Keywords: Image Processing and Pattern Recognition, Neural Networks and their Applications, Deep Learning
Abstract: With the rapid advancement of computer vision and graphics, high-quality single-view 3D reconstruction has become increasingly significant in applications such as autonomous driving, robotics, and virtual reality. Traditional methods often struggle with the inherent ill-posed nature of single-view reconstruction, leading to limited accuracy and reduced detail retention. To address these challenges, we introduce an innovative framework that integrates an asymmetric U-Net architecture with 3D Gaussian splatting for single-view 3D reconstruction. Furthermore, our approach incorporates a novel 3D smoothing filter that constrains the maximum frequency of the 3D representation, effectively mitigating high-frequency artifacts in out-of-distribution rendering. By synergistically combining implicit and explicit representations, our method leverages the strengths of both to enhance reconstruction efficiency and quality. Experiments conducted on the SRN-Cars dataset demonstrate that our framework outperforms existing methods in quantitative metrics and qualitative assessments, achieving higher reconstruction accuracy and smoother surfaces.

Keywords: Deep Learning, Machine Vision, Transfer Learning
Abstract: Knowledge distillation transfers knowledge from a complex network (teacher) to a lightweight network (student).Existing knowledge distillation methods typically employ a loss function comprising task and distillation losses, and they use a hyper-parameter to balance the two losses. However, in this paper, we observe an inconsistency between the gradientdirections of these two losses, which introduces a trade-off between two gradients, hindering the student from learning the entire knowledge. To overcome this challenge, we propose a Universal Knowledge Distillation Framework Based on Decoupling Gradients (DGKD). DGKD breaks the trade-off by decoupling the gradients of the two losses and distributing these to separate branches of the student. Additionally, we introduce a Knowledge Interaction Module (KIM) and Inference-stage Simplifications to optimize our DGKD, enhancing its flexibility and simplicity. Extensive experiments validate the superiority of DGKD. For example, DGKD achieves a +2.80% accuracy improvement on CIFAR-100 for the VGG13-MN-V2 pair.

Keywords: Deep Learning, Neural Networks and their Applications, Computational Intelligence
Abstract: In recent years, predictor-based Neural Architecture Search (NAS) methods have gained significant attention due to their ability to rapidly estimate architecture properties, such as accuracy and latency. However, GNN-based predictor methods often suffer from convergence issues that lead to suboptimal performance, and Transformer-based predictor face quadratic computational complexity which limits their efficiency. To address these challenges, we propose GMP, a GNN-Enhanced Mamba-2 neural predictor that efficiently learns the feature representations of architectures and accurately predicts architecture properties. Specifically, we leverage a novel Mamba-2 model, which employs an advanced State Space Duality (SSD) framework to achieve competitive performance with linear complexity. To adapt to neural prediction tasks and fully improve the predictive capabilities, we propose a bidirectional Mamba-2 module and design a flow-enhanced GNN to integrate with it. To the best of our knowledge, we are the first to apply such SSM architecture to neural predictors. Extensive experiments conducted on NAS-Bench-101, NAS-Bench-201, DARTS, and NNLQP datasets demonstrate the feasibility of our proposed method. Compared to state-of-the-art methods, our GMP not only achieves a significant speedup (7.47 times faster) but also delivers satisfactory performance. Our code is available at https://github.com/estar445/GMP.

Keywords: Deep Learning, Neural Networks and their Applications, Expert and Knowledge-Based Systems
Abstract: Most of recent attention-guided feature masking distillation methods perform knowledge transfer via global teacher attention maps without delving into fine-grained clues. Instead, performing distillation at finer granularity is conducive to uncovering local details supplementary to global knowledge transfer and reconstructing comprehensive student features. In this study, we propose a Spatial-aware Adaptive Masking Knowledge Distillation (SAMKD) framework for accurate object detection. Different from previous feature distillation methods which mainly perform single-scale feature masking, we develop spatially hierarchical feature masking distillation scheme, such that the object-aware locality is encoded during coarse-to-fine distillation process for improved feature reconstruction. In addition, our spatial-aware feature distillation strategy is combined with a masking logit distillation scheme in which region-specific feature difference between teacher and student networks is utilized to adaptively guide the distillation process. Thus, it can help the student model to better learn from the teacher counterpart with improved knowledge transfer and reduced gap. Extensive experiments for detection task demonstrate the superiority of our method. For example, when FCOS is used as teacher detector with ResNet101 backbone, our method improves the student network from 35.3% to 38.8% mAP, outperforming state-of-the-art distillation methods including MGD, FreeKD and DMKD.

Keywords: Deep Learning, Neural Networks and their Applications, Image Processing and Pattern Recognition
Abstract: Metal artifacts frequently emerge in reconstructed computerized tomography (CT) images of cultural relics, primarily due to the presence of metal elements within these cultural relics during the scanning process. It degrades the quality of CT images, posing a challenge to cultural relic restoration efforts. Traditional metal artifact reduction (MAR) methods include projection-domain interpolation and iterative reconstruction. The former offers high computational efficiency but may introduce secondary artifacts, while the latter achieves high accuracy at the cost of increased computational complexity. Recent studies have employed CNN-based image segmentation models to automatically separate metal artifacts from anatomical structures in medical CT images. These models demonstrate superior accuracy compared to conventional approaches. However, compared to medical CT images, the CT images of cultural relics are more complex and blurred at the edges due to their age and the diversity of their constituent materials. This makes artifact removal in cultural relics a challenging task. To address this problem, we first constructed a dataset of CT images of Chinese cultural relics and annotated the cultural relic regions in each image. Secondly, we classified the images into five classes based on the shape of the artifact, then proposed the Annotation and Structural Variability Estimator (ASVE) to assess the complexity of each class for division of test set and training set. Finally, we proposed a CT image artifact removal framework for cultural relics, utilizing a domain-specific pre-trained encoder and a designed attention-based scSEConv Block to enhance the artifact awareness and boundary recovery capabilities of the Unet. The experimental results showed that our model achieved an IOU of 0.8445 and a recall of 0.9179 on the test set, outperforming other compared segmentation models.

