatial personalization—adjustments in display method, position, and depth—can improve the user experience of AR captions. We conducted two experiments. Experiment 1 focused on 2D position personalization, revealing that user-controlled positioning enhanced visibility and comfort, though comprehension gains were modest. Experiment 2 introduced a novel concept, Simulated Depth Personalization, which emulates different viewing distances despite a fixed AR focal plane (4 m). This revealed clear individual differences: participants varied in whether their preferences aligned more with the presentation method (e.g., monocular vs. binocular) or viewing distance. Our findings emphasize that effective AR captioning is not one-size-fits-all but must flexibly adapt to individual spatial preferences. We argue for incorporating spatial personalization as a core design principle for accessible AR systems.

Keywords: Human-centered Learning, Human Factors, Human-Machine Interaction
Abstract: Human Activity Recognition (HAR) represents a significant domain within pervasive computing, facilitating a diverse array of applications ranging from healthcare to smart environments. Traditional HAR models suffer from several challenges, including data privacy and the distributed participation of heterogeneous resource-constrained devices. To mitigate these challenges, the research community popularly uses federated learning (FL). However, selecting clients in FL is a critical issue, mainly when there is a combination of resource-constrained heterogeneous devices. This paper proposes a resource- and performance-aware client selection algorithm for HAR that amalgamates the benefits of FL with energy efficiency. The framework allows for the training of machine learning models across multiple devices without the necessity of sharing raw data, thereby preserving user privacy. Additionally, it employs energy-aware strategies to diminish the carbon footprint and reduce the computational costs typically linked to traditional cloud-based HAR systems. Experimental results indicate that the proposed framework achieves more than 90% accuracy, comparable to centralized models, while significantly lowering energy consumption and improving the robustness of the model. This work contributes to the evolving field of green AI, delivering an effective, privacy-preserving, and environmentally sustainable approach for HAR applications.

Keywords: Media Computing, Deep Learning, Application of Artificial Intelligence
Abstract: The task of controlled music generation has been well developed, but the lack of modeling of music control attributes and neglect of music structure affect the quality of music creation. In order to solve the above problems, we first classify music into three levels according to its essentially different nature and extract the corresponding interpretable control attributes. Then by adding the control attributes to the music representation, a multi-level connection between music generation process and human composition is established. Finally, we propose a multi-level controllable music generation model with fine-grained control (MCMG). This model considers the structural relationships between music levels, enabling the music generation process highly controllable. The experiments show that our generative model not only improves on the basic music metric compared to the baseline, but also performs well on the controllability metric.

Keywords: Multimedia Computation, Deep Learning, Application of Artificial Intelligence
Abstract: Abstract— Visual Question Answering with Natural Language Explanations (VQA-NLE) requires models to provide logically grounded explanations while generating accurate answer. Existing methods struggle with complex questions that demand comprehensive logical reasoning and world knowledge integration. We propose that decomposing the target image and question into question-answer-explanation triples containing visual cues, world knowledge, and object relationships can enhance image understanding, question comprehension, and knowledge retrieval for final answer reasoning. We call the proposed method Sub-QAE Prompting that includes three key steps. We train a visual question generation (VQG) model using instruction-aligned question contexts which subsequently utilized to generate the sub-questions from image. Then generate image-based and text-based sub-questions via the VQG model and frozen large language model respectively. Last we construct question-answer-explanation triples by deriving corresponding answers and explanation for generated sub-questions, and encode question-answer-explanation triples into visual-aware prompting module for MLLM to generate the final answer and explanation. Experimental results on the two challenge benchmark VQA-X and A-OKVQA demonstrate that our method achieves state-of-the-art performance compared to existing VQA-NLE approaches

Keywords: Multimedia Computation, Deep Learning, Artificial Social Intelligence
Abstract: Cross-modal retrieval aims to retrieve semantically similar heterogeneous modal data based on queries from any modality. However, existing methods struggle to capture complex semantic relationships, particularly in small scale datasets, where data scarcity impairs alignment accuracy. To address these issues, we propose small scale data adaptive enhancement for Cross-modal Retrieval (SSDAE). The framework combines a Variational Autoencoder to generate high-quality enhanced samples, which alleviates data scarcity effectively. For further improve the effectiveness of data enhancement,we introduce a dynamic enhancement probability control mechanism, which selects the optimal enhancement probability according to dataset characteristics. The SSDAE incorporates Graph Attention Networks to capture high-order feature relationships of different modalities, dynamically learning the importance of different features. Besides, a modality consistency constraint is utilized to further enhance the accuracy of semantic alignment. Experimental results on three small scale benchmark datasets demonstrate that SSDAE significantly improves retrieval performance and accuracy.

Keywords: Multimedia Computation, Image Processing and Pattern Recognition, Deep Learning
Abstract: In cross-modal retrieval, the core challenge of image-text retrieval lies in the mismatch of heterogeneous modality representation spaces. Traditional methods overly rely on inter-modal relationships or optimize a single modality, neglecting cross-modal feature understanding and the preservation of original data details. In this paper, we propose a multimodal collaborative knowledge augmentation framework. Unlike single-path enhancement, we design a novel generation-reconstruction dual-path architecture. The image-to-text generation driven by BLIP-2 and the text-to-visual reconstruction driven by Stable Diffusion form a closed-loop knowledge cycle, achieving multi-granular feature enhancement through bidirectional cross-modal semantic transformation and deeply constructing the relationship within the visual-language joint embedding space. Then, we design a sparse-gated dynamic aggregation mechanism, which precisely focuses on key cross-modal semantics through feature decoupling, spatially sparse masking, and dynamic weight allocation. Additionally, the temporal-progressive feature fusion strategy controls the feature integration ratio via dynamic fusion coefficients, realizing a smooth transition from single-modal representation learning to cross-modal association reinforcement, overcoming the modal interference issue in traditional fixed-weight fusion. Experimental results demonstrate that our method outperforms existing mainstream methods on the Flickr30K and MS-COCO datasets.

Keywords: Multimedia Computation, Image Processing and Pattern Recognition, Deep Learning
Abstract: Existing cross-modal hashing retrieval methods exhibit limitations in capturing semantic correlations and co-existing information across modalities. Most approaches primarily rely on single-level feature representations and fail to fully exploit both coarse-grained and fine-grained semantics. To address these issues, this paper proposes a novel framework, Multi-Granularity Semantic Association Learning for Cross-modal Hashing Retrieval (MSAH), which integrates semantic correlation mining at both coarse-grained and fine-grained levels. The framework processes redundant information in fine-grained features while leveraging coarse-grained features to capture global-level sample relationships, subsequently propagating semantics into the hash function via knowledge distillation. In addition, we designed an Indirect Association Learning strategy to capture potential relationships between instances through the descriptive consistency of image-text pairs. This approach not only reduces the distance between samples of the same category but also mitigates the limitations of contrastive learning in handling category-overlapping samples during hash function training. Experimental results demonstrate that the proposed method significantly outperforms existing deep cross-modal hashing approaches in retrieval efficiency across multiple datasets.

Keywords: Multimedia Computation, Machine Vision, Machine Learning
Abstract: In the age of data explosion, data collection is everywhere, but data may contain a lot of private information. Therefore, anonymization techniques are used to protect privacy information while preserving important data information. For human images, personal privacy information is present not only in the face; parts of the body also contain a lot of personal identification information. Thus, human anonymization should be extended from the face to the full-body. However, current human anonymization mainly focuses on face anonymization. In this paper, we propose a multi-choice full-body anonymization framework that provides different anonymization choices at the semantic level while ensuring the image diversity and face diversity of full-body anonymized data. For full-body anonymized data, we propose a comprehensive evaluation system. We conduct a detailed evaluation of the anonymized data from three aspects: data quality and diversity, anonymization performance, and data utility. We have demonstrated through extensive experiments that our method provides more anonymous choices and better diversity performance compared to state-of-the-art methods while ensuring data quality and anonymization performance.

Keywords: Multimedia Computation, Neural Networks and their Applications, Image Processing and Pattern Recognition
Abstract: Vision and tactile are the most widely used perception modes in robotic interactions. We propose a Multimodal Visual-Tactile Sensor (MVTS) using a single camera to synchronously capture visual and tactile information. Unlike the traditional design with a fixed opaque gel layer, the MVTS consists of a color camera, a transparent elastomer layer embedded with color markers, and multiple LEDs. It can acquire images for vision like a normal camera, even during object contact. It can also use the markers to acquire tactile information while interacting with objects. In order to simultaneously acquire and separate tactile and visual information in single-shot, we designed a multimodal information separation framework based on a multi-task learning deep neural network. It adopts a shared feature encoder and two separate and parallel decoders to restore visual images and extract tactile maps. The MVTS was evaluated on multiple tasks such as force estimation, contact surface reconstruction, texture reconstruction, and volume estimation of grasped simple-shaped objects, and the results show that it has good performance in multimodal information acquisition, which can help robots better interact with the environment.

Keywords: Human-Machine Interaction
Abstract: When the society delves into the realm of Human-AI collaboration, we encounter a complex web of reliability, accountability, trust, and safety considerations. Crafting workflows and processes that up-hold stringent AI system requirements is not just a technical challenge, but a design imperative given the challenge with increasingly AI decision-making capabilities. To address the challenge, SMC Society (SMCS) started a new initiative in 2025 to formulate a Collaborative Human-AI Symbiosis (CHAIS) program and prepare, enable, and demonstrate SMCS as a world leader in CHAIS. It benefits from the internationally recognized SMCS interdisciplinary knowledge, expertise, and technology in Systems, Man, Cybernetics, and particularly Human-AI Teaming. The goal is to design, develop, and demonstrate the intricacies of Human-AI teaming and unlock the potential of this transformative alliance. The proposed workshop is the first event of the Initiative for engaging with world leaders in related expertise areas as well as SMCS and IEEE regular and student members working in the area of concepts and tools in human and AI decisionmaking processes. Specifically, the objective of this workshop is to brainstorm, identify challenges and priorities, and develop a plan for a SMCS roadmap to address the expertise and capability gaps towards a new IEEE program.

Keywords: Agent-Based Modeling
Abstract: Store layout design is of paramount importance in the realm of retail management. To enhance the layout design from a micro-level perspective, building accurate and explainable agent-based models on customers' behaviors accounting for the layout characteristics is critical yet challenging. Various methods have been developed to build data-driven agent-based models yet they are incapable to elucidate causal relations between customers' behaviors and layout characteristics to explicitly explaining the constructed model. To fill this research gap, we develop a framework specialized for retail management to refine the causal understanding of how layout characteristics influence customers' behaviors, to support explainable design and modeling of such behaviors, and based on which to enable retailers' informed decisions on layout design. By applying the proposed framework on a small retail store case with only limited real data, we demonstrated that the method outperforms conventional optimization methods in saving simulation costs and in constructing an explainable model that more accurately replicates the customers' in-store traffic.

Keywords: AI and Applications
Abstract: Text-to-image and image-to-text models allow automated (but imperfect) semantic translation across modalities. This paper presents results and preliminary analysis of an empirical study of recursive information processing in popular open-weight generative artificial intelligence (genAI) models such as FluxSchnell and BLIP-2. Through clustering and topological data analysis we show some of the ways that different genAI models and initial prompts give rise to different semantic embedding trajectories, and suggest some ways forward for understanding how semantic information is transmitted through these types of complex information-processing systems.

Keywords: AI and Applications, Application of Artificial Intelligence, Knowledge Acquisition
Abstract: This paper presents an experimental analysis of neuroscience-inspired memory architectures for agricultural information systems. Drawing from biological memory principles, we implement and evaluate four distinct approaches—Vector Database, Knowledge Graph, Finite State Machine, and a novel Hybrid Memory architecture that integrates these components. Our controlled experiments across diverse agricultural queries demonstrate that the biomimetic hybrid architecture achieves superior relevance scores (0.753) compared to single-component approaches while maintaining acceptable performance trade-offs. Matrix-based performance analysis reveals how different architectures excel in specific query types: Vector DB for factual retrieval, Knowledge Graph for relational queries, FSM for procedural information, and Hybrid Memory maintaining strong performance across all categories. This research provides quantitative evidence supporting specialized memory designs for agricultural knowledge systems, though current implementations lack multimodal capabilities that would further enhance agricultural decision support.

Keywords: AI and Applications, Deep Learning, Machine Learning
Abstract: While Large Language Models (LLMs) have demonstrated impressive capabilities in code generation and mathematical reasoning, their ability to produce correct and highly optimized numerical computation code remains largely unexplored. This paper presents a systematic evaluation framework for assessing LLMs' performance in generating computationally efficient numerical code. We introduce a benchmark comprising carefully curated classical numerical computation problems where human experts typically achieve 3-10× speedups compared to straightforward implementations. These problems require deep understanding of numerical computation and computer architecture, identification of performance bottlenecks, and application of multi-level optimization techniques. To address potential dataset contamination, we also develop a benchmark generator that creates novel variants of these optimization challenges. Our evaluation of mainstream LLMs reveals that even state-of-the-art models consistently fall short of human expert optimization levels. The benchmark is open-sourced to facilitate further research in this direction.

Keywords: Big Data Computing,, AI and Applications, Transfer Learning
Abstract: Existing Convolutional Neural Networks (CNNs) suffer from limitations such as purely using features from the final layers for decision making, without leveraging important information obtained from middle layers. To tackle such limitations, we propose a novel CNN variant with multi-layer feature aggregation and pooling permutations for sound classification. We also introduce a method reshaping intermediate features from different layers to summarize in time-axis without losing important information. The proposed model results in significant boosting on model performance over the original networks for diverse sound classification problems.