Keywords: Computational Intelligence, Expert and Knowledge-Based Systems, Hybrid Models of Computational Intelligence
Abstract: About 50% of Type 2 diabetes patients develop diabetic neuropathy, which can damage nerves throughout the body, including those controlling the vocal cords, leading to issues like vocal fold paralysis, hoarseness, or vocal strain. Therefore, this research pioneers automated diabetes diagnosis using voice and speech recordings, with the attempt to provide a non-invasive, easily accessible tool for early detection of diabetes and prediabetes conditions. Firstly, five speech/voice datasets have been generated, including audio recordings of vowel letters (‘A’, ‘E’, ‘O’) and high and low speed countings of digital numbers (1-20), provided by participants with and without diabetes conditions. Initially, Mel-frequency Cepstral Coefficient (MFCC) features from various voice clips of diabetic and non-diabetic participants are extracted. 1D Convolutional Neural Network (1D CNN) and Bidirectional Long Short-Term Memory (BiLSTM) incorporating various hybrid attention mechanisms are proposed for diabetes classification. Specifically, attention methods, i.e. Squeeze and Excitation Block (SE), Convolutional Block Attention Module (CBAM), and Global Context (GC) blocks, as well as their hybrid strategies, i.e. SE+CBAM and SE+CBAM+GC, have been proposed to extract the most important acoustic features and incorporated with BiLSTM and 1D CNN, respectively. Moreover, within and cross-dataset hybrid attention feature maps of diverse resulting networks are constructed to further emphasize the most crucial disease-indicative patterns from a global perspective, leading to the best accuracy rate of 82%. The empirical studies suggest that speech analysis could be a viable method for the preliminary diagnosis of diabetes.

Keywords: Neural Networks and their Applications
Abstract: Event causality identification (ECI) aims to predict the causal relationships between events mentioned in text. Although existing prompt-based methods have achieved significant improvements on the ECI task, these methods face the static prompt challenge, i.e., the model encodes all examples within the prompt equally. In this paper, we proposes a Fusion Dynamic In-Context in Decoder (FDICD) model. Specifically, FDICD is inspired by the in-context learning paradigm, encoding dynamic in-context examples separately and then fusing this information in the decoder. Additionally, we incorporate external knowledge to enhance the semantic representation of causal tokens in the prompt. We conducted extensive experiments on two widely used benchmarks, and the results demonstrate that FDICD achieve promising improvements over baseline methods.

Keywords: Neural Networks and their Applications
Abstract: 用于软件缺陷预测的深度学习技术 提高大规模缺陷检测效率 项目。 但是，预测结果通常包括 一定比例的误报和遗漏， 需要手动验证实际缺陷。

这导致研究人员对以下方面的兴趣增加 开发方法以自动确定 可疑缺陷的真实性。 然而，现有的自动缺陷方法 识别是有限的，因为它们通常无法识别 有效区分真伪缺陷 阳性。 在本文中，我们提出了一种自动方法 验证基于深度学习的预测结果 软件缺陷模型，利用静态分析 技术。 这种方法有效降低了人工成本 与验证潜在缺陷相关。 它通过聚合来自 程序入口点到可疑位置 缺陷。 该方法涉及编译和完善 潜在缺陷的执行路径轨迹， 对结构化采用筛选和优先级 分析。 符号执行用于验证 可疑缺陷。 在Ö

Keywords: Neural Networks and their Applications, AI and Applications, Deep Learning
Abstract: Pedestrian trajectory prediction is important in the research of mobile robot navigation in environments with pedestrians. Most pedestrian trajectory prediction algorithms require as input complete historical trajectories. If a pedestrian is unobservable in any frame in the past, then its historical trajectory becomes incomplete and the algorithm does not predict its future trajectory. To address this limitation, we propose STGN-IT, a spatio-temporal graph network allowing incomplete trajectory input. STGN-IT is able to predict the future trajectories of pedestrians with incomplete historical trajectories. STGN-IT uses the spatio-temporal graph with an additional encoding method to represent the historical trajectories and observation states of pedestrians. Moreover, STGN-IT introduces static obstacles in the environment that may affect the future trajectories as nodes to further improve the prediction accuracy. A clustering algorithm is also applied in the construction of spatio-temporal graphs. Experiments on public datasets show that STGN-IT outperforms state-of-the-art algorithms.

Keywords: Neural Networks and their Applications, Application of Artificial Intelligence, Machine Learning
Abstract: Graph Convolutional Networks (GCNs) have excelled in skeleton-based action and gesture recognition due to their strong feature extraction capabilities. However, existing methods often overlook the aggregation of motion information from low-level to high-level features. To address this, we propose an Adaptive Memory Multi-level Feature Graph Convolutional Network (AMF-GCN), which preserves low-level features during propagation and progressively passes them to the output. We also introduce a Motion Capture Module (MCM) to capture intrinsic relationships between motion and spatial features, and a Dynamic Fusion Graph Convolution (DFGC) algorithm to efficiently transmit multi-level features while preserving low-level information. Additionally, a Motion Feature Enhancement Mechanism (MFEM) uses attention to highlight important features. Experimental results indicate that AMF-GCN demonstrates competitive performance compared to mainstream models, achieving 93.5% accuracy on NTU RGB+D (X-sub) and 97.5% on X-view, while delivering favorable results across multiple benchmark datasets.

Keywords: Neural Networks and their Applications, Deep Learning, AI and Applications
Abstract: In this study, we propose a novel deep neural model, CWTLNet, specifically designed to address the unique characteristics of cryptocurrency trading, including significant short-term volatility, which operates around the clock, and is prone to sudden price jumps and drops. The architecture consists of two branches. The CWT branch employs a continuous wavelet transform (CWT) to extract time-frequency feature maps from the input time series. These feature maps are then processed by a stack of CWT Blocks, which are composed of residual structures with two Conv1D layers, allowing the model to capture local, ultra-short-term patterns in the data. The linear branch first decomposes the time series into its trend and residual components. Each component is modeled separately using linear models and subsequently recombined. The linear branch is particularly effective at capturing the periodic patterns embedded in the time series. Finally, the outputs of the two branches are fused through a gated mechanism with a temperature coefficient, enabling adaptive weighting of the branch outputs. Experimental results demonstrate that CWTLNet outperforms commonly used models — including LSTM, Transformer, and DLinear — on minute-level trading data for seven different cryptocurrencies. For example, The performance of CWTLNet on the Bitcoin dataset, in terms of Mean Squared Error (MSE) and Mean Absolute Error (MAE), shows improvements of at least 2.7% and 1.7%, respectively, compared to the aforementioned three models.

Keywords: Neural Networks and their Applications, Deep Learning, AI and Applications
Abstract: Recent studies have shown that although DNNs perform well on visual tasks, they are still vulnerable to clean-label backdoor attacks. As poisoned samples come from the target class, the model learns both original and trigger features, weakening the association between the trigger and the target label and thereby reducing the attack success rate (ASR). To address this issue, we propose a novel clean-label backdoor attack framework, named CLEAR, i.e., Clean-Label Embedding Attack with Representation-Guidance, which strengthens the correlation between triggers and target labels while maintaining high stealth. Specifically, the“benign-label push” mechanism perturbs clean samples selected from the target class (the samples designated for poisoning), pushing them away from their original class in the feature space, while the “target-label pull” mechanism pulls the poisoned sample representations closer to the target class through saliency-guided embedding. CLEAR consists of three key components: integrating a diffusion model with PGD to generate natural but semantically perturbed adversarial samples; selecting backdoor-susceptible samples near the decision boundary based on classification loss; and embedding triggers into high-saliency regions identified using a Feature Pyramid Network (FPN) combined with a local self-attention mechanism. Experimental results on CIFAR10 and GTSRB with ResNet18 and VGG16 demonstrate that CLEAR generally improves ASR while maintaining strong stealth.

Keywords: Evolutionary Computation, Metaheuristic Algorithms, Optimization and Self-Organization Approaches
Abstract: In evolutionary computation, the study subject for algorithms capable of effectively resolving multi-objective optimization problems remains at the forefront of research. Most existing multi-objective evolutionary algorithms (MOEAs) often struggle with maintaining diversity and avoiding premature convergence when solving complex or high-dimensional optimization problems. This study proposes an innovative iteration of the Non-dominated Sorting Genetic Algorithm III (NSGA-III), which infuses chaotic dynamics alongside estimated convergence point strategies to enhance solution quality and diversity. Our research undertakes a comprehensive evaluation of this enhanced algorithm against a suite of benchmark problems. We compare it with its predecessor, Non-dominated Sorting Genetic Algorithm II (NSGA-II), and variations incorporating chaotic dynamics alongside estimated convergence point strategies. The performance analysis results indicate that NSGA-III, using chaotic dynamics and estimated convergence point strategies across most test functions, demonstrates superior performance over traditional and singly-enhanced MOEAs. The statistical analysis strongly suggests that the dual enhancements embedded in the novel algorithm contribute significantly to its optimization capabilities.

Keywords: Machine Learning, Evolutionary Computation, Neural Networks and their Applications
Abstract: Balancing exploration and exploitation is a fundamental challenge in reinforcement learning. In Maximum Diffusion Reinforcement Learning (MaxDiff RL), this balance is regulated by a temperature parameter that controls exploration, but optimizing it across different tasks is challenging. To tackle this issue, we propose a method that dynamically adjusts the temperature parameter via evolutionary algorithms during training. The training process is divided into multiple stages, where different optimization strategies are applied to adaptively evolve the temperature parameter, ensuring a proper balance between exploration and exploitation. We evaluate our method on two continuous control tasks in robotics, i.e., Swimmer and HalfCheetah. Experimental results demonstrate that our method outperforms baseline algorithms with randomly initialized or default temperature parameters, achieving faster convergence and higher cumulative rewards, particularly in tasks demanding greater exploration.

Keywords: Evolutionary Computation, Metaheuristic Algorithms, Machine Learning
Abstract: Surrogate-assisted evolutionary algorithms (SAEAs) address expensive optimization problems using surrogate models to approximate costly evaluation functions. To develop effective SAEAs, evaluating the generalization performance of surrogate models during the search process is crucial. In practice, surrogate accuracy is typically estimated using validation accuracy computed from a holdout subset of the limited training data. This estimate is then used to guide parameter tuning or surrogate model selection. However, whether validation accuracy reliably reflects test accuracy on unseen data is unclear, especially when only a small training dataset is available in SAEAs. Despite its practical importance, the relationship between validation and test accuracies in SAEAs has not been systematically investigated. This study examines the reliability of validation accuracy as an indicator of surrogate model generalization. Toward this goal, we implement surrogate-assisted particle swarm optimization with a generation-based strategy and employ radial basis function (RBF) models, which are commonly used in recent SAEA research. Experiments are conducted on CEC 2015 benchmark problems to analyze the relationship between validation and test accuracies. The results reveal that validation accuracy does not consistently correlate with test accuracy, suggesting that it is not a reliable indicator of surrogate model generalization in SAEAs.

Keywords: Application of Artificial Intelligence, Deep Learning
Abstract: Path planning is an essential task for the mission execution of unmanned surface vehicle (USV). However, existing advanced techniques based on deep reinforcement learning (DRL) often suffer from low learning efficiency and insufficient environmental perception from single-sensor configurations. To address these issues, this paper proposes a hybrid framework named A*D3QN, which integrates the heuristic efficiency of the A* algorithm with the adaptive decision-making of a dueling double deep Q-network (D3QN), associated with the multimodal data fusion for precise environment modeling. The proposed A*D3QN incorporates prior knowledge from the global paths generated by A* to initialize and guide the D3QN learning process. The prior knowledge, formatted as RL transition tuples, is used in the reward function design and the N-step prioritized experience replay, which significantly accelerates overall learning efficiency. Moreover, an improved D3QN architecture is designed to dynamically fuse visual data and navigation states via a cross-entropy attention mechanism, enabling multimodal perception in partially unknown environments. Extensive experiments across three scenarios with varying obstacle densities demonstrate that A*D3QN significantly outperforms state-of-the-art DRL baselines. Ablation studies further validate the necessity of each component.

Keywords: Fault Monitoring and Diagnosis, Manufacturing Automation and Systems, Cyber-physical systems
Abstract: Rapid technological advances in recent years have established Big Data, Industrie 4.0, Internet of Things (IoT) and Artificial Intelligence (AI) as data-driven concepts that are increasingly being implemented and proving value in real-world scenarios, transforming the future across all sectors—including manufacturing. Especially in the area of industrial automation, these fast-paced and transformative changes collide with a long-established industry. Despite the current state-of-the-art in technology, the lifetime of equipment is measured in decades—with the need to provide long-term support even to outdated (legacy) systems in heterogeneous production environments. Incorporating changes in such systems is challenging and mostly driven by software due to its adaptability. This paper contributes a generalized, lightweight approach for integrating machine-level data collection and analysis with agent-based systems, aiming to facilitate the adoption of CPPS in the industry and easing the integration with legacy control systems. As a use case for production performance optimization, the monitoring and detection of outliers for interconnected transportation systems has been defined. A key feature of the approach is that it utilizes only data from presence sensors. Thus, only minimal information about the system and no further configuration or development is required while maintaining full functionality. The generalizability and transferability of this approach have been validated using two representative laboratory production systems, demonstrating the potential of PLC-based data (pre-) processing and its scalability to multiple IEC 61131-3 conform control applications.