Keywords: Cybernetics for Informatics, Soft Computing, Socio-Economic Cybernetics, Agent-Based Modeling
Abstract: The blockchain framework has increasingly moved from purely finance and the digital world to acquire a greater role as an indispensable diaphragm between the physical and the information processing parts of Cyber-Physical Systems. A significant need is the requirement for trustworthy contact between the two domains, which today can be handled with the Oracle concept and its inherent limitations. This proposal calls on the holonic paradigm to mitigate the Oracle problem, employing a constructivistic and second-order cybernetic stance that allows the trust model of the oracle to evolve and follow the dynamics of complex reality.

Keywords: Cyborgs,, Neural Networks and their Applications, Deep Learning
Abstract: In this work, we explore group-equivariant neural networks into reinforcement learning through Proximal Policy Optimization (PPO) to enhance locomotion learning in high- dimensional humanoid agents. By leveraging the symmetry properties of the robot’s morphology through Equivariant Multi-Layer Perceptrons (EMLP), we aim to improve sample efficiency and training stability. Our experiments with the 21-DoF Unitree H1 humanoid suggest that while EMLP combined with PPO improves sample efficiency, vanilla PPO achieves marginally higher performance in terms of gait quality and biomechanical realism. For assessment, we propose a suite of human-inspired biomechanical metrics, such as joint trajectory deviation, gait symmetry, phase consistency, energy efficiency, and motion smoothness—comparing learned policies against human motion capture (MoCap) data. We also provide these metrics as a framework for quantitatively evaluating the similarity between human gait and reinforcement learning-based humanoid locomotion. Surprisingly, despite the theoreticalbenefits of equivariance, our findings suggest that excessive symmetry constraints may limit expressivity and impede the emergence of human-like locomotion in complex agents. This study provides key insights into the trade-offs of incorporating symmetry priors in deep reinforcement learning for humanoid control.

Keywords: Expert and Knowledge-Based Systems, Computational Intelligence, Information Assurance and Intelligence
Abstract: The lateral movement phase in Advanced Persistent Threat (APT) represents a critical stage where attackers navigate internal systems to expand their foothold and escalate privileges. To improve the detection of these threats, this paper presents a process mining-based approach for detecting APTs, focusing specifically on the lateral movement phase. A controlled experimental environment was developed to simulate attack scenarios, enabling the collection and analysis of system and process logs. The results demonstrate the feasibility of using process mining to model attack progression, enhance incident response, and support the early detection of stealthy intrusions in Windows-based environments.

Keywords: Knowledge Acquisition, Big Data Computing,, Machine Learning
Abstract: In this study, we focus on explainable artificial intelligence (XAI), which improves the interpretability of learning models that tend to become black boxes, and propose a method to visualize the factors that affect customer satisfaction. We have developed a method to visualize trends across features based on t-SNE using the contribution of LIME, one of the most representative XAIs, as a variable. By expanding the output results of t-SNE into a bubble chart based on the contribution of variables, it becomes possible to grasp overall trends regarding satisfaction at cram schools, in contrast to LIME, which provides an interpretation for each sample, and to support an understanding of overall needs. We confirmed its effectiveness by applying it to survey data from the Japanese cram school industry, where it is desirable to operate cram schools based on an understanding of the factors that affect the needs of students and parents.

Keywords: Machine Learning, Application of Artificial Intelligence
Abstract: Local Interpretable Model-agnostic Explanations (LIME) is an interpretable method used to explain the predictions of machine learning models. It generates perturbed samples around an instance and fits a simple surrogate model, such as linear regression, to approximate the local behavior of the black-box model. This paper introduces Additive Local Interpretable Model (ALIME), a LIME variant based on generalized additive models (GAMs), where feature effects are modeled by penalized splines (P-splines), providing flexibility for capturing nonlinear relationships. Experimental results

show that ALIME outperforms the original LIME method in terms of local fidelity. In addition, the shape functions generated by ALIME can clearly capture local feature contributions, providing insight into the relationship between features and model outputs, and enhancing overall interpretability.

Keywords: Machine Learning, Optimization and Self-Organization Approaches, AIoT
Abstract: In federated learning, model aggregation is performed at a common server leading to a single point of failure, which aggravates with increasing number of clients causing large network delays that may slow down the entire system. In such cases, hierarchical aggregation is performed with edge devices deployed at different geographical locations. Typically, edge devices employ online learning algorithms for processing the incoming data stream and so, one-way aggregation strategies might end up overwriting the local model updates after the clients receive the aggregated model. Hence, we propose a two-way aggregation strategy wherein models are aggregated both on the client and server. Also, since the incoming distribution may be temporal in nature, we need to dynamically update the aggregation tree. We quantify the non IIDness among different geo-spatialy distributed data-spouts without compromising data privacy. We propose a novel dynamic re-clustering algorithm which is triggered by the intra-cluster and inter-cluster divergence. We implement the above algorithm in an end-to-end FL system that can perform two-step aggregation for aggregation of parent-side and child-side models in the aggregation tree to preserve the model updates in asynchronous execution of the system. Using benchmark datasets, we demonstrate that our proposed framework achieves 95% accuracy, which we plan to further improve.

Keywords: Cloud, IoT, and Robotics Integration
Abstract: The advancement of Industry 4.0 and the rise of Industry 5.0 are transforming human-robot collaboration (HRC) within industrial environments. Addressing the challenges of operator safety, ergonomic health, and production efficiency remains essential for modern manufacturing and logistics systems. This talk offers an overview of cutting-edge control and optimization strategies aimed at improving safety, ergonomics, and the operational performance in collaborative robotics and drone-supported automation. An optimization-based approach for planning time-efficient and safety-compliant trajectories for robotic manipulators is presented, with a particular emphasis on ergonomic factors and compliance with safety standards. Moreover, innovative control frameworks for human-drone collaboration in industrial warehouse settings are explored, where drones support pick-and-delivery operations by adapting their paths in real time to human movement. These methods are validated through realistic use cases, paving the way for safer and more efficient HRC in the industrial landscapes of the future.

Keywords: Manufacturing Automation and Systems, Cyber-physical systems, Distributed Intelligent Systems
Abstract: Scaling is key to economic success—but why does it pose greater challenges in Industrial AI compared to Consumer AI? This presentation examines the key obstacles and explores proven concepts such as assistance systems, foundation models, human-in-the-loop versus closed-loop approaches, and the multi-stage adaptation of models to specific use cases. The question is raised: Is a foundation model for industrial manufacturing applications realistic? Real-world examples of implemented Industrial AI solutions in Digital Manufacturing are presented to demonstrate how these approaches are applied in practice.

Keywords: Machine Vision, Deep Learning, Application of Artificial Intelligence
Abstract: We propose LLM-DETR, an enhanced version of the DETR (DEtection TRansformer) framework that integrates advanced attention mechanisms inspired by large language models (LLMs) for object detection. Applied to the MS COCO 2017 dataset, LLM-DETR demonstrates a 5% increase in average precision (AP) over the original DETR, while exhibiting efficient GPU training. This improvement underscores the potential of LLM-inspired attention mechanisms for advancing object detection accuracy in various domains. The code for our LLM-DETR is publicly available on GitHub: https://github.com/mingruWang/LLM-DETR.

Keywords: Deep Learning, Application of Artificial Intelligence
Abstract: Multimodal sentiment analysis (MSA) integrates and processes data from multiple sources, like audio and text, to better understand human emotions through crossmodal interactions. Previous research generally obtain a feature embedding from different modalities and fuse at the utterance level directly, making it difffcult to capture ffne-grained multimodal representations and susceptible to irrelevant information from heterogeneous modalities. Moreover, these methods often overlook explicit word-level alignment between modalities, which limits effective representation learning and multimodal fusion. Therefore, we propose a novel method, Fine-grained Multimodal Fusion Network(MMTA) for MSA. Speciffcally, a ffne-grained representation alignment module is introduced to extract representation at the word level by montreal forced aligner (MFA).A local-global multi-scale fusion method is then designed to further capture more ffne-grained interaction from local features, i.e., word-level features. Finally, we also investigate the effectiveness of incorporating word-level contextual information into the ffne-grained fusion process. Extensive experiments on the public MSA datasets, i.e., MOSI and MOSEI, show that our approach surpass previous baselines.

Keywords: Deep Learning, Application of Artificial Intelligence, Image Processing and Pattern Recognition
Abstract: In recent years, visible-infrared object detection has achieved significant progress. However, most existing methods primarily emphasize the shared features between the two modalities while overlooking their feature differences. To address this limitation, we propose the Difference-Guided Modality Fusion Network, which can effectively improve the fusion and detection performance of modalities. Specifically, we propose a cross-modal data augmentation strategy to overcome the limitations of single-modality reliance by exchanging the partial modal information. To further capture and analyze feature differences between modalities, we introduce a differential attention fusion approach that models a difference matrix across modal channels, thereby quantifying and strengthening the salient features of the two modalities. Additionally, we develop a modality-aware dynamic learning mechanism that employs a loss function that can simultaneously focus on the differences and common parts of the modalities, guiding the model to adaptively learn features between the modalities. Experimental results on FLIR, LLVIP and M3FD datasets demonstrate the effectiveness of the proposed method, with mAP reaching 42.3%, 67.5% and 59.0% respectively.

Keywords: Deep Learning, Application of Artificial Intelligence, Image Processing and Pattern Recognition
Abstract: Despite the significant advancements made in industrial defect detection, accurately and timely identifying complex and small-sized defects remains a challenge. Most current lightweight defect detectors are unable to fully extract both global and local contextual information due to their simplified network architectures. To address the above issues, this paper introduces a novel real-time detector DSK-YOLO, which efficiently enhance global and local contextual information under lower number of parameters. Specifically, DSK-YOLO comprises two key components: DSKblock and DSKSR. The DSKblock employs dilated separable kernels to expand the effective receptive fields (ERFs) without deep layer stacking, thereby identifying complex defects. For small-sized defects detection, we develop a feature-level super resolution (SR) auxiliary branch to enhance local contextual information in the training phase. Moreover, the train-only SR branch brings no extra computational overhead for inference, making it an impressive choice for real-time tasks. Experimental results demonstrate that, on the industrial datasets NEU-DET and ESD, DSK-YOLO achieves mAP of 45.7% and 64.8%, which is 1.4% and 1.0% higher than the baseline model YOLOv8n. Our proposed DSK-YOLO offers a favorable trade-off between precision and parameters compared to state-of-the-art models.

Keywords: Neural Networks and their Applications, Deep Learning, Machine Learning
Abstract: With the rapid progress of urbanization, city governance faces growing challenges such as traffic violations and environmen-tal pollution. Traditional manual monitoring methods are inef-ficient and costly. To enhance the efficiency of monitoring and managing uncivil behaviors in urban environments, we pro-pose a self-supervised video hashing retrieval framework for uncivil behavior recognition. Leveraging deep learning tech-niques, our method generates compact binary hash codes for both video and text modalities via a contrastive masked auto-encoder (ConMH), enabling efficient large-scale retrieval. We further improve ConMH by introducing cross-attention mech-anisms in the text hashing branch to better handle context de-pendencies. To optimize retrieval results, we integrate five multimodal fusion and ranking strategies, including a novel Hybrid Distance-Rank Fusion method that balances similarity scores and rank information. Experiments conducted on MSRVTT and MSVD datasets demonstrate that our approach achieves superior performance in mAP@K and NDCG metrics. The framework significantly enhances cross-modal semantic coverage, ensures high retrieval precision, and maintains low computational and storage overhead through binary encoding.

Keywords: Deep Learning, Application of Artificial Intelligence, Neural Networks and their Applications
Abstract: There are several options in specifying the action for robot control. Typically, robot motion is the natural choice since it is a readily observable quantity. However, applied force may be more suitable for tasks that involve contact with the environment. Moreover, many tasks that exhaustively interact with the environment typically require a specification of both motion and force simultaneously. The wave variable combines them into one quantity related to the power supplied to the robot. Therefore, wave variable may be used to specify the action. This helps the robot perform interactive tasks better than using motion or force alone. It also helps in creating high quality dataset of task demonstration since human perceives wave reaction as an additional modality to generate the wave action properly.

Keywords: Deep Learning, Application of Artificial Intelligence, Representation Learning
Abstract: With the advancement of autonomous driving technology, trajectory prediction has become a critical task for ensuring traffic safety and intelligent decision-making. Existing motion forecasting models suffer from HD (High-Definition) map dependency, leading to high costs and poor adaptability. Furthermore, their accuracy sharply declines when maps are unavailable, motivating research into map-free alternatives. However, map-free models typically exhibit lower accuracy. To address this issue, we propose a universal enhancement framework that employs trajectory-map contrastive learning, utilizing a trajectory-to-map encoder to extract implicit map representations from raw trajectories, thereby improving performance. Extensive experiments on the Argoverse dataset demonstrate that, after incorporating our trajectory-to-map encoder into map-free models, the average minADE and minFDE are improved by 2.7% and 3.5%, respectively. These results underscore our method's robustness and generalizability in enhancing map-free models, confirming the efficacy of implicit map representation learning and offering a promising solution for HD-map-free autonomous driving in dynamic open-road environments.

Keywords: Deep Learning, Artificial Social Intelligence
Abstract: This paper proposes a method based on the Optimal Transport algorithm and Kolmogorov-Arnold Networks to address the issue of lacking explainability in current legal case matching tasks. The model first employs the Optimal Transport algorithm to construct semantic and legal feature matrices for case pairs, capturing deep semantic relations and legal feature disparities between the cases. Based on the information in these matrices, the model extracts rationales that support the decision-making process. To further enhance the model's ability to capture complex relationships in legal data, we integrate Kolmogorov-Arnold Networks. By decomposing high-dimensional functions into simple univariate mappings, KAN effectively captures the intricate semantic and legal dependencies between case pairs. Experimental results demonstrate that our approach outperforms four baseline methods on the CAIL and ELAM datasets, improving the F1 score by 5% and 12%, respectively, thus validating the effectiveness of our model.