Keywords: Distributed Intelligent Systems, Adaptive Systems, Fault Monitoring and Diagnosis
Abstract: Accurate financial forecasting demands scalable models that adapt to volatility in high-dimensional, temporally dynamic markets. Traditional approaches—such as centralized Hidden Markov Models (HMMs) and GARCH (Generalized Autoregressive Conditional Heteroskedasticity)—struggle with real-time responsiveness and robustness. We propose HMM-DFC, a distributed framework that integrates Fuzzy C-Means (FCM) clustering with HMMs across temporally partitioned nodes to capture evolving market regimes. To enhance adaptability and consistency, we introduce two novel modules: Volatility-Aware Transition Refinement (VATR) for sharper regime transitions and Entropy-Based Node Regularization (EBNR) for stable cross-node synchronization. Extensive evaluation on S&P 500 data shows that HMM-DFC outperforms GARCH, centralized HMMs, and distributed FCM in volatility detection and clustering quality. The framework supports real-time, and scalable forecasting for applications in trading.

Keywords: Fault Monitoring and Diagnosis, Communications, Homeland Security
Abstract: Domain fronting is a covert communication technique, which evades detection by connecting with legitimate domains to imitate normal network traffic. But the imitation is flawed, so current detection methods usually treat domain fronting as abnormal traffic. However, these methods show very low precision due to normal-abnormal traffic imbalance in practice. In the paper, we find that domain fronting's imitation is limited to popular applications (apps), such as Chrome and Firefox browser. Thus, we mitigate traffic imbalance by defining domain fronting detection as an app discrimination problem, rather than previous anomaly detection task.

According to the revised definition, we propose FakeApp, a high-precision method for domain fronting detection in real networks. Using frequent item analysis, FakeApp first extracts the imitated app information from domain fronting tools as symbolic features. Then, through deep neural networks, it discriminates whether traffic belongs to genuine or spoofed apps. Finally, FakeApp integrates symbolic features and neural networks together to identify domain fronting in real networks. Evaluations over 2 million flows show that the precision of FakeApp is over 95%, far surpassing state-of-the-art methods on four domain fronting tools. These results also indicate that we have effectively mitigated the traffic imbalance issue.

Keywords: Intelligent Power Grid, Infrastructure Systems and Services, Cyber-physical systems
Abstract: This study investigates how fuzzy-logic humidity thresholds influence HVAC energy of a residential house. A validated fuzzy controller is embedded in the Vertical City Weather Generator (VCWG v3.0.0) and subjected to a one-at-a-time sensitivity analysis on the four relative-humidity limits—Low-Low, Low, High and High-High—while all other inputs are held constant. Results show that sensible loads remain essentially unchanged regardless of threshold adjustments, whereas humidification and dehumidification respond strongly and in opposite directions to shifts at the dry and humid ends of the comfort band. Widening the allowable humidity range therefore offers a straightforward means to cut latent energy without affecting heating or cooling demand. The findings highlight which fuzzy thresholds matter most and provide practical guidance for tuning rule-based HVAC control in cold, dry climates.

Keywords: Fault Monitoring and Diagnosis, Intelligent Power Grid, Smart Metering
Abstract: Arc fault diagnosis constitutes a critical challenge in fault detection, focusing on rapid and precise identification of arc-induced safety risks in power systems. Previous methodologies hard to strike a balance between real-time performance and detection accuracy in arc fault diagnosis systems. To address this challenge, we propose a linear-time Mamba-based model enhanced with a Spatial Awareness Module(SAM), achieving real-time DC arc fault diagnosis. Specifically, our approach leverages a state-space model (SSM) framework and employs a hardware-aware parallel algorithm for efficiency. To further improve accuracy while maintaining the computational efficiency of the base model, we integrate a spatial awareness module to capture global features, enabling precise fault diagnosis. Experimental results demonstrate that our method achieves 96.72% accuracy with a 1.87 ms response time, making it highly suitable for industrial applications where rapid and reliable arc fault diagnosis is critical. This advancement holds significant promise for enhancing safety in industrial electrical systems.

Keywords: Fault Monitoring and Diagnosis, Large-Scale System of Systems, Discrete Event Systems
Abstract: 随着云存储系统的发展，“fail-slow” 越来越大 注意，在此期间，驱动器 I/O 线程会体验 延迟响应和系统性能始终如一 低于预期。本文介绍了 Patch-sfp，一种 检测云中 fail-slow 的实用框架 存储系统。它采用基于 PatchTST 的预测模型 LPM 来 预测未来的驱动器延迟趋势，并开发 延迟阈值设计机制，准确识别 fail-slow 事件。

考虑到驱动器的负载不同，动态分块 策略旨在从 不同工作环境下的驱动器，增强了 LPM 捕获一段时间内依赖关系的能力。此外，还引入了 MoH-Attention 机制 提高 LPM 学习复杂关系的能力 在驱动器数据之间。实验结果表明，Patch-sfp 有效 检测云存储系统中的故障慢速，性能优于 现有的 fail-slow 检测方法包括 性能、适应性和健壮性。

Keywords: Image Processing and Pattern Recognition, Machine Vision, Deep Learning
Abstract: Accurate classification of leaf diseases is crucial for plant health and effective crop management. Existing deep learning approaches are predominantly categorized into Convolutional Neural Network (CNN)-based and Vision Transformer (ViT)-based methods. However, inherent limitations in both approaches constrain further performance gains. CNNs excel at capturing local lesion details but struggle with long-range dependencies. In contrast, ViTs leverage self-attention to model global feature relationships but often overlook fine-grained local information. Moreover, most models rely solely on three-channel (RGB) inputs, underutilizing the texture details present in grayscale data. To address these problems, we propose a hybrid CNN-Transformer model enhanced with a novel Gray-Aware (GA) Attention Module and an improved Residual Convolutional Block Attention Module (Rs-CBAM). Specifically, the hybrid architecture effectively balances fine-grained detail preservation and global contextual understanding. GA strengthens texture representation, while Rs-CBAM further enhances attention to critical regions. Comparative experiments conducted on the PlantVillage and AI Challenger 2018 datasets demonstrate that our model outperforms existing models. Ablation studies further confirm the effectiveness of each proposed enhancement. Overall, the proposed approach provides a promising direction for fine-grained plant disease classification, and GA shows potential in broader image processing tasks requiring enhanced texture representation. The code is available on GitHub: https://github.com/Governeson/GA-CCB.

Keywords: Image Processing and Pattern Recognition, Machine Vision, Multimedia Computation
Abstract: Grayscale resolution constitutes a fundamental performance metric in digital imaging systems for computer vision applications such as natural light imaging or medical imaging. Although frame accumulation has been conventionally employed to enhance grayscale resolution, existing implementations demand stringent hardware capabilities and operational conditions, thereby restricting their practical applicability. This research introduces an innovative grayscale super-resolution technique that integrates frame accumulation with shaped-function signal, mitigating analog-to-digital converter quantization errors. Specifically, our approach involves superimposing an auxiliary illumination with periodic luminance onto the original image signal. The composite signal is continuously sampled by an imaging sensor, with subsequent grayscale reconstruction achieved through an innovative algorithm developed in this research. The experimental results demonstrate that our method is effective in enhancing the quality of images captured by natural light digital cameras by improving grayscale resolution. Quantitative analysis reveals advancements across multiple image quality metrics, including structural similarity index (with +4% gains) and peak signal-to-noise ratio (with +3.5db gains). Furthermore, this methodology enhances operational convenience by addressing two key constraints in existing approaches: the requirement for precise waveform control of auxiliary illumination and the necessity for phase synchronization between auxiliary optical signals and image sensor sampling cycles. These features significantly facilitate the deployment of this method in practical application scenarios.

Keywords: Image Processing and Pattern Recognition, Machine Vision, Neural Networks and their Applications
Abstract: The rapid growth of user-generated content (UGC) on social platforms has created a pressing need for effective outdoor video quality assessment. Evaluating video quality in uncontrolled environments is challenging due to the absence of reference videos and distortions caused by compression and transmission artifacts, limiting the applicability of traditional metrics. In this paper, we propose a novel no-reference video quality assessment model that reduces computational complexity by identifying frames with significant visual changes. Optical flow detection is then applied to these frames to capture perceptually important regions, enabling focused processing. Experiments on three public outdoor video quality databases—KoNViD-1k, LIVE-Qualcomm, and CVD2014—demonstrate the effectiveness of our method. Furthermore, ablation studies highlight the critical roles of frame selection and optical flow-based region analysis in improving model performance. The source code is available at https://github.com/Hsiang417/SCCOME.

Keywords: Image Processing and Pattern Recognition, Neural Networks and their Applications
Abstract: Although SAM-based single-source domain generalization models for medical image segmentation can mitigate the impact of domain shift on the model in cross-domain scenarios, these models still face two major challenges. First, the segmentation of SAM is highly dependent on domain-specific expert-annotated prompts, which prevents SAM from achieving fully automated medical image segmentation and therefore limits its application in clinical settings. Second, providing poor prompts (such as bounding boxes that are too small or too large) to the SAM prompt encoder can mislead SAM into generating incorrect mask results. Therefore, we propose the FA-SAM, a single-source domain generalization framework for medical image segmentation that achieves fully automated SAM. FA-SAM introduces two key innovations: an Auto-prompted Generation Model (AGM) branch equipped with a Shallow Feature Uncertainty Modeling (SUFM) module, and an Image-Prompt Embedding Fusion (IPEF) module integrated into the SAM mask decoder. Specifically, AGM models the uncertainty distribution of shallow features through the SUFM module to generate bounding box prompts for the target domain, enabling fully automated segmentation with SAM. The IPEF module integrates multiscale information from SAM image embeddings and prompt embeddings to capture global and local details of the target object, enabling SAM to mitigate the impact of poor prompts. Extensive experiments on publicly available prostate and fundus vessel datasets validate the effectiveness of FA-SAM and highlight its potential to address the above challenges.

Keywords: Image Processing and Pattern Recognition, Neural Networks and their Applications, Deep Learning
Abstract: Detecting whether two images depict the same location, totally or partially, from different viewpoints is a key challenge in mobile robotics known as Visual Loop Detection (VLD). VLD is an essential component of visual Simultaneous Localization and Mapping (SLAM), enabling robots to estimate their position while constructing a consistent map of the environment. Many existing approaches rely on global image descriptors or signatures such as Bag of Words (BoW), Vector of Locally Aggregated Descriptors (VLAD), and Hash-based Loop Closure (HALOC), which encode images into compact feature vectors for efficient comparison.

Recent deep learning methods have surpassed these classical techniques but typically require large datasets and substantial computational resources, particularly during training. Addressing these challenges is crucial in underwater robotics, which is the focus of this work, as data collection is costly and onboard computational resources are often limited.

This paper proposes a hybrid approach that combines deep learning with a signature-based method. Specifically, a triplet-based neural network is trained on HALOC descriptors to learn a discriminative embedding space where images of the same location are mapped closer together while different locations remain distinct. By operating directly on HALOC signatures, the model remains lightweight, thus being trainable on small datasets and involving low computational requirements.

The proposal is experimentally evaluated with underwater imagery grabbed in coastal areas of Mallorca.

Keywords: Machine Learning, Neural Networks and their Applications, Deep Learning
Abstract: Hybrid modeling is an approach that integrates data-driven models with physics-based models to compensate for discrepancies between model predictions and real-world behaviors, enabling us to obtain accurate dynamical system models. Previous studies have pointed out that in the training phase, applying constraints on the data-driven parts, called regularizers, is necessary when the physics models include unidentified parameters. However, the appropriate formulation of such regularizers remains unclear. In this work, we conducted experiments to investigate effective ways of introducing regularizers. We found that we can strike the balance between parameter estimation and state prediction by utilizing correlational information between physics and whole models for regularization. Furthermore, we extended the hybrid modeling approach to stochastic differential equations (SDEs), proposing a novel SDE learning algorithm using the probability distributions for the regularizers. Our experimental results showed that the proposed approach improved parameter estimation accuracy and enabled us to acquire correct stochastic dynamics.

Keywords: Machine Learning, AI and Applications, Application of Artificial Intelligence
Abstract: Explaining autonomy is becoming a crucial factor in the design of trustworthy autonomous platforms in both transport and smart living sectors. Interpretable reinforcement learning (RL) is an emerging research area that aims to explain why an autonomous platform adopts an action or set of actions. However, the state-of-the-art has focused on the design of explainable tools as independent modules that are not involved in the decision-making process of the RL agent. In this paper, we propose a novel belief-desire-intention RL (BDI-RL) approach that incorporates the explainable module as a belief model that enhances the learning capabilities of the RL as well as actions interpretability. To this end, we combine the merits of Dyna-Q algorithm as backbone RL model and belief maps as explainable element. The combined contribution of these models provides a robust model that emulates better the reasoning process of humans by leveraging beliefs and on-line agent-environment interactions. Simulations experiments are conducted in a grid environment of different sizes and obstacles. Comparisons are also provided to show the benefits of the proposed methodology.