Keywords: Machine Vision, Deep Learning, Image Processing and Pattern Recognition
Abstract: We present SFTNet, a lightweight (7.1–8.5M parameters) deepfake video detection model that fuses spatial and frequency‐domain features to combat advanced face‐swapping manipulations. SFTNet employs EfficientNet and MobileNet backbones to extract rich RGB and spectral representations, which are aggregated via average pooling and modeled temporally with bidirectional LSTMs. This architecture captures subtle inconsistencies across frames, enabling effective discrimination between genuine and synthesized videos.

Designed for on‐device, real‐time operation, SFTNet eliminates reliance on cloud processing, making it suitable for social media, video conferencing, messaging, and financial authentication platforms. Trained and evaluated on FF++ and CelebDF, it achieves 93.44 % accuracy and 96.34 % AUC under diverse test conditions. By delivering low-latency, on-device inference with minimal resource overhead, SFTNet supports ubiquitous deployment, providing a first line of defense against misinformation, scamming, and digital extortion. Its compact, privacy‐preserving design bolsters digital security and media authenticity in resource‐constrained environments.

Keywords: Deep Learning, AI and Applications, Neural Networks and their Applications
Abstract: Depression is one of the most common psychological disorders. In recent years, gait data-based depression recognition methods have drawn a lot of attentions. However, in existing gait-based depression recognition models, the extensive use of dropout leads to non-negligible inconsistencies between the training and inference stages, which may limit further improvements of model performance. In this paper, we propose a multimodal gait-based depression recognition model, which integrates skeleton and silhouette modalities of gait and incorporates consistency training method to address the inconsistencies caused by dropout. Experiments conducted on the D-Gait depression dataset demonstrate that the proposed model with consistency training yields better and more stable performance. Additionally, through extended experiments, we provide substantial insights into the trade-offs between advantages and disadvantages of the consistency training method.

Keywords: Application of Artificial Intelligence, Neural Networks and their Applications
Abstract: As global efforts to explore the Moon intensify, the need for high-quality 3D lunar maps becomes increasingly critical—particularly for long-distance missions such as NASA's proposed Endurance rover, which aims to traverse 2,000 km across the South Pole–Aitken basin. Kaguya TC images, though globally available at 10 m/pixel, suffer from altitude inaccuracies caused by stereo matching errors and JPEG-based compression artifacts. This paper presents a method to improve the quality of 3D maps generated from Kaguya TC images, focusing on mitigating the effects of compression-induced noise in disparity maps. We analyze the compression behavior of Kaguya TC imagery, and identify systematic disparity noise patterns, especially in darker regions. In this paper, we propose a method to enhance 3D map quality by reducing residual noise in disparity images derived from compressed images. Our experimental results show that the proposed method effectively reduces elevation noise, enhancing the safety and reliability of terrain data for future lunar missions.

Keywords: Application of Artificial Intelligence, AI and Applications, Machine Learning
Abstract: In the evolving landscape of software engineering, accurate identification of defect-inducing commits is critical to improving software quality and reducing development costs. This paper revisits the widely adopted SZZ algorithm, which is utilized for labeling commits as clean or defect-inducing, to address one of its main limitations, i.e., its reliance on outdated corrective commits identification strategies. We propose an innovative approach that integrates the semantic understanding capability of the GPT OpenAI model into the SZZ flow to better interpret commit messages. Our experiments reveal, for some projects, a large number of commits incorrectly interpreted as defect-fixing, consequently, leading to the misclassification of commits as defect-inducing. As an example, for the Postgresql dataset, the number of defect-inducing commits was reduced in 21% when compared to the original SZZ. Furthermore, results of our experiments strongly suggest that, as a result of the proposed SZZ labeling process, the JIT-SDP problem has been shown to be more challenging than originally reported by previous works.

Keywords: Application of Artificial Intelligence, AIoT, Deep Learning
Abstract: Deep neural networks (DNNs) have revolutionized multiple generations by harnessing the power of advanced GPUs and extensive datasets. In fields such as audio and video processing, DNN models are deployed on edge servers to achieve millisecond-level latency to meet stringent service-level objectives (SLOs). However, the dynamic and unpredictable nature of real-world applications, characterized by erratic surges in inference requests, poses significant challenges to existing edge systems. To address these challenges, this study presents a novel scheduling algorithm that utilizes deep reinforcement learning to maximize the minimum margins of all GPUs.This strategic approach significantly enhances the system’s capacity to manage unexpected high-demand tasks, thereby improving resilience and overall response capabilities under diverse fluctuating workloads. The experimental results confirm that the proposed algorithm consistently outperforms traditional algorithms, demonstrating superior performance in task completion rate, resource utilization efficiency, robustness, and responsiveness across different load conditions.

Keywords: Application of Artificial Intelligence, Biometric Systems and Bioinformatics, Deep Learning
Abstract: Cardiovascular disease (CVD) remains one of the leading global health challenges, calling for accurate and scalable detection methods. This work presents MoViE, a multimodal fusion model based on sparsely activated experts, designed to effectively fuse electrocardiogram (ECG) and electronic health record (EHR) data for CVD detection. MoViE adopts a dual-branch architecture. In the ECG branch, the feed-forward layers of the Vision Transformer (ViT) are replaced with a Mixture-of-Experts (MoE) module that incorporates top-k routing and shared experts, enabling the model to capture both fine-grained waveform features and long-range temporal dependencies. The EHR branch encodes structured clinical records using TF-IDF, followed by a multilayer perceptron (MLP) to extract semantic representations. To support modality interaction, a MoE-based fusion module is introduced, where a gating network adaptively selects experts to combine complementary features. Extensive experiments on the MIMIC-IV-ECG dataset for myocardial infarction detection demonstrate that MoViE outperforms existing unimodal and multimodal mainstream baselines, achieving 87.98% Accuracy, 90.70% AUC, 66.15% F1-score, and 59.12% MCC, highlighting the potential of MoViE as a general framework for intelligent disease detection.

Keywords: Application of Artificial Intelligence, Cloud, IoT, and Robotics Integration, AIoT
Abstract: We investigate the problem of path planning and task offloading for UAV clusters in a UAV-assisted edge computing scenario. UAVs autonomously make decisions regarding path planning, continuous service provision, and task offloading based on collected information. In this setting, terminal equipment (TE) cannot directly connect to servers; thus, UAVs act as both edge servers and communication relays, proactively providing services to TEs.We construct a fine-grained temporal scale model that decomposes UAV actions into atomic time units, transforming decision-making on specific behaviors into state transition decisions. This approach better accommodates the needs of time-varying environments. Regarding path planning, given the time-sensitive nature of TE requests, we focus on how to provide stable and timely computational services to TEs. We propose a multi-agent reinforcement learning algorithm capable of dynamically sensing task priorities to enhance Quality of Service (QoS), with optimizations made in terms of task completion rate, UAV energy consumption, and processing delay.About task offloading decision, we introduce a dual-keyword-based offloading algorithm to optimize the binary offloading process. Finally, we conduct simulation experiments to demonstrate the effectiveness of the proposed algorithms, and comparative experiments confirm their superiority.

Keywords: Application of Artificial Intelligence, Deep Learning
Abstract: Deep learning’s remarkable performance in natural language processing (NLP) faces critical interpretability challenges, particularly in high-stakes domains like healthcare and finance where model transparency is essential. While concept bottleneck models (CBMs) have enhanced interpretability in computer vision by linking predictions to human-understandable concepts, their adaptation to NLP remains understudied with persistent limitations. Existing approaches either enforce rigid binary concept activations that degrade textual representation quality or obscure semantic interpretability through latent concept embeddings, while failing to capture dynamic concept interactions crucial for understanding linguistic nuances like negation or contextual modification. This paper proposes the Concept Language Model Network (CLMN), a novel neural-symbolic framework that reconciles performance and interpretability through continuous concept embeddings enhanced by fuzzy logic-based reasoning. CLMN addresses the information loss in traditional CBMs by projecting concepts into an interpretable embedding space while preserving human-readable semantics, and introduces adaptive concept interaction modeling through learnable neural-symbolic rules that explicitly represent how concepts influence each other and final predictions. By supplementing original text features with concept-aware representations and enabling automatic derivation of interpretable logic rules, our framework achieves superior performance on multiple NLP benchmarks while providing transparent explanations. Extensive experiments across various pre-trained language models and datasets demonstrate that CLMN outperforms existing concept-based methods in both accuracy and explanation quality, establishing a new paradigm for developing high-performance yet interpretable NLP systems through synergistic integration of neural representations and symbolic reasoning in a unified concept space.

Keywords: Application of Artificial Intelligence, Deep Learning
Abstract: In the domain of dense retrieval, training with hard negatives is widely used. However, the existence of a large number of false negatives has given rise to some hard-to-train queries, which often have a significant impact on the loss at the same time. The training process overemphasizes these queries that account for a disproportionately high proportion in the training loss while neglecting the positive examples with moderate rankings, thus resulting in suboptimal model outcomes. We theoretically analyze the excessive influence of hard-to-train queries and introduce an in-batch loss weighting algorithm that adaptively assigns weights based on the in-batch ranking position of its positive document. Specifically, our method will calculate the in-batch ranking positions of the similarity between positive examples and queries, construct query weights based on the Gaussian distribution, and balance the attention in the training process. This strategy mitigates the effects of false negatives while retaining the benefits of hard negatives. Experiments on the MS-MARCO dataset show statistically significant improvements: a 1.3% increase in MRR@10, higher Recall@1000, and consistent NDCG@10 gains on the TREC test set across diverse hard negative sources. The relevant code has been open-sourced on the GitHub repository: https://github.com/TRcreeper/IBRWLoss .

Keywords: Application of Artificial Intelligence, Deep Learning, Artificial Social Intelligence
Abstract: 变量中的无监督异常检测 时间序列在 工业监控和 IoT 设备管理。然而 现有方法在以下方面仍面临重大挑战 对复杂的时间模式进行建模。虽然 变压器 模型在逐点 表示学习，他们的注意力机制 主要关注时间步长之间的直接依赖关系， 使得在 Segment 级别。为了解决这些问题，我们建议 PETformer，原型增强的多尺度无监督 Transformer 架构。PETformer 引入了 多头尺度卷积注意力 （MHSCA） 模块， 它采用并行卷积核来提取 多尺度功能。此设计增强了模型的 能够捕获本地和全局依赖项。 此外，它还包含多尺度余弦原型 （MSCP） 作为与输入交互的电感偏置 序列，加强法向模式的先验建模 并改进模型的 对潜在的模式偏差敏感。实验的 结果表明，PETformer 的性能始终优于 PETformer 现有最先进的无

Keywords: Application of Artificial Intelligence, Deep Learning, Expert and Knowledge-Based Systems
Abstract: Abstract—Recommender systems based on collaborative filtering (CF) effectively model user-item interactions but struggle with data sparsity and cold-start issues. On the other hand, large language models (LLMs) offer strong semantic understanding but fail to capture structured user-item relationships.To bridge this gap, we propose AlignGenRec, a framework that integrates collaborative and textual knowledge for generative recommendation. Specifically, we introduce an embedding alignment mechanism that aligns item embeddings from a pre-trained CF model with text embeddings from item descriptions. These aligned embeddings are then transferred to the LLM without requiring fine-tuning, enabling structured knowledge integration while preserving generative capabilities.Additionally, constrained sequence decoding ensures that generated recommendations correspond to valid items, improving recommendation accuracy. Experimental results demonstrate that AlignGenRec outperforms both CF-based and LLM-based baselines, particularly in cold-start scenarios, with performance improvements of 7 - 11% across three benchmark datasets. Beyond recommendation tasks, AlignGenRec also supports preference prediction and user profiling, highlighting its versatility in real-world applications. Index Terms—Recommendation Systems, Collaborative Filtering, Large Language Models, Generative Recommendation.

Keywords: Deep Learning, Image Processing and Pattern Recognition, Application of Artificial Intelligence
Abstract: Abstract— Lane detection plays a crucial role in autonomous driving. Though modern anchor-based deep lane detection methods have demonstrated remarkable performance on stan- dard benchmarks, they continue to struggle with complex topological variations. In this work, we propose the Dual Weighting Attention Feature Fusion Network(DW AFFNet), which enhances detection performance through two key innova- tions: (1) hierarchical feature fusion for improved location and (2) anchor’s IoU classification score consistency optimisation. In computer vision, shallow low-level features provide precise spatial localization while deep high-level features capture es- sential global information. Therefore, complementary global- local representations for enhanced lane detection accuracy are established via Iterative Coordinate Attention Feature Fusion (ICAFF) module, which systematically combines these hierar- chical features through coordinate-sensitive attention mecha- nisms. Furthermore, we introduce the Dual Weighting Label Assignment Scheme (DW Scheme) to align IoU and classifi- cation scores through importance-aware dynamic weighting, significantly improving sample discrimination capability. We evaluate our method on two benchmarks of lane detection and the results demonstrate its effectiveness. Our method surpasses baseline on CULane. On CULane, it obtains 56.45 mF1 with 64.21/55.76/22.16 F1@75/80/90 scores, outperforming CLRNet by 1.52%/2.27%/3.06%/7.4% respectively. Significant improvements are observed across most scenarios in complex road conditions.