Keywords: Machine Learning, AI and Applications, Application of Artificial Intelligence
Abstract: Federated learning effectively safeguards data privacy by enabling local model training across multiple clients. However, it remains susceptible to label flipping attacks, which can significantly degrade the global model's performance, even when the proportion of malicious clients is small. Existing defense methods often rely on assumptions regarding client data distribution or attacker proportions, which limits their effectiveness in real-world scenarios characterized by data heterogeneity and unknown attack scale. This paper introduces a novel and robust defense mechanism, FedDLFA. FedDLFA is grounded in a key insight: Due to the adversarial nature of the training objective, malicious clients exhibit significantly distinct neural activation patterns under standardized inputs compared to normal clients. FedDLFA extracts neural activation vectors from all client models, calculates their cosine similarity, constructs a similarity matrix, and applies clustering techniques to divide clients into two groups. Subsequently, it employs a density-size joint scoring mechanism to identify potential clusters of malicious clients. Experiments conducted on the MNIST, FMNIST, and CIFAR10 datasets, under both IID and Non-IID settings, demonstrate that FedDLFA achieves superior accuracy and effectively mitigates attack success rates compared to existing state-of-the-art methods.

Keywords: Machine Learning, Computational Intelligence, AI and Applications
Abstract: Autonomous systems operating in high‑stakes search‑and‑rescue (SAR) missions must continuously gather mission‑critical information while flexibly adapting to shifting operational priorities. We propose CA‑MIQ (Context‑Aware Max‑Information Q‑learning), a lightweight dual‑critic reinforcement learning (RL) framework that dynamically adjusts its exploration strategy whenever mission priorities change. CA‑MIQ pairs a standard extrinsic critic for task reward with an intrinsic critic that fuses state‑novelty, information‑location awareness, and real‑time priority alignment. A built‑in shift detector triggers transient exploration boosts and selective critic resets, allowing the agent to re‑focus after a priority revision. In a simulated SAR grid‑world, where experiments specifically test adaptation to changes in the priority order of information types the agent is expected to focus on, CA‑MIQ achieves nearly four times higher mission‑success rates than baselines after a single priority shift and more than three times better performance in multiple‑shift scenarios, achieving 100% recovery while baseline methods fail to adapt. These results highlight CA‑MIQ’s effectiveness in any discrete environment with piecewise‑stationary information‑value distributions.

Keywords: Machine Learning, Computational Life Science, Neural Networks and their Applications
Abstract: Fast learning remains a fundamental challenge in deep learning. Inspired by the biological insect olfactory system, we study fast learning in a model with structured random connectivity, adaptive thresholds, and sparsely activated neurons. In contrast to conventional random feature learning, our model regulates neuronal activity through dynamic threshold adaptation. Experimental results indicate that the model performs well with fewer training iterations while improving generalization and robustness compared to a traditional MLP. The model integrates principles from random feature learning and sparse coding, similar to the mechanisms observed in the insect olfactory system, particularly in the Antennal Lobe and Kenyon cells. These findings support that our Bio-Inspired Olfaction Neural Network (BONN) is a biologically plausible and computationally efficient alternative to fast learning in neural networks.

Keywords: Multimedia Computation, Neural Networks and their Applications, Transfer Learning
Abstract: Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) can deliver outstanding performance when appropriately optimized for specific tasks. Although these architectures typically require substantial computational resources, lightweight CNN models may achieve comparable or even superior efficiency in well-defined and constrained scenarios. Therefore, the effectiveness of the approach depends on the careful tuning of the hyperparameter and the deliberate selection of an architecture aligned with the characteristics of the target problem. This work presents LIGA, a LIghtweight CNN-based architecture designed to classify popular music Genres from the Amazonian region, including andean, brega, carimbó, cumbia, merengue, pasillo, salsa, and vaqueirada, originating from countries such as Bolivia, Brazil, Colombia, Ecuador, French Guiana, Peru, the Dominican Republic, and Venezuela. In addition to low computational resource usage and improved training speed, LIGA achieved higher precision and accuracy compared to the EfficientNet, MobileNet, ResNet, VGG, Xception, MobileViT, and MaxViT models.

Keywords: Human-Computer Interaction, Human-Machine Cooperation and Systems, Human-Machine Interface
Abstract: Plant simulation models play a significant role in agricultural production and ecological research. However, traditional simulation software is often highly specialized and involves complex parameters, posing substantial entry barriers for non-expert users. With the emergence of large language models (LLMs) and advancements in their function calling capabilities, we propose an interactive simulation system that integrates LLM-based function calling with plant simulation models. This system uses the function-calling capabilities of large models (e.g., DeepSeek-V3) to translate natural language instructions into simulation parameters and commands for the GreenLab platform, thereby enabling interactive control of plant growth models. Furthermore, the system provides real-time visualization of simulation outcomes through a front-end 3D interface. Preliminary results demonstrate that this approach effectively lowers the usability threshold of plant simulation models, allowing non-expert users to conveniently conduct simulations and analyze results using natural language.

Keywords: Human-Machine Cooperation and Systems, Human Factors, Assistive Technology
Abstract: Accurately estimating the phase of oscillatory systems is essential for analyzing cyclic activities such as repetitive gestures in human motion. In this work we introduce a learning-based approach for online phase estimation in three-dimensional motion trajectories, using a Long Short-Term Memory (LSTM) network. A calibration procedure is applied to standardize trajectory position and orientation, ensuring invariance to spatial variations. The proposed model is evaluated on motion capture data and further tested in a dynamical system, where the estimated phase is used as input to a reinforcement learning (RL)-based control to assess its impact on the synchronization of a network of Kuramoto oscillators.

Keywords: Human-Machine Cooperation and Systems, Human Factors, Human-Collaborative Robotics
Abstract: Smooth and safe interactions between pedestrians and Autonomous Mobile Robots (AMRs) are crucial for integrating robotic systems into shared environments. Previous studies on external Human-Machine Interfaces (eHMIs) often employed static information presentation, neglecting dynamic interaction contexts. This study investigates the effectiveness of gaze-based nudging—subtle, intuitive communication via gaze behavior—in facilitating pedestrian role selection (leader or follower) during perpendicular crossing interactions with AMRs. Virtual reality (VR) experiments using Unity and Cybershoes are conducted to evaluate two gaze patterns generated by the AMR: `Leader’s gaze', a brief gaze directed toward pedestrians in the early stages of interactions, and `Follower's gaze', a gaze keeping track of pedestrians throughout the interaction until crossing completion. The impact of gaze timing (Early/Late) and initial positional relationships between pedestrians and AMRs (initial difference in TTCP) are systematically analyzed. The results indicate that gaze nudging significantly enhances pedestrians' subjective ratings of safety, smoothness, and understanding of the robot's intention compared to no-gaze conditions.Leader's gaze effectively encourages pedestrians to adopt the follower role under conditions favoring AMR priority (small or negative TTCP difference), whereas Follower's gaze promotes pedestrians adopting the Leader role under conditions naturally favoring pedestrian priority (larger TTCP difference). Additionally, the effectiveness of gaze nudging strongly depends on interaction timing, with early-stage gaze presentations exhibiting greater influence. These findings confirm the potential of gaze-based nudging as a nonintrusive, context-sensitive strategy for pedestrian-AMR interactions, emphasizing the importance of precisely timed gaze presentations for facilitating pedestrians' natural and intuitive role selection.