Keywords: Machine Vision, Image Processing and Pattern Recognition, Multimedia Computation
Abstract: Open-world general object counting is a critical task in computer vision, with applications in image understanding, environmental monitoring, and surveillance. Traditional methods rely on large-scale annotated data, which is expensive and time-consuming to obtain, and often fail to manage diverse object categories and complex scenes. To address this, we propose AugCount, a novel framework that enhances open-world object counting by generating high-quality, diverse synthetic data during testing. For the first time, we employ a diffusion model to produce conditional images based on density maps specifying object locations for general object counting. Our framework is highly versatile and adaptable to various counting tasks. Experiments demonstrate that AugCount significantly improves performance on benchmark datasets like FSC-147, reducing the average counting error by 1 per image and achieving a state-of-the-art MAE of 4.70. AugCount effectively addresses data scarcity and model generalization challenges, offering enhanced robustness and adaptability for practical counting systems.

Keywords: Machine Vision, Machine Learning, AI and Applications
Abstract: Recent years have seen the rise of many competent pose estimation models. Unfortunately, many of these still rely on manually labelled training data. With many arguing the importance of larger and more diverse pools of training data, this study proposes the use of virtually created hand images for pose estimation training. Synthetically generating virtual pose data provides a faster and more accurate method for creating labelled training data, due to the tedious and taxing nature of manual labelling. This study looks to employ 3D modelling software to simulate and record randomised, realistic, human hand postures. For this, virtual hands were recorded with ever-changing joint angles and randomised backgrounds. This created data was used to retrain an existing YOLOv11 pose estimation network. The model trained with solely synthetic data was able to detect hands in real-world images, though it scored a low accuracy for these images. Injecting an existing dataset with these synthetic data was able to provide a significantly reduced error in the estimations, showcasing a 54.0% increase in accuracy — compared to a model trained without the addition of these data. The results achieved show promise for the utilisation of synthetic data in the training of pose estimation models.

Keywords: Machine Vision, Machine Learning, Application of Artificial Intelligence
Abstract: Eye state detection plays a vital role in safety-critical applications such as driver monitoring systems and medical diagnostic or therapeutic devices. This study explores ultra-lightweight convolutional neural network architectures tailored for real-time deployment on resource-constrained embedded platforms, where power and processing capabilities are severely limited.

Unlike traditional approaches that rely on complex models or post-processing techniques, our solution emphasizes low-latency inference while maintaining high classification accuracy. To achieve this, we combine efficient eye region extraction based on facial landmarks with a custom-designed CNN featuring just over 260,000 parameters. The model is evaluated on the MRL Eye State dataset and demonstrates competitive results, surpassing existing architectures in terms of inference speed and model size.

Our implementation is validated on a Raspberry Pi 5 using the Camera Module 3, highlighting the system’s practical applicability in embedded environments. This research is part of the DeepLightAI project, where continuous eye monitoring is required to ensure patient safety under high-intensity medical lighting conditions.

The results show that lightweight models can effectively bridge the gap between performance and efficiency, offering a viable path for real-world, real-time eye state detection in embedded applications.

Keywords: Machine Vision, Machine Learning, Application of Artificial Intelligence
Abstract: 在对象检测领域，特征金字塔网络 在对象检测中得到广泛采用 算法，因为它简单、高效且健壮 特征生成功能。尽管有其优点，但 特征金字塔网络在其 建筑设计。在本文中，我们 旨在剖析 对 Pyramid 网络进行特征化并引入新网络 架构，称为 Hybrid-Residual Feature Pyramid 网络 （HR-FPN），旨在有效缓解 这些已确定的问题。HR-FPN 主要由 两个关键模块：Hybrid-Operation Module 和 残差特征增强模块。混合运营 Module 有效地整合了来自 high-level 功能转换为 low-level 功能，而 残差特征增强模块缓解信息 最高金字塔层要素映射中的损失 提取尺度不变的上下文信息。然后，我们 设计了稀疏 R-CNN 的增强版本，称为 特定提案特征 R-CNN（SPF R-CNN），其中 整合了 Learnable Proposal Classification 功能，并且 可学习的提案回归功能，可有效 缓解 分类和本ࢸ

Keywords: Machine Vision, Machine Learning, Deep Learning
Abstract: In recent years, Transformer-based methods have remained the dominant approach for 3D human pose estimation (3D HPE). While these methods have achieved continuous improvements in accuracy, the inherently high computational complexity of Transformer modules makes current 3D HPE models excessively resource-intensive, rendering them impractical for deployment on resource-constrained devices. To address this challenge, we propose Token Pruning-based Spatio-Temporal Topology Transformer (TTTFormer), which consists of a Spatio-Temporal Topology Encoder (STTE) and a Token Pruning-based Spatio-Temporal Decoder. The STTE incorporates Spatial Position Embedding (SPE) to model spatial correlations between joints and a Temporal Topology Block (TTB) to effectively capture the temporal relationships of the same joint across frames. These components enable the STTE to extract richer spatio-temporal features. Furthermore, to mitigate the high computational cost of transformers when handling long sequences, we introduce a lightweight design in the Token Pruning-based Spatio-Temporal Decoder, which preserves representative tokens while discarding redundant ones that contribute disproportionately to computational overhead. This approach achieves a balance between efficiency and accuracy in complex 3D HPE tasks. Extensive experiments on the Human3.6M dataset demonstrate that our model significantly enhances both efficiency and precision. The results show that our model achieves a MPJPE of 40.8mm and better performance with 364M FLOPs per frame.

Keywords: Machine Vision, Neural Networks and their Applications, AI and Applications
Abstract: Monitoring the slagging status during the tapping process of a converter is an important task in the metallurgical process. Currently, the monitoring of the slagging process during converter tapping mainly relies on manual visual inspection methods, which suffer from problems such as low efficiency, strong subjectivity, and low accuracy. Fortunately, monitoring of the slagging during converter tapping is essentially an object detection task. Considering the harsh scenarios and tiny targets, the object detection model for this task should be carefully designed. To address this issue, this paper proposes an efficient object detection model named ESD-YOLO based on an improved YOLO algorithm. ESD-YOLO aims to enhance detection accuracy and real-time performance, especially for small targets such as minor slag streams in complex scenarios. It integrates feature guidance from DINOv2, to strengthen fundamental feature learning capabilities, while introducing an efficient channel attention module to optimize multi-scale feature information fusion. Additionally, the NWD loss function is employed to further elevate focus on small target features. Experimental results demonstrate that ESD-YOLO achieves 96.8% mAP and 92.6 FPS on an actual converter slagging dataset, outperforming mainstream object detection models and reaching state-of-the-art performance.

Keywords: Transfer Learning, Machine Vision, Deep Learning
Abstract: Test-time adaptation (TTA) aims to improve the robustness of segmentation models to an unlabeled target domain without requiring access to the source data. While existing TTA methods have achieved promising results on image classification, they often fail to translate effectively to semantic segmentation due to the spatial complexity and fine-grained nature of dense predictions. We propose PromptCAL, a lightweight and effective TTA framework tailored for semantic segmentation, built upon the SegFormer architecture. Our method addresses two central challenges: (1) Which model component to adapt remains underexplored. Using Grad-CAM visualization and sensitivity analysis, we identify Stage 2 of the transformer backbone as the most domain-sensitive and restrict adaptation to this stage. (2) How to identify reliable supervision during adaptation is critical. We introduce a confidence-aware self-training mechanism based on per-pixel entropy filtering to guide pixel selection for model adaptation, ensuring label quality and model transferability. In addition, we incorporate lightweight prompt injection to enhance the adaptability of mid-level features. Our method achieves competitive improvements over the state-of-the-art while maintaining high adaptation efficiency and significantly reducing runtime overhead. Extensive experiments on corrupted semantic segmentation benchmarks, including ACDC (A-fog, A-night, A-rain, and A-snow), Cityscapes-foggy (CS-fog) and Cityscapes-rainy (CS-rain) demonstrate that PromptCAL achieves comparable or superior accuracy to state-of-the-art TTA baselines, while reducing adaptation time by over 50% per domain. This makes it a practical solution for efficient TTA in smart cities and edge-deployed vision systems. The source code is available at https://github.com/ml4papers/PromptCAL.

Keywords: Neural Networks and their Applications, Machine Vision, Application of Artificial Intelligence
Abstract: lex deep networks, such as transformers, have shown better dehazing performance. However, their practicability is restricted by low adaptability to industrial scenarios. It is necessary to realize an effective and lightweight model that can be applied in industrial sites. For such a purpose, an effective image dehazing model named MFDehaz-Net is proposed. MFDehaz-Net has two specially designed components, i.e., Multi-scale Fusion Block and Point-Depth wise Block, which help it achieve a deep fusion of image features of different scales to obtain better global understanding. Besides, to reduce the damage to the original color of the image caused by image dehazing, MFDehaz-Net integrates the supervision signal in the frequency domain with a specially designed loss function. Experimental results demonstrate that MFDehaz-Net outperforms SOTA models in terms of dehazing ability with much shorter inference time.

Keywords: Machine Vision, Application of Artificial Intelligence, Computational Intelligence
Abstract: Zero-shot visual question answering (VQA) poses a formidable challenge at the intersection of computer vision and natural language processing. Traditionally, this problem has been tackled using end-to-end pre-trained vision-language models (VLMs). However, recent advancements in large language models (LLMs) demonstrate their exceptional reasoning and comprehension abilities, making them valuable assets in multi-modal tasks, including zero-shot VQA. LLMs have been previously integrated with VLMs to solve zero-shot VQA in a conversation-based approach. However, while the focus in VQA tasks is often on specific regions rather than the entire image, this aspect has been overlooked in previous approaches. Consequently, the overall performance of the framework relies on the ability of the pre-trained VLM to locate the region of interest that is relevant to the requested visual information within the entire image. To address this challenge, this paper proposes Grounded Multi-modal Conversation for Zero-shot Visual Question Answering (GMC-VQA), a region-based framework that leverages the complementary strengths of LLMs and VLMs in a conversation-based approach. We employ a grounding mechanism to refine visual focus according to the semantics of the question and foster collaborative interaction between VLM and LLM, effectively bridging the gap between visual and textual modalities and enhancing comprehension and response generation for visual queries. We evaluate GMC-VQA across three diverse VQA datasets, achieving substantial average improvements of 10.04% over end-to-end VLMs and 2.52% over the state-of-the-art VLM-LLM communication-based framework, respectively. Our code is publicly available at https://github.com/mrzarei5/GMC-VQA.

Keywords: Machine Vision
Abstract: Dynamic gesture recognition faces critical challenges due to complex backgrounds, multi-scale feature variations, and inefficient multi-modal fusion. To address these challenges, this study proposes the Wavelet-Enhanced Multiscale Transformer (WEMT), a unified framework that integrates three key innovations: wavelet-based featuredecomposition for noise-robust detail enhancement, hierarchicalattention mechanisms for global-local spatiotemporalmodeling, and confidence-driven adaptive fusion for heterogeneous modality integration. The wavelet domain module selectively amplifies discriminative gesture features through frequency-domain analysis and multi-scale pooling,while the pyramid-structured attention heads capture both coarse-grained postures and fine-grained motions. The adaptive fusion framework dynamically prioritizes complementary modalities through entropy-based gating, ensuring robustness in diverse environments. Evaluations on NVGesture and Briareo datasets demonstrate state-of-the-art performance across single-modal and multi-modal configurations, with notable advantages in low-light scenarios. The model also maintains computational efficiency, making it suitable for real-world human-computer interaction applications. This work provides a systematic and scalable solution for accurate dynamic gesture recognition, bridging the gap between theoretical advances and practical deployment.

Keywords: Decision Support Systems
Abstract: In Granular Computing, a key approach has been the handling of incomplete fuzzy preference relations through an allocation of information granularity and its corresponding optimization procedure. This methodology enables the transformation of numerical models into granular versions, offering a more accurate representation of reality by recovering missing information. In decision-making contexts involving fuzzy preference relations, it has played a crucial role in advancing procedures for estimating incomplete information. However, although several granularity allocation protocols have been proposed, only one has been employed so far: the one based on a uniform and symmetric allocation of information granularity. To address this gap, we aim to determine the efficacy of existing protocols for allocating information granularity in estimating missing values of incomplete fuzzy preference relations. Numerical tests are presented to demonstrate the effectiveness of each protocol.

Keywords: Decision Support Systems
Abstract: Psychological biases, such as confirmation bias, make individuals particularly vulnerable to believing and spreading fake news on social media, leading to significant consequences in domains such as public health and politics. Machine learning–based fact-checking systems have been widely studied to mitigate this problem. Among them, ensemble methods are particularly effective in combining multiple classifiers to improve robustness. However, their performance heavily depends on the diversity of the constituent classifiers—selecting genuinely diverse models remains a key challenge, especially when models tend to learn redundant patterns. In this work, we propose a novel automatic classifier selection approach that prioritizes diversity, also extended by performance. The method first computes pairwise diversity between classifiers and applies hierarchical clustering to organize them into groups at different levels of granularity. A HierarchySelect then explores these hierarchical levels to select one pool of classifiers per level, each representing a distinct intra-pool diversity. The most diverse pool is identified and selected for ensemble construction from these. The selection process incorporates an evaluation metric reflecting each classifier’s performance to ensure the ensemble also generalises well. We conduct experiments with 40 heterogeneous classifiers across six datasets from different application domains and with varying numbers of classes. Our method is compared against the Elbow heuristic and state-of-the-art baselines. Results show that our approach achieves the highest accuracy on three of six datasets. The implementation details are available on the project’s repository: https://github.com/SaraBCoutinho/HSFN.

Keywords: Decision Support Systems, Discrete Event Systems, Enterprise Information Systems
Abstract: The rapid dissemination of multimodal fake news on social media platforms poses a significant challenge, as deceptive narratives often combine text and images to mislead users. To address this, we propose MEVE-FN, a novel fake news detection framework built upon a Mixture-of-Experts (MoE) architecture. Our model leverages modality-specific vision and language experts to extract specialized features. These features are then dynamically integrated using a Laplace Gating mechanism for adaptive fusion and a MEVE-Adapter module that enhances deep cross-modal interaction. Comprehensive experiments on three public benchmarks (Weibo, Weibo21, and Twitter) demonstrate that MEVE-FN consistently outperforms strong baseline models, achieving notable improvements in detection accuracy and robustness. Furthermore, ablation studies validate the effectiveness of our proposed components, confirming the critical contribution of both the Laplace gating and the MEVE-Adapter to the model's superior performance.

Keywords: Decision Support Systems, Enterprise Information Systems, Service Systems and Organizations
Abstract: In light of the increasing recognition of environmental and social challenges, technology-driven corporations often face significant barriers in aligning their innovation-driven goals with sustainable practices. Despite their capacity for rapid technological advancements, these companies struggle to integrate sustainability objectives into their core strategies due to competing priorities and a lack of cohesive frameworks. This study explores Knowledge Management's (KM) capacity to facilitate and implement sustainable practices. We use the Rapid Review methodology to systematically review the literature and achieve this objective. A total of 35 articles were analyzed, examining factors related to organizational knowledge, the adoption of KM within technology-driven companies, and effective sustainable practices. The results highlight the importance of KM in facilitating and implementing sustainable practices, showing that technology-driven companies that invest in organizing their knowledge can improve their effectiveness in implementing sustainable initiatives, increase their resilience and competitiveness, and strengthen trust with stakeholders.

Keywords: Decision Support Systems, Infrastructure Systems and Services, Distributed Intelligent Systems
Abstract: Deploying AI services on battery-powered mobility platforms such as autonomous vehicles, mobile robots, and large scale IoT sensor networks requires determining the most suitable execution environment for each workload across the cloud, edge, and device computing options. Because every placement option imposes different trade‑offs among Accuracy, Latency, and Energy efficiency (ALE), stakeholders face a difficult, mission‑critical decision that existing studies seldom address in a holistic, mission‑aware manner. To fill this gap, we introduce the Mission‑driven ALE (MALE) evaluation method. MALE couples ALE metrics with explicit mission objectives by allowing analysts to apply customizable weights to each criterion. The evaluation results are aggregated and visualized as heatmaps, helping transform a previously heuristic and opaque placement decision (black-box) into a more transparent and interpretable process (white-box). We examine the applicability of MALE through three representative case studies: Autonomous Vehicles, Real-Time Robotics, and IoT Sensor Networks, each reflecting distinct ALE priorities. By supplying a structured, mission‑aware decision‑support method, MALE strengthens stakeholder confidence and accelerates the optimization of AI service placement across the cloud–edge–device continuum, offering a practical foundation for future validation in real‑world deployments.

Keywords: Decision Support Systems, Modeling of Autonomous Systems, Infrastructure Systems and Services
Abstract: In Vehicular Cloud Computing (VCC), the volatility of mobile resources presents a fundamental challenge: will a task be successfully delivered or ultimately discarded? This paper addresses this “discard or deliver” dilemma by proposing a Stochastic Petri-Net (SPN) model for a Static Vehicular Cloud (SVC) in the context of a car rental business. The model captures key dynamics, including task queuing, resource allocation, vehicle unavailability due to rentals or failures, and broker availability. System performance and availability are evaluated through five core metrics: Queue Discard Probability (QDP), Utilization of Available Resources (UAR), Unavailability (UA), Global Capacity-Oriented Availability (GCOA), and Internal Capacity-Oriented Availability (ICOA). To establish model correctness, its queueing behavior is validated against a classical M/M/m/k model, showing close agreement through experiments. Sensitivity analysis reveals that rental times, task arrival and task service times, and broker times have the greatest impact on the metrics. Case studies reveal key trade-offs between task loss, resource utilization, and availability, offering practical guidance for network planners seeking to deploy effective SVC infrastructures.

Keywords: Decision Support Systems, Service Systems and Organizations
Abstract: In situations of overcrowding of hospitals, a fast and effective evaluation of the severity of patient illnesses is necessary to anticipate the allocation of limited resources. Patient triage provides a first assessment of patient acuity before admission to the Emergency Department (ED) but requires a complex management of medical data performed by specially trained staff. This need to anticipate resource allocation is further amplified when a patient must be admitted for hospitalization after a stay in the ED. This study presents a system for the automation of patient triage in the ED and the prediction of post-ED hospitalizations. A Large Language Model (LLM)-enhanced expert system architecture is designed for the assessment of patient acuity through the processing of rules expressed in natural language and converted into structured rules. The triage evaluation is reused by Machine Learning classifiers, based on eXtreme Gradient Boosting, Random Forest, Logistic Regression, and Decision Tree, predicting, based on triage/ED data and medical history, whether patients require observation, hospitalization, or can be safely discharged after their stay in the ED. Experiments underscore a high accuracy (between 72% and 82%) for best LLMs experimented while eXtreme Gradient Boosting classifier obtains the best performance in the prediction of post-ED orientation with an accuracy between 59.1% and 73.5% depending on the scenario. The proposed architecture stands out for its adaptability to other problems.

Keywords: Decision Support Systems, Smart Sensor Networks, Manufacturing Automation and Systems
Abstract: The Lightweight On-line Detector of Anomalies (LODA) is one of the best performing generic unsupervised anomaly detection algorithms. This method returns an anomaly score using an ensemble of one-dimensional histograms in a random projected space. Deploying LODA in a streaming, dynamically evolving environment remains a challenging task as the method proposed by Pevn´y (2016) tends to forget the learned behaviors too quickly, meaning the model will not recognize reemerging normal behavior and thus often produces high anomaly scores for normal samples.

In this paper, we propose a new algorithm, demonstrating that the proposed modifications lead to improved performance on benchmark datasets, facilitate smoother transitions during concept drifts, and provide more robust and significantly better anomaly scoring on sensor data and in streaming, dynamic environments.

Keywords: Decision Support Systems, System Modeling and Control, Distributed Intelligent Systems
Abstract: This paper introduces the Trend-Enhanced CNN Adversarial Autoencoder (TE-CNN-AAE), a novel approach for generating low-dimensional representations of intraday stock market activity from 5-minute interval quotes. The proposed model extends traditional autoencoders by incorporating an adversarial component that enhances the quality of embeddings by reconstructing input sequences while simultaneously estimating market trends. Using five years of intraday data from Dow Jones Industrial Average (DJIA) assets, we demonstrate that TE-CNN-AAE outperforms baseline methods in capturing intraday patterns and predicting daily price movements. Qualitative evaluations via UMAP visualizations show improved class separability in the latent space, further confirmed by the Silhouette Score and Davies-Bouldin Index, while quantitative analysis using an LSTM classifier validates the superior predictive utility of the embeddings. Ablation tests confirm the importance of the adversarial discriminator in generating robust representations. Our results suggest that TE-CNN-AAE effectively captures the complex dynamics of financial time series and holds the potential for improving decision-support systems in trading.

Keywords: Enterprise Information Systems, Discrete Event Systems, Decision Support Systems
Abstract: Background: The generation of synthetic event logs is fundamental to the advancement and evaluation of process mining techniques. However, synthetic log generation in the context of multi-perspective declarative modeling faces challenges due to its flexible nature, where any behavior that is not constrained by rules is allowed. In addition, the synthetic event log needs to simultaneously meet process, data access and privacy constraints. Objective: This study presents a general-purpose, configurable, multi-perspective log generator capable of simultaneously ensuring compliance with the rules of the flow, data access and privacy perspectives in a flexible context such as that found in declarative processes. Method: A multi-perspective synthetic log generator was developed, capable of integrating declarative process models, data access and organizational structure. The tool was structured in three stages: processing the input models, generating the event log with Declare4Py, and generating the access log respecting the restrictions. Results: Two experiments were carried out with 1,000 and 2,000 cases respectively. The logs generated respected all the rules of conformity between activities, accesses and permissions. There was a linear increase in file size and a moderate increase in execution time, compatible with the complexity of the models. Conclusions: The proposed tool proved effective in generating coherent synthetic logs with multiple perspectives, supporting research into process mining and the evaluation of compliance algorithms. It is a relevant solution for contexts where real data is scarce or sensitive

Keywords: Image Processing and Pattern Recognition
Abstract: Image super-resolution (SR) aims to generate a high-resolution (HR) image from a low-resolution (LR) input. Traditionally, full-reference image quality assessment (FR-IQA) models have been widely used to evaluate the perceptual quality of super-resolved images, relying on pristine reference images as the gold standard. However, in real-world SR applications, such reference images are often unavailable, posing challenges for the use of FR-IQA. While blind image quality assessment (BIQA) models can assess the perceptual quality of super-resolved images without requiring a reference, there remains a lack of comprehensive studies evaluating the effectiveness of existing BIQA models for real-world SR tasks. This dilemma can largely be attributed to the high cost of subjective testing required to collect sufficient human quality annotations, which in turn hinders the development of effective SR-IQA models. In this study, we tackle this challenge with a data-efficient approach. We first generate super-resolved images from LR inputs using state-of-the-art real-world SR methods. Then, we use the maximum differentiation competition (MAD) to select a diverse set of images for subjective testing, allowing us to efficiently gather human preferences and assess the alignment between BIQA model predictions and human judgments. The resulting global ranking of SR methods not only indicates the relative performance of recent real-world SR models, but also gives us an opportunity to develop a new BIQA model tailored for real-world SR-IQA. By utilizing the global rankings of SR algorithms as prior knowledge, we can refine pretrained BIQA models using vast amounts of super-resolved images textbf{without} any supervisory signal. Experimental results show that our approach substantially enhances IQA performance for real-world SR while preserving robust predictive accuracy across various distortion scenarios. The dataset and the code are available at url{https://github.com/cschenjunlin/SR-IQA-SMC25}.

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Automatic MRI brain tumor classification plays an important role in the clinical diagnosis of brain tumors. Meanwhile, deep learning methods recently attract wide attention in this medical application. Different from existing methods that are mainly based on CNN and Transformer architectures, this paper attempts to explore Mamba-related models for MRI brain tumor classification, motivated by the recent success of state space models in capturing global features of medical images. More specifically, this paper introduces a novel vision prompt Mamba model, i.e., VPT-Mamba, for MRI brain tumor classification, which implants a visual prompt into the improved Vision Mamba to capture global and local important information of brain images. The VPT-Mamba utilizes the Mamba® as the backbone to mitigate feature artifacts of brain tumors by introducing register tokens into Vision Mamba, and it further incorporates the visual prompt tuning in input features to enhance feature representation. We evaluate the VPT-Mamba on two public and private brain tumor classification datasets. Our VPT-Mamba achieves the best ACC and AUC results of 99.510% and 99.966% on the public dataset, and the corresponding values are 96.582% and 99.125% on the private brain tumor dataset, respectively. Moreover, comparative experimental results demonstrate the effectiveness of VPT-Mamba for this medical image classification application.

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Stable diffusion is an outstanding image generation model for text-to-image, but its time-consuming generation process remains a challenge due to the quadratic complexity of attention operations. Recent token merging methods improve efficiency by reducing the number of tokens during attention operations, but often overlook the characteristics of attention-based image generation models, limiting their effectiveness. In this paper, we propose local representative token guided merging (ReToM), a novel token merging strategy applicable to any attention mechanism in image generation. To merge tokens based on various contextual information, ReToM defines local boundaries as windows within attention inputs and adjusts window sizes. Furthermore, we introduce a representative token, which represents the most representative token per window by computing similarity at a specific timestep and selecting the token with the highest average similarity. This approach preserves the most salient local features while minimizing computational overhead. Experimental results show that ReToM achieves a 6.2% improvement in FID and higher CLIP scores compared to the baseline, while maintaining comparable inference time. We empirically demonstrate that ReToM is effective in balancing visual quality and computational efficiency.

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Video question answering (VQA) is a multimodal task that requires the interpretation of a video to answer a given question. Existing VQA methods primarily utilize question and answer (Q&A) pairs to learn the spatio-temporal characteristics of video content. However, these annotations are typically event-centric, which is not enough to capture the broader context of each video. The absence of essential details such as object types, spatial layouts, and descriptive attributes restricts the model to learning only a fragmented scene representation. This issue limits the model's capacity for generalization and higher-level reasoning. In this paper, we propose a fundamental question generation with the integration of question embeddings for video question answering (FIQ), a novel approach designed to strengthen the reasoning ability of the model by enhancing the fundamental understanding of videos. FIQ generates Q&A pairs based on descriptions extracted from videos, enriching the training data with fundamental scene information. Generated Q&A pairs enable the model to understand the primary context, leading to enhanced generalizability and reasoning ability. Furthermore, we incorporate a VQ-CAlign module that assists task-specific question embeddings with visual features, ensuring that essential domain-specific details are preserved to increase the adaptability of downstream tasks. Experiments on SUTD-TrafficQA demonstrate that our FIQ achieves state-of-the-art performance compared to existing baseline methods. Code is available at https://github.com/juyoungohjulie/FIQ

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Recently, personalized portrait generation with a text-to-image diffusion model has significantly advanced with Textual Inversion, emerging as a promising approach for creating high-fidelity personalized images. Despite its potential, current Textual Inversion methods struggle to maintain consistent facial identity due to semantic misalignments between textual and visual embedding spaces regarding identity. We introduce ID-EA, a novel framework that guides text embeddings to align with visual identity embeddings, thereby improving identity preservation in a personalized generation. ID-EA comprises two key components: the ID-driven Enhancer (ID-Enhancer) and the ID-conditioned Adapter (ID-Adapter). First, the ID-Enhancer integrates identity embeddings with a textual ID anchor, refining visual identity embeddings derived from a face recognition model using representative text embeddings. Then, the ID-Adapter leverages the identity-enhanced embedding to adapt the text condition, ensuring identity preservation by adjusting the cross-attention module in the pre-trained UNet model. This process encourages the text features to find the most related visual clues across the foreground snippets. Extensive quantitative and qualitative evaluations demonstrate that ID-EA substantially outperforms state-of-the-art methods in identity preservation metrics while achieving remarkable computational efficiency, generating personalized portraits approximately 15 times faster than existing approaches.