Keywords: Human-Machine Cooperation and Systems, Human-Machine Interaction, Information Systems for Design
Abstract: Interpreting complex engineering drawings requires substantial manual effort in industrial workflows, with engineers spending hundreds of hours verifying correlations between visual elements and structured specifications across thousands of documents. This challenge is particularly acute in the Engineering, Procurement and Construction (EPC) industries, where interpretation errors are propagated into costly procurement and construction mistakes. We propose a cybernetic systems approach using Large Multimodal Models (LMMs) as cognitive partners in engineering documentation workflows, enhancing human capabilities through intelligent assistance in verification and information extraction tasks. To validate this solution, we systematically evaluated five LMMs on 60 piping isometric drawings with varying template structures, measuring both optical character recognition accuracy and correlation capabilities between drawings and their associated Bills of Materials. The results demonstrated significant performance variation between systems, with Claude 3.5 Sonnet and GPT-4o achieving greater accuracy 70%, while open source alternatives faced challenges with complex layouts and ambiguous visual information. These findings establish benchmarks for human-machine collaboration in engineering documentation processing and provide a framework for integrating intelligent systems into technical workflows across multiple industrial domains.

Keywords: Human-Machine Cooperation and Systems, Interactive Design Science and Engineering, Assistive Technology
Abstract: In the landscape of systems engineering, the integration of Generative Artificial Intelligence (GenAI) offers transformative potential for Requirements Engineering (RE). This study introduces Self-Generative Requirements Engineering (Self-GenRE), a human-AI reflexive methodology leveraging GenAI with humans in the loop to facilitate continuous requirement gathering improvement in the RE process. The methodology is demonstrated through two use cases: 1) a reflexive application of GenRE to itself, utilizing NLP4ReF – a GenRE tool designed to automate system requirements generation from a partial set of initial requirements, and 2) the application of the enhanced GenRE tool to an IoT project. Self-GenRE streamlines requirement gathering, demonstrating a significant reduction in requirements gathering time from days to minutes, while increasing previously overlooked requirements identification – 42% before and 71% after enhancement. These improvements illustrate the Self-GenRE's effectiveness, enabling GenRE tools to enhance themselves based on the system of interest (SoI) and its requirements, thereby continuously improving both the process and the GenRE tools, ensuring alignment with evolving project needs. Presently, modifying GenRE tools to adhere to new requirements is performed manually; however, we envision future automation of this process, paving the way for self-improving agentic generative RE. The Self-GenRE methodology, process, and tools were developed within a Model-Based Systems Engineering (MBSE) framework, facilitating efficient design modifications. This research highlights the efficacy of integrating GenAI methodologies and processes into systems engineering, paving the way for more robust and adaptable frameworks capable of addressing the evolving challenges in modern technology development and innovation.

Keywords: Human-Machine Cooperation and Systems, Interactive Design Science and Engineering, Augmented Cognition
Abstract: Engineering, Procurement, and Construction (EPC) projects often encounter critical inefficiencies in Interdisciplinary Check (IDC) processes, where engineers manually validate design consistency across drawings and specifications, resulting in bottlenecks in large-scale projects. This paper introduces a proof-of-concept(PoC) human-AI cooperative framework for automating IDC workflows. The framework combines deep learning for document processing, heuristic algorithms for feature extraction, and a Large Multimodal Model (LMM) for cross-domain validation, while preserving engineering decision authority. The framework extracts key engineering data, links components across disciplines, and checks compliance with industry standards and client requirements. Testing on controlled data from a single EPC project (with 25 different engineering drawings) successfully identified all design discrepancies within the predefined validation scope, demonstrating notable workflow efficiency potential. As Phase 1 of a three-phase research cycle, this confirms technical feasibility before expanding to multiple projects and company-wide deployment, showing how human-AI collaboration can revolutionize engineering validation while preserving oversight of engineering expertise.

Keywords: Control of Uncertain Systems
Abstract: Prescribed-time stable systems frequently suffer from infinite gain issues that lead to practical infeasibility. This paper investigates the problem of stabilization for uncertain Euler-Lagrangian systems and proposes a singularity-free prescribed-time stabilization controller. Based on the time space deformation approach, the designed controller ensures singularity avoidance, effectively eliminating the existence of infinite gain. Besides, the considered Euler-Lagrangian systems are subject to unknown nonlinear functions and derivative-bounded external disturbances with unknown bounds. To achieve system stabilization under such uncertainty issue, this paper formulates multiple sliding manifolds with prescribed-time stability. The controller's effectiveness is validated through simulation studies on a rendezvous formation problem.

Keywords: Control of Uncertain Systems, Adaptive Systems, System Modeling and Control
Abstract: Numerous practical applications have revealed significance efficiency and high robustness of the Active Disturbance Rejection Control (ADRC) scheme in the problem of tracking control of highly uncertain and perturbed dynamical systems. The error-based ADRC (eADRC) allows additionally reducing the information content about a reference trajectory, where the availability of the reference signal's time derivatives is not required. The main limitation, however, of the conventional ADRC (also in the error form) applied with a linear extended state observer (LESO) comes from a need of a wide enough observer's bandwidth, which leads to large observer gains, and in the case of a noisy feedback can lead to a control performance degradation and to an excessive amplification of high parasitic frequencies in a control signal. In this paper we propose to mitigate the above mentioned limitation by augmenting the eADRC with an add-on mechanism called the emph{residual compensation} which enables reducing a bandwidth of LESO while keeping (or even improving) the resultant tracking control performance in both a~tracking accuracy and a~control cost. A presentation of the residual compensation idea for the error-based ADRC is followed by its experimental verification using a highly uncertain laboratory-scale aerodynamic plant.