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Composed Image Retrieval (CIR) is the task of retrieving a target image from a gallery using a composed query consisting of a reference image and a modification text. Among various CIR approaches, training-free zero-shot methods based on pre-trained models are cost-effective but still face notable limitations. For example, sequential VLM-LLM pipelines process each modality independently, which often results in information loss and limits cross-modal interaction. In contrast, methods based on multimodal large language models (MLLMs) often focus exclusively on applying changes indicated by the text, without fully utilizing the contextual visual information from the reference image. To address these issues, we propose multi-faceted Chain-of-Thought with re-ranking (MCoT-RE), a training-free zero-shot CIR framework. MCoT-RE utilizes multi-faceted Chain-of-Thought to guide the MLLM to balance explicit modifications and contextual visual cues, generating two distinct captions: one focused on modification and the other integrating comprehensive visual-textual context. The first caption is used to filter candidate images. Subsequently, we combine these two captions and the reference image to perform multi-grained re-ranking. This two-stage approach facilitates precise retrieval by aligning with the textual modification instructions while preserving the visual context of the reference image. Through extensive experiments, MCoT-RE achieves state-of-the-art results among training-free methods, yielding improvements of up to 6.24% in Recall@10 on FashionIQ and 8.58% in Recall@1 on CIRR.

Keywords: Image Processing and Pattern Recognition, Deep Learning
Abstract: Composed image retrieval (CIR) is a vision-language task that retrieves a target image using a reference image and modification text, enabling intuitive specification of desired changes. While effectively fusing visual and textual modalities is crucial, existing methods typically adopt either early or late fusion. Early fusion tends to excessively focus on explicitly mentioned textual details and neglect visual context, whereas late fusion struggles to capture fine-grained semantic alignments between image regions and textual tokens. To address these issues, we propose FAR-Net, a multi-stage fusion framework designed with enhanced semantic alignment and adaptive reconciliation, integrating two complementary modules. The enhanced semantic alignment module (ESAM) employs late fusion with cross-attention to capture fine-grained semantic relationships, while the adaptive reconciliation module (ARM) applies early fusion with uncertainty embeddings to enhance robustness and adaptability. Experiments on CIRR and FashionIQ show consistent performance gains, improving Recall@1 by up to 2.4% and Recall@50 by 1.04% over existing state-of-the-art methods, empirically demonstrating that FAR-Net provides a robust and scalable solution to CIR tasks.

Keywords: Image Processing and Pattern Recognition, Machine Vision, Neural Networks and their Applications
Abstract: Considering the influence of reflected and transmitted light on physical imaging in real-world scenarios, we propose a method that leverages reflection and refraction coefficients. By calculating these coefficients between the captured image and the transmission image, our approach effectively guides the separation of reflection layers within the captured image. The proposed three-branch framework integrates the Feature Enhancement Module (FEM), which is specifically designed to recover additional details and produce high-quality transmission images. These enhancements significantly improve the performance of the reflection removal process. Comprehensive experiments conducted on various datasets and in comparison with state-of-the-art reflection removal methods demonstrate the effectiveness of our approach. The results show that our method excels in eliminating reflection artifacts and correcting intensity distortions, yielding superior image quality. Especially, the PSNR and SSIM scores of our model outperform existing methods. Furthermore, the simplicity of the input and architecture underscores the practicality of the proposed method. The source code of the proposed method is available at https://reurl.cc/LnAmqx.

Keywords: Image Processing and Pattern Recognition, Deep Learning, Machine Learning
Abstract: Breast cancer (BC) remains one of the leading causes of cancer-related mortality among women, despite recent advances in Computer-Aided Diagnosis (CAD) systems. Accurate and efficient interpretation of multi-view mammograms is essential for early detection, driving a surge of interest in Artificial Intelligence (AI)-powered CAD models. While state-of-the-art multi-view mammogram classification models are largely based on Transformer architectures, their computational complexity scales quadratically with the number of image patches, highlighting the need for more efficient alternatives. To address this challenge, we propose Mammo-Mamba, a novel framework that integrates Selective State-Space Models (SSMs), transformer-based attention, and expert-driven feature refinement into a unified architecture. Mammo-Mamba extends the MambaVision backbone by introducing the Sequential Mixture of Experts (SeqMoE) mechanism through its customized SecMamba block. The SecMamba is a modified MambaVision block that enhances representation learning in high-resolution mammographic images by enabling content-adaptive feature refinement. These blocks are integrated into the deeper stages of MambaVision, allowing the model to progressively adjust feature emphasis through dynamic expert gating, effectively mitigating the limitations of traditional Transformer models. Evaluated on the CBIS-DDSM benchmark dataset, Mammo-Mamba achieves superior classification performance across all key metrics while maintaining computational efficiency.

Keywords: Image Processing and Pattern Recognition, Deep Learning, Machine Vision
Abstract: We propose DA-MSIRNet, a lightweight yet innovative architecture for high-quality image inpainting that significantly enhances the standard U-Net through four key innovations: (1) Context Anchor Attention (CAA) for efficient global context modeling via adaptive region selection, (2) Sparse Self-Attention (SpA), inspired by Spa-former, for dynamic and precise local detail refinement by focusing on salient relationships, (3) Multi-Scale Inverted Residual (MSIR) modules for enhanced multi-scale feature fusion through optimized skip connections, and (4) Structural Similarity (SSIM) Loss for improved perceptual quality and fidelity. DA-MSIRNet effectively addresses critical limitations of existing methods, including GAN instability, U-Net's restricted receptive field, and Transformer computational complexity. Comprehensive evaluations on Places2 and CelebA-HQ datasets demonstrate that DA-MSIRNet achieves state-of-the-art performance in both quantitative metrics (PSNR/SSIM/FID) and visual quality, while maintaining superior computational efficiency. The code for our DA-MSIRNet is publicly available on GitHub: https://github.com/nutcliu2507/DA-MSIRNet.

Keywords: Virtual and Augmented Reality Systems, User Interface Design, Human-Machine Interface
Abstract: We have created a virtual reality (VR) system designed to enhance the understanding of dementia symptoms among third parties, including young people who may be unfamiliar with dementia, as well as caregivers and family members of patients. The VR system can effectively simulate the disorientation and visual field constriction associated with Alzheimer's disease, enabling users to navigate a realistic cityscape. By altering the position of the numerical landmarks on a round trip, the VR system increased the completion time for maze tasks, thereby simulating the experience of disorientation and the sensation of being lost. Furthermore, the VR system replicated the symptoms of visual field constriction and suppressed parasympathetic nervous system activity, along with subjective symptoms such as decreased sleepiness and increased discomfort. Such VR experiences will be instrumental in improving the understanding of patients with dementia and could be useful in patient care.

Keywords: Assistive Technology, Human Enhancements, Virtual/Augmented/Mixed Reality
Abstract: A key problem in the adoption of exoskeletons in industry is that workers are incorrectly fitting the device, leading to discomfort and suboptimal functioning of the exoskeleton. Although biomechanical modeling and design optimization strategies have tried to resolve this issue, we propose a user-centric, real-time fitting aid to guide and control the correct fitting process, which has not been achieved in practice. Inspired by previous work that uses exergame-like feedback to instruct a user, we compared augmented reality (AR)-based visual cues to guide the accurate fitting of a passive upper limb exoskeleton. We selected visual cues through a systematic literature review and evaluated their efficacy and usability for different aspects of exoskeleton fitting in a study with sixteen participants. The study outcome suggests a statistically significant preference for a semi-transparent overlay instead of a more abstract arrow-based method. Moreover, the results indicate high usability and satisfaction with our approach, improved user acceptance, and potentially enhanced fitting accuracy. These findings advance understanding of the viability of exergame-like real-time guidance as a means to increase exoskeleton acceptance and adoption in industrial settings.

Keywords: Virtual/Augmented/Mixed Reality, Brain-Computer Interfaces, Augmented Cognition
Abstract: Attention is a critical cognitive process required to perform everyday tasks, yet it has become increasingly difficult to maintain due to constant distractions. While several systems aim to support attentional control, many cannot adapt to the user’s cognitive state in real time. This study proposes an adaptive virtual reality platform based on the Stroop task that measures cognitive engagement (CE) using realtime electroencephalography (EEG) data. A subject study was implemented, comparing a control group (non-adaptive version) with an experimental group (adaptive version). Results showed that the adaptive environment was significantly more effective at sustaining attention than the static condition. Additionally, cognitive engagement increased as task difficulty advanced, but stabilized after reaching higher difficulty levels, suggesting the presence of a cognitive ceiling or overload threshold. These results suggest that an adaptive VR Stroop task can enhance attentional regulation, provided that the difficulty remains within the cognitive capacity of the user. This system has promising implications for real-time, personalized cognitive training.

Keywords: Virtual/Augmented/Mixed Reality, Human-Computer Interaction, Interactive and Digital Media
Abstract: Tangible User Interfaces (TUI) for human--computer interaction (HCI) provide the user with physical representations of digital information with the aim to overcome the limitations of screen-based interfaces. Although many compelling demonstrations of TUIs exist in the literature, there is a lack of research on TUIs intended for daily two-handed tasks and processes, such as cooking. In response to this gap, we propose SPICE (Smart Projection Interface for Cooking Enhancement). SPICE investigates TUIs in a kitchen setting, aiming to transform the recipe following experience from simply text-based to tangibly interactive. SPICE uses a tracking system, an agent-based simulation software, and vision large language models to create and interpret a kitchen environment where recipe information is projected directly onto the cooking surface. We conducted comparative usability and a validation studies of SPICE, with 30 participants. The results show that participants using SPICE completed the recipe with far less stops and in a substantially shorter time. Despite this, participants self-reported negligible change in feelings of difficulty, which is a direction for future research. Overall, the SPICE project demonstrates the potential of using TUIs to improve everyday activities, paving the way for future research in HCI and new computing interfaces.

Keywords: Virtual/Augmented/Mixed Reality, User Interface Design, Human-Computer Interaction
Abstract: Overlaying real-world objects with virtual information is a fundamental aspect of augmented reality (AR). Traditional techniques typically rely on marker-based or feature-based computer vision methods, each with inherent limitations regarding robustness and environmental adaptability. In this paper, we introduce the Three-Touch (TT) technique, a novel approach that enables users to manually align virtual content with physical objects by selecting corresponding feature pairs. We conduct a comparative study between the TT technique and a conventional grab gesture (G) method, evaluating usability, user experience, task load, positioning accuracy, and task completion time. Our findings indicate that while the TT technique reduces physical and mental effort, it may lead to greater positional offsets and reduced task performance compared to the G technique.

Keywords: Affective Computing, Virtual and Augmented Reality Systems, Human-Computer Interaction
Abstract: Simulator sickness remains a major challenge in immersive simulation systems, particularly in high-fidelity driving environments. While previous research has utilized machine learning with multimodal physiological data, it often depends on restricted feature sets and overlooks comprehensive eye movement and head motion data. In this study, we propose a deep learning framework using a hybrid 1CNN-BiLSTMAttention model for real-time quantification of simulator sickness severity. Data were collected using Deakin University’s Genesis Simulator—an immersive 360° environment with a six degrees-of-freedom motion platform. Eye-tracking and head movement features were extracted, and the Fast Motion Sickness (FMS) scale was used for severity labeling. The proposed model achieved 87.6% accuracy and an F1-score of 0.91 for high-severity detection. A 5-fold cross-validation demonstrated a significant benefit of attention over baseline models (p = 0.0019). This study offers a scalable solution for adaptive simulation and intelligent vehicle systems through integrated eye and head movement data.

Keywords: Human-centered Learning, Virtual and Augmented Reality Systems, Virtual/Augmented/Mixed Reality
Abstract: This study investigates how virtual professor proximity influences student comprehension and attention in an immersive learning environment. 27 participants experienced three lectures in virtual reality (VR) under varying spatial conditions: close distance (personal space intrusion), optimal distance (user defined preferred proximity), and far distance (outside social interaction proximity). Attention was tracked via eye movement, and comprehension was evaluated through multiple-choice tests. Results show that close distance reduced comprehension despite increased gaze toward the professor, likely due to discomfort and attentional overload. Far distance increased visual distraction, while the optimal distance supported comfort, engagement, and performance. Our findings offer actionable design guidelines for avatar placements in virtual learning systems to optimize user experience and cognitive effectiveness.

Keywords: Human Factors, Design Methods, Human-Computer Interaction
Abstract: Coding the gestures people use when interacting with tangible devices can aid interface design by making interfaces more intuitive and consistent. Past work devotes very little space to developing coding schemes for user-defined gesture sets. A generalizable coding scheme for gestures made by the human hand is developed for tangible devices using inspiration from predetermined time systems (PTS) from human factors engineering. The coding scheme is generic, adaptable to any hand motion proposal and has features such as duration invariance. Case study examples demonstrate its ability to provide consistent labels for hand interactions with tangible devices. Coding scheme labels are also balanced, having neither too many nor too few distinct gestures. This paper can help teach and create coding schemes for gesture elicitation studies performed by human-computer interaction researchers. The larger societal impact of this work is to help advance scientific methods for beneficial interface design within HCI/HRI by combination with human factors theory.

Keywords: Human Factors, Human-Computer Interaction, Human-centered Learning
Abstract: This study aims to conduct an experimental evaluation of personalized interventions to promote physical activity. Promoting health behavior such as physical activity to improve well-being is an important social issue. Maintaining health behavior based solely on an individual's will without any support is difficult due to lifestyle and environmental/social factors. Therefore, it is important to increase the intrinsic motivation for health behavior through external interventions. In recent years, services that promote physical activity by intervening with users through mobile health applications have become increasingly popular. In such services, users are generally encouraged to increase their health behavior through positive interventions such as supportive messages. However, such one-size-fits-all interventions have limited impact because of a mismatch with individual needs, resulting in inadequate or sometimes counterproductive effects. Therefore, interventions tailored to individual needs (i.e., personalized interventions) are needed. We used PLS-SEM to model the relationship between individually different personalities (i.e., personality traits) and the motivational ability of intervention for physical activity (i.e., persuasiveness), with the goal of enabling personalized interventions. We conducted an experiment to examine the effect of physical activity promotion through the personalized intervention based on the PLS-SEM model. As a result of the experiment, the personalized intervention group showed a significantly higher implementation rate and continuation rate of physical activity compared to the non-personalized intervention group and the control group. These results suggest that it is possible to promote physical activity more effectively through an intervention tailored to personality traits. This may be a solution to the problem of insufficient intervention effect with a one-size-fits-all approach.

Keywords: Human Factors, Systems Safety and Security
Abstract: Children's widespread use of digital platforms has intensified concerns about the adequacy of privacy protections. Current legislation places the responsibility for managing children's privacy on parents and guardians, assuming they possess the necessary knowledge to make informed decisions. In light of this, this work assesses parental maturity in managing children's privacy on social networks. First, we identify the main privacy attributes relevant to children’s online data protection by analyzing existing laws and regulations, including the GDPR, COPPA, and LGPD. This phase establishes a regulatory baseline for evaluating parental responsibilities and expectations. In the second phase, we surveyed 77 parents and guardians to assess their level of maturity in managing privacy-related measures and to evaluate how effectively they can protect their children's data in digital environments. Our results reveal a significant discrepancy between perceived and actual knowledge, suggesting that many parents may not be adequately prepared to fulfill the role expected by current regulations. These findings support the need for clearer policies and a shared responsibility model between platforms and guardians to ensure child privacy.

Keywords: Artificial Social Intelligence, AI and Applications, Application of Artificial Intelligence
Abstract: Network Operation and Maintenance (NOM) is an important part of the IT and telecommunications infrastructure. To improve the efficiency of NOM personnel, predict base station alarms, and classify base stations more precisely, we build a new labeled time series dataset, called BSAlarm, which collects 1000 different time series from processed real-world network alarm log data. Furthermore, we develop a NOM pipeline that employs B-spline interpolation alongside the moving average method to create and utilize this dataset for both long- and short-time series forecasting (LTSF), as well as base station classification tasks. Unlike prior NOM methods, our pipeline is centered on the generation of time series data, thereby enhancing alarm prediction performance and the visualization of changes under base station alarm conditions. Extensive experiments show that the proposed BSAlarm dataset and pipeline achieve remarkable results on time-series-data-based NOM.

Keywords: Artificial Social Intelligence, Application of Artificial Intelligence
Abstract: Ironic language, prevalent on social media, Existing sarcasm detection methods face three main limitations: (1) the implicit and context-dependent nature of irony, requiring advanced reasoning; (2) substantial computational overhead in locally fine-tuning large language models (LLMs); and (3) insufficient detection performance due to a narrow analysis of sarcastic expressions. To address these issues, we propose the textbf{Multi-Agent Sarcasm Detection framework with Chain-of-Thought (MASD)}. MASD introduces a Chain-of-Thought prompting paradigm that combines syntactic cues with deep semantic reasoning. A KNN-based sampling technique identifies semantically similar examples to produce a robust LLM emulation template. Leveraging LLM inferential capabilities, MASD enables effective sarcasm detection with minimal reliance on extensive pre-trained datasets. Moreover, its multi-agent architecture allows for the collaborative extraction of contextual, emantic and rhetorical insights. By incorporating a self-reflection mechanism, LLMs continuously refine their reasoning logic. Experiments on benchmark datasets across real-world scenarios demonstrate Macro-F1 score improvements of 5.04%, 1.57%, and 8.82%, underscoring the framework's effectiveness and scalability, with substantial implications for advancing sarcasm detection methodologies.

Keywords: Artificial Social Intelligence, Application of Artificial Intelligence, Cyborgs,
Abstract: This paper proposes a comprehensive framework that integrates the YOLOv5 object detection model with the Proximal Policy Optimization (PPO) algorithm enhanced by a multi-head attention mechanism, aiming to enable a humanoid robot to autonomously locate itself in front of a door and complete the task of opening the door. Firstly, the real-time and precise positioning and navigation of the door are achieved through the object detection model. Secondly, the improved PPO algorithm dynamically extracts key features and establishes deep associations through the multi-head attention mechanism to optimize the arm control strategy. Experiments are carried out on the Webots simulation platform. The results show that this method not only improves the accuracy and efficiency of the robot in reaching the front of the door but also significantly enhances the success rate of the door-opening action and the adaptability to environmental changes. It verifies the effectiveness of the integration of the attention mechanism and deep reinforcement learning technology in improving the autonomous operation skills of humanoid robots, providing new ideas and technical support for the future development of humanoid robots.

Keywords: Artificial Social Intelligence, Deep Learning, Cyborgs,
Abstract: Cherry tomato harvesting is generally performed manually. Robotic harvesting is gaining increasing interest from both academia and industry. This paper proposes a cherry tomato cluster detection algorithm based on YOLOv8, named YOLOv8-CTCD. First, the YOLOv8 input channels are adjusted to enable 4-channel RGB-D images as input. Subsequently, a CARAFE-M module is designed to replace the upsampling method in YOLOv8n. It maintains a lightweight architecture while achieving a larger receptive field, allowing effective aggregation of contextual information. In addition, it assigns greater weight to more important features. Moreover, a C2f-MLCA module is introduced into YOLOv8, which integrates information from feature maps at different levels and enhances the network's capability of feature extraction. It also integrates the SPPELAN module to strengthen its feature fusion capability. YOLOv8-CTCD has been evaluated using a private cherry tomato dataset obtained from a greenhouse farm. The experimental results show that it achieves an mAP@50 of 93.8% and an mAP@50:90 of 68%, which represents improvements of 2.1% and 3% over YOLOv8n, respectively.

Keywords: Artificial Social Intelligence, Deep Learning, Neural Networks and their Applications
Abstract: Remote sensing scene classification in mining areas is crucial for evaluating deposit occurrence states and supporting environmental monitoring and sustainable development. However, existing methods are limited by homogeneous and heterogeneous spectral spatial and topographic features of mining areas, large intra-class variations, and small target sizes. To overcome these limitations, this study integrates RGB and SAR data to construct a multi-modal dataset and proposes an RGB-SAR mining scene classification model with dual feature enhancement and adaptive cross-modal attention interaction. The model includes: (1) Dual feature enhancement module that suppresses irrelevant features and enhances discriminative multi-scale representations of mining targets; (2) BifocalNet based feature extraction module using a CNN-Transformer hybrid architecture to capture local textures and model global context; (3) Attention based adaptive cross-modal interaction module that achieves deep spectral geometric feature complementarity through the fusion of RGB and SAR modalities. Experiments show the model achieves an OA of 84.58%, outperforming other models and ranking first or second in most evaluation metrics. The proposed dataset and model thus advance mining scene classification.

Keywords: Artificial Social Intelligence, Machine Vision, Neural Networks and their Applications
Abstract: Spacecraft pose estimation from monocular images presents significant challenges due to complex environmental conditions such as occlusions, illumination variations, and background interference, as well as estimation inaccuracies caused by multi-scale variations in object distance and viewpoint. To address these issues, this paper proposes a novel monocular pose estimation algorithm. In concrete terms, a high-resolution feature extraction framework is constructed using Higher-HRNet, a variant of the High-Resolution Network (HRNet), to generate multi-scale feature maps and enhance spatial feature representation. To balance estimation accuracy with real-time performance under limited computational resources, lightweight Ghost-BasicBlock and Ghost-Bottleneck modules are designed to reduce model complexity. Moreover, to mitigate the loss of feature representation capacity induced by model compression, a Biformer-Receptive Attention (BRA) mechanism is incorporated in the decoding stage to strengthen spatial-context modeling and improve keypoint localization accuracy. As the culminating process, the six-degree-of-freedom pose of the spacecraft is estimated by integrating the Efficient Perspective-n-Point (EPnP) algorithm with Random Sample Consensus (RANSAC). Experimental results on the SPEED dataset demonstrate that the proposed method achieves a mean translation error (meanET) of 0.0077m and a mean rotation error (meanER) of 0.0232°, with respective medians (medianET and medianER) corresponding to 0.0038m and 0.0112°, respectively—outperforming current state-of-the-art methods.

Keywords: Computational Intelligence, AI and Applications, Artificial Social Intelligence
Abstract: AlphaZero’s success in games like chess, Go, and shogi has raised questions about its learning behaviour and limitations. Its play often mirrors human reasoning, applying familiar and novel concepts alike. In this work, we apply AlphaZero to Tic-Tac-Toe to examine its ability to learn a simple rules-based hierarchy that models human understanding. Despite the game’s simplicity, AlphaZero exhibits counter-intuitive, sub-optimal behaviour not typical of human players. However, we show that modifying the reward structure can enforce rule hierarchies in pure MCTS and supervised settings, and improves AlphaZero’s rule learning while reducing sub-optimal play.

Keywords: Machine Learning, Computational Intelligence, Computational Intelligence in Information
Abstract: In software defect prediction, achieving a balance between computational efficiency and accuracy is critical yet challenging. Traditional approaches often depend on high-dimensional feature sets, introducing redundant parameters that escalate complexity without enhancing predictive capability. This limitation becomes particularly acute in large-scale industrial software systems, where both feature selection and interaction analysis are vital for reliable predictions due to the complexity and scale of real-time operations. In such environments, defects can disrupt critical processes, making efficient and accurate prediction essential for maintaining system reliability and safety. To overcome these issues, we propose a two-phase method combining feature reduction and interaction optimization tailored for industrial software applications.The first phase employs mutual information and Lasso regularization to eliminate 43% of redundant features, significantly lowering model complexity. The second phase utilizes polynomial feature expansion to construct interaction terms, with Partial Dependence Plots (PDPs) decoding their nonlinear impacts. Experimental results demonstrate a 5%-25% acceleration in training speed across diverse models while preserving accuracy, proving the method’s efficacy for scalable defect prediction tasks in industrial software environments, where rapid deployment and robustness are paramount.

Keywords: AIoT, AI and Applications
Abstract: The Mobile Edge Computing (MEC) paradigm is widely believed to be highly potent in boosting computation-intensive AI applications, where massive amounts of data are processed at the internet-of-things end. Nevertheless, in reality, it remains a great challenge in filling the gap between resource capacity and user need especially when MEC users are socially connected and content requests can be highly dynamic in terms of their spatial-temporal patterns. Traditional approaches in this direction tend to leverage mobile trace prediction algorithms for yielding mobility-aware content acquisition schedules. In this research, however, we consider that the interest in scocially-connected MEC clusters of users can be highly useful in aiding the design of effective MEC service allocation mechanisms. For this purpose, we develop a novel interest-aware method for content placement. It consists of a clustering model for dividing servers based on their physical connection, a clustering model for grouping users based on their interest similarity, and a predictive content placement model by using a transformer-based method.Simulative results upon real-world datasets show our method outperforms traditional ones across multiple performance metrics.

Keywords: Intelligent Transportation Systems, Decision Support Systems
Abstract: Existing algorithms for the flight string scheduling problem often fall short due to their NP-hard nature and difficulties with local optima. To enhance efficiency, we propose a two-stage deep reinforcement learning column generation optimization algorithm (2-Stage DRLCG). This method leverages "offline training and online decision-making" to improve scheduling. In the first stage, we create a flight connection diagram based on airport connections and minimum transfer times. The second stage utilizes a deep Q network (DQN) to optimize column generation through a reduced cost (RC) approach. Experiments using data from a major Chinese airline demonstrate significant improvements in cost reduction and computational efficiency, highlighting the potential of deep reinforcement learning in flight scheduling.

Keywords: Intelligent Transportation Systems, Decision Support Systems
Abstract: In order to accurately predict the flight transit time at airports of different sizes, this paper proposes a two-stage flight transit time dynamic prediction model based on ensemble learning and Transformer to jointly and dynamically predict the flight transit time of multiple airports. First, considering the continuity of flight operation, the flight information is associated, and the key features are extracted from the historical information of the flight to construct a multi-airport flight transit data set; secondly, the parameter transfer and connection of the flight transit time are carried out; thirdly, the Bayesian optimization algorithm is introduced to optimize the hyperparameters of the LightGBM and XGBoost models, and the optimal hyperparameters are used to make a preliminary prediction of the flight transit time in the first stage; finally, combined with the real-time delay data and the actual operation of the previous flight, the Transformer model is used to dynamically adjust the initial flight prediction time in the second stage to obtain the final prediction result. The experimental results show that the two-stage prediction results reduce the MAE by 2.42 and 2.49 minutes compared with the one-stage MAE, which proves the effectiveness of this method, and it is better than the random forest, support vector machine and neural network models in terms of prediction error.

Keywords: Intelligent Transportation Systems, Decision Support Systems
Abstract: 频繁的航班延误已成为 全球空中交通，严重影响出行 乘客体验和运营效率 航空系统。为了有效地捕获 航班延误的时空动态特征， 本文提出了一种新的时空预测 航班延误模型，称为逐层 时空图卷积网络（LSTGCN），其中 增量捕获时空依赖关系 通过分层图卷积。该型号 集成时间建模和逐层图 卷积技术，使用时间卷积 网络 （TCN） 与 Transformer 机制相结合，以 提取航班延误的时间特征，介绍 图卷积网络 （GCN） 和图注意力 深入探索拓扑的机制（GAT） 机场之间的关系和动态影响 逐层交替的方式，构建多个 图邻接结构。与特定节点结合 聚合方法，这种方法准确地捕获了 拓扑结构特征及动态相互作用 机场网络内的关系。最后， 提取的时空特征融合成一个统一

Keywords: Intelligent Transportation Systems, Robotic Systems, Smart Buildings, Smart Cities and Infrastructures
Abstract: With the rapid development of UAVs, object detection in aerial imagery has become an important techniques. However, deploying real-time detection models on UAV platforms remains highly challenging due to limited computational costs, as well as the multi-scale objects and dense distribution caused by high-altitude imaging. Numerous studies have contributed valuable insights to address these challenges, yet opportunities remain for improving the balance between model efficiency and detection accuracy. Moreover, current research mainly focuses on small object detection, without fully considering the detection requirements for multi-scale and rotated objects in high-altitude imagery. To address these issues, this paper proposes a lightweight object detection model specifically designed for multi-scale and rotated object detection in high-altitude imagery. Furthermore, to better evaluate the model's performance in UAV-based object detection in high-altitude imagery, this work uses VisDrone 2019 dataset to assess the model’s real-world performance. As a results, compared to existing object detection approaches, the proposed model achieves a strong balance between detection accuracy, model efficiency, and inference speed, reducing parameters by approximately 31% and improving inference speed by 25%, while maintaining the same detection accuracy as the best baseline methods.

Keywords: Communications, Cooperative Systems and Control
Abstract: Wireless rechargeable sensor networks (WRSNs) play a pivotal role in ground-based monitoring within sixth-generation (6G) networks. However, these networks frequently face challenges in ground-to-air communication, stemming from urban obstructions, suboptimal antenna configurations, and other factors. To address these issues, this paper proposes a novel framework that integrates WRSNs with unmanned aerial vehicle (UAV) control. Specifically, we consider a system utilizing a mobile charging vehicle (MCV) equipped with an intelligent reflecting surface (IRS) to concurrently meet sensor charging needs and improve communication between the base station (BS) and the UAV. We formulate a dynamic multi-objective optimization problem to maximize sensor charging utility and BS-to-UAV communication rate while minimizing the MCV’s energy consumption. This is accomplished through the coordinated optimization of the MCV’s trajectory and the IRS’s phase configurations. To tackle the computational complexity of continuous two-dimensional trajectory optimization, we present a Pareto-optimal waypoint generation strategy. Furthermore, we propose an enhanced proximal policy optimization (IPPO) algorithm, incorporating shared feature extraction and LSTM-enhanced policy networks, to effectively manage high-dimensional state processing and temporal dependencies. Simulation results demonstrate that our approach outperforms benchmark algorithms in terms of charging utility, communication rate, and energy consumption optimization.

Keywords: Communications, Infrastructure Systems and Services, Smart Sensor Networks
Abstract: The widespread deployment of Internet of Things (IoT) devices, coupled with their often limited security capabilities, has significantly increased the network attack surface. Consequently, network asset managers need to continuously monitor and assess vulnerable IoT devices to mitigate potential threats. However, existing passive IoT device identification approaches, which rely on network traffic analysis, are hindered by substantial labeling requirements and computational overhead, severely limiting their practicality in real-world environments. To address these challenges, we propose FS-IoT, a fast few-shot IoT device identification framework. FS-IoT introduces the concept of packet bursts as the fundamental unit of recognition and systematically explores their extraction, representation, classification, and aggregation for IoT device identification. Experimental evaluations on two public datasets demonstrate that FS-IoT achieves superior accuracy (99.94% and 98.15%) while requiring only 2% of the labeled training data needed by state-of-the-art methods. Furthermore, FS-IoT improves recognition speed by an order of magnitude, making it highly suitable for practical, large-scale deployments.

Keywords: Homeland Security, Decision Support Systems, Communications
Abstract: Social media’s rise has enabled rapid content dis- semination and fostered ideological echo chambers. These platforms, by design, often promote insular communities and targeted content exposure, making them effective tools for extremist recruitment and propaganda. Existing research has primarily focused on detecting extremist-related content particularly on Twitter (now X), which ignores the increasing adoption of other platforms such as Bluesky and Mastodon. To capture this evolving landscape, we propose SOCINT (Social Observations with Chronic Interactions and Network Tracking), a novel framework for user-level sentiment classification at specific timestamps. Unlike existing approaches that analyze individual posts in isolation, SOCINT leverages both temporal activity patterns and social network interactions to assess user sentiment and track its evolution over time. We also introduce a new multi-platform dataset enriched with social and temporal metadata. Experimental results demonstrate that SOCINT outperforms existing baselines, offering a dynamic and context-aware solution for detecting and monitoring extremist behavior across social media ecosystems.

Keywords: Decision Support Systems, Manufacturing Automation and Systems, System Modeling and Control
Abstract: The paper proposes a novel decomposition method aimed at supporting decision-making in manufacturing systems. The main goal of the presented technique is an adequate proposal for dividing the production process flow in the manufacturing system by delegating the selected activities to the outsourcing mode. In order to obtain such separated activities within a production process, the modelled system is decomposed into the set of components (sets of places) that are executed externally. The presented idea involves Petri net and hypergraph theories. In particular, the system is modelled by a Petri net and further decomposed with the use of hypergraph’s transversals. The key advantage of the proposed algorithm is many decomposition possibilities, since the activities performed as part of the production process in an outsourcing mode may be specified precisely, exactly conforming to the production strategy. It can be treated as a tool supporting decision-making by managers in the context of delegating tasks from the production process in an outsourcing mode. Such study complements the results of the analysis of cost and efficiency of the production process realized in a manufacturing enterprise.

Keywords: System Modeling and Control, Discrete Event Systems
Abstract: This paper aims to analyse the efficiency and effectiveness of two invariant coverage verification methods for Petri net-based concurrent systems. The first algorithm is based on the classical approach, where the complete analysis of the Petri net’s incidence matrix is performed. This technique assures the correctness of results, but it may be inefficient due to the exponential computational complexity. The second algorithm applies an innovative idea of Petri net’s incidence matrix transformation and examination. The main advantage of such an approach is its polynomial computational complexity. On the other hand, the method may not always be accurate and may not detect all errors in the examined system. This paper exhaustively verifies the efficiency (run-time) and effectiveness (correctness of the achieved results) of both methods. The theoretical analyses are supported by the results of experiments performed on 386 benchmarks (Petri net models). In addition, besides the examination of the run-time of methods, the number of iterations performed by algorithms is computed and analysed.

Keywords: Adaptive Systems, Smart Metering, Intelligent Transportation Systems
Abstract: Although current ReID (Re-Identification) methods have become relatively mature, they still require manual extraction of pedestrian images and annotation of features. They lack semantic understanding capabilities in open scenes. While some ReID models integrated with LLMs (Large Language Models) offer more comprehensive functions and better performance, they still fall short in terms of semantic understanding and cross-modal retrieval. To solve these problems, we introduce SUO-ReID, a semantic understanding-based approach for ReID in open scenes. SUO-ReID combines LVLM (Large Vision-Language Model) with ResNet (Residual Network) to extract high-level semantic features of targets in open scenes. It can also perform more flexible and complex functions, such as searching for or comparing targets with specified features, through instruction inputs. Experimental results show that SUO-ReID achieves an accuracy rate of 95.73% on datasets such as Market-1501 and DukeMTMC and exhibits excellent semantic understanding capabilities in open scenes, supporting cross-modal retrieval. It can also provide a detailed description of the features of the identified object and its surrounding scene. This study provides new insights into the application of large vision-language models in the field of ReID.

Keywords: Cybernetics for Informatics, Hybrid Models of Computational Intelligence, Hybrid Models of Neural Networks, Fuzzy Systems, and Evolutionary Computing
Abstract: This paper addresses co-design control strategies for continuous-time Markov jump systems where subsets of transition rate matrices are fixed a priori, challenging conventional co-design methodologies. A synchronously mode-dependent parametric framework is developed to address partial transition rate optimization alongside output feedback controller synthesis. Novel criterion is derived to guarantee mean-square stability by reconstructing adjustable switching parameters while preserving fixed system transitions. Crucially, the formulation decouples decision variables associated with predefined and designable transition rates, streamlining computational complexity and eliminating the need for high-dimensional parameter optimization. Stability analysis and controller design are unified through hybrid control principles. A numerical case studies validate the proposed approach, demonstrating enhanced feasibility compared to existing methods.

Keywords: Cyborgs,, Swarm Intelligence
Abstract: This paper presents a universal and adaptive control framework for reconfigurable vehicles subject to composite motion disturbances, incorporating a collaborative fault-tolerant mechanism. A model-based cascaded control architecture is developed based on the vehicle’s kinematic and dynamic models. To ensure safety, an improved adaptive geofencing strategy is proposed which integrates capability constraints with barrier functions in the kinematic loop. For dynamic feedback, an adaptive gain-filtered extended state observer enables accurate disturbance estimation from noisy outputs and improves robustness via feedforward compensation. Meanwhile, a collaborative fault-tolerant mechanism further allocates control inputs to mitigate actuator faults. Finally, experimental results validate the proposed method’s effectiveness under complex interference scenarios.

Keywords: Machine Learning
Abstract: How to effectively optimize the operation for a complex supply chain environment has been high on the agenda. Although the existing deep reinforcement learning methods have achieved success in certain applications, they still face limitations in complex supply chain environments, including difficulties in data sharing and a lack of digital collaboration, especially in multi-agent systems. In order to meet this challenge, we propose a novel multi-agent reinforcement learning algorithm based on dynamic optimistic weights (DO-QMIX), aiming at solving the shortcomings of the traditional weighted QMIX algorithm (WQMIX) in the simplicity of the weighting function. WQMIX employs two weighting schemes to handle multi-agent issues. However, its fixed weighting function restricts algorithm performance and hinders adaptability to dynamic, complex supply chain challenges. Therefore, we propose a dynamic weighting mechanism, which can adjust the weighting function in real time based on the changes in the environment, thus improving the overall efficiency. We construct a complex multistage supply chain environment in the real-world supply chain scenario and conduct many experiments using both real-world and simulated datasets. The experimental results demonstrate that DO-QMIX is significantly superior to the traditional multi-agent reinforcement learning algorithm in complex supply chain scenarios, especially in dealing with dynamic changes and complex decisions.

Keywords: Swarm Intelligence, Cybernetics for Informatics, Agent-Based Modeling
Abstract: In this paper, a finite-time robust safe cooperative control strategy is proposed for heterogeneous multi-agent systems (HMAS) in cluttered obstacle environments under input saturation, external disturbances, and Denial-of-Service (DoS) attacks. An adaptive event-triggered observer is designed at the cyber layer to achieve distributed resilient tracking under DoS-induced communication networks. At the physical layer, a distributed control scheme based on finite-time robust control barrier function (FT-RCBF) is first developed to ensure fast obstacle avoidance for HMAS. The proposed method is applied to a cooperative forest traversal and monitoring task for ground-air autonomous systems, and its effectiveness and robustness are verified through simulations.

Keywords: Optimization and Self-Organization Approaches, Computational Intelligence, Application of Artificial Intelligence
Abstract: This paper presents a prescribed performance finite-time observer-based super-twisting controller (PPFTOST) for cooperative aerial suspended transport systems. The proposed controller integrates a prescribed performance framework, an appointed-time disturbance observer (ATDO) for disturbance estimation, and a fast terminal sliding mode super-twisting controller (FTSMSTC) for rapid convergence with reduced chattering. Through this combination, the closed-loop system achieves finite-time convergence of both the disturbance estimation error and the sliding variables, while ensuring that the tracking errors remain within prescribed performance bounds. In simulations against an ATDO-only baseline, the PPFTOST reduced payload-tracking root mean square error by 8.2%, 6.7%, and 9.3% on the x, y, and z axes, respectively, with all actuator commands remaining within limits. These results demonstrate that the proposed method enables accurate, robust, and smooth multi-UAV load transport under realistic disturbance conditions.

Keywords: Cloud, IoT, and Robotics Integration, Swarm Intelligence, Agent-Based Modeling
Abstract: This paper investigates a discrete higher-order sliding mode control (DHOSMC) framework integrated with cloud-based predictive control (CBPC) to address time-varying formation control in discrete multi-agent systems (DMAS). The approach ensures robustness against model uncertainties and external disturbances while compensating for communication delays. The DHOSMC guarantees finite convergence of tracking errors, and the CBPC leverages cloud resources for multi-step prediction and coordination among agents. Both delay-free and delayed scenarios are considered. Lyapunov-based analysis is used to prove system stability. Simulation results demonstrate the effectiveness of the proposed method in achieving accurate and robust time-varying formation of DMAS.