Keywords: Control of Uncertain Systems, Large-Scale System of Systems, Intelligent Power Grid
Abstract: In this study, we present a method for online estimation of the mean performance output in microgrid systems subject to high-dimensional and dynamic uncertainties. We integrate an efficient Multivariate Probabilistic Collocation Method (MPCM) based sampling strategy with a Copula-based conditional probability distribution. This integrated method enables online evaluation of system outputs with high estimation accuracy and efficiency. The online evaluation algorithm is developed, and its theoretical analysis is provided. Real Time Digital Simulator (RTDS) experiments validate the method, demonstrating its feasibility for practical applications.

Keywords: Control of Uncertain Systems, Quality and Reliability Engineering, Technology Assessment
Abstract: Data-dependence analyses (DDA) and data-flow analyses (DFA) are essential software engineering processes automatically carried out in compilers, debuggers, and, in general, in most program analyses for termination, parallelization, optimization, etc. The question "For which classes of programs DDA and DFA are decidable?" remains unanswered, even though it is of fundamental importance because it can pose limits to the algorithms that try to solve the DDA/DFA problem. It is a well-known result that DDA/DFA is undecidable for non-terminating programs. In 2000, Thomas Reps took a big step forward. He proved the undecidability of one specific DDA for programs with recursive functions and composite data structures, and he explained the important implications of that result. It was a big step in the path to prove the undecidability of other analyses. This paper uses Reps' results to prove that DDA and DFA are also undecidable for programs without composite data structures.

Keywords: Control of Uncertain Systems, System Modeling and Control, Robotic Systems
Abstract: Flexible cable-driven robotic arms (FCRAs) offer dexterous and compliant motion. Still, the inherent properties of cables, such as resilience, hysteresis, and friction, often lead to particular difficulties in modeling and control. This paper proposes a model predictive control (MPC) method that relies exclusively on input-output data, without a physical model, to improve the control accuracy of FCRAs. First, we develop an implicit model based on input-output data and integrate it into an MPC optimization framework. Second, a data selection algorithm (DSA) is introduced to filter the data that best characterize the system, thereby reducing the solution time per step to approximately 4 ms, which is an improvement of nearly 80%. Lastly, the influence of hyperparameters on tracking error is investigated through simulation. The proposed method has been validated on a real FCRA platform, including five-point positioning accuracy tests, a five-point response tracking test, and trajectory tracking for letter drawing. The results demonstrate that the average positioning accuracy is approximately 2.070 mm. Moreover, compared to the PID method with an average tracking error of 1.418°, the proposed method achieves an average tracking error of 0.541°.

Keywords: Control of Uncertain Systems, System Modeling and Control, Robotic Systems
Abstract: This paper presents a robust Model Predictive Control (MPC) framework for trajectory tracking of unmanned aerial vehicles (UAVs), enhanced by a reinforcement learning (RL) policy for warm-start initialization and a sliding mode observer (SMO) for disturbance estimation. The proposed architecture addresses multifaceted performance degradation, including residual trajectory tracking errors, reduced control responsiveness, and slow early-stage convergence, primarily caused by external disturbances, model uncertainties, and time-varying environmental conditions. An enhanced adaptive super-twisting SMO is developed to estimate lumped disturbances in real time. These estimates are integrated into the MPC prediction model to compensate for mismatches between nominal dynamics and true system behavior. To accelerate control convergence, particularly in improving early stage performance, an offline RL warm-start policy is used to generate an initial control sequence for MPC. The overall framework preserves the constraint handling and predictive capabilities of conventional MPC, while significantly improving robustness, stability, and tracking accuracy. The simulation results validate the effectiveness of the proposed method in achieving reliable and precise trajectory tracking under challenging and uncertain operating scenarios.

Keywords: Wearable Computing
Abstract: This study suggests a lightweight algorithm to estimate relative grip strength using forearm muscle deformation measured by infrared optical sensors. Unlike conventional approaches relying on machine learning or electromyography (EMG), the suggested method estimates grip strength by calculating the absolute difference between sensor values in a relaxed and contracted state, thereby enabling computation without complex signal processing. To validate the method’s generalizability across individuals, the algorithm was applied to data collected from seven participants. Each participant wore an optical sensor band (FirstVR) and a pressure-sensing glove and repeatedly gripped and released a cylindrical object while synchronized time-series data were recorded. Relative grip strength was computed using a moving average filter, and correlations with actual grip force values were analyzed. The results showed that the estimated grip strength values generally followed the trends of the measured force data, achieving a coefficient of determination (R²) above 0.5 for most participants. The findings indicate that the suggested method can provide a simple yet effective approach to grip strength estimation with minimal processing, which is potentially useful for wearable systems or real-time applications with limited computational resources.

Keywords: Wearable Computing
Abstract: Accurate localization is critical for a wide range of applications, including data navigation, augmented reality, safety management, and healthcare. However, achieving accurate positioning remains challenging, especially in GPS-degraded environments. Pedestrian Dead Reckoning (PDR), which estimates position based on inertial sensor data, offers a promising alternative but typically demands substantial effort in algorithm design and system integration. This complexity often leads to prolonged development timelines and elevated costs. To address these challenges, we present the PDR Sensor (PDRS), a plug-and-play, black-box PDR module optimized for seamless integration into wearable and IoT platforms. PDRS integrates a microprocessor and an inertial measurement unit (IMU) within a compact, low-power module. It executes real-time sensor fusion onboard, providing position updates via standard interfaces (I²C, SPI, UART). Experimental evaluations demonstrate its high robustness, achieving a step detection error of just 0.16% over 1250 steps and a traveled distance error of 1.87% under diverse conditions. The system further exhibits low latency (2.5 ms) and modest power consumption (52.35 mA). Unlike conventional studies, the PDRS leverages a holistic co-design of hardware and software to enhance performance and integration aspects, simplifying adoption without requiring deep expertise. By lowering technical barriers, PDRS enables developers and researchers to prioritize innovation in IoT and wearable technologies, accelerating prototyping, reducing development time, and advancing industry progress. Future work aims to enhance accuracy, reduce power usage and module size, and integrate AI-driven techniques to better mitigate drift.

Keywords: Wearable Computing, Haptic Systems
Abstract: This paper presents a wearable haptic device that can perceive eight directions of the forearm utilizing cloth deformation. Inducing kinesthetic perception through wearable haptic devices is expected to have applications in VR, rehabilitation, and teleoperation. The demand for haptic devices made of flexible materials that can provide intuitive feedback has rapidly increased. In this paper, we developed a novel haptic device for the forearm that consists of thin artificial muscles and existing clothing. The cloth deformation caused by the contraction of the artificial muscle allows the wearer to perceive four directions of the forearm, and the results of the perception experiment showed 98.8% accuracy. Furthermore, experimental results showed that synthesis stimuli can extend the wearer's kinesthetic perception in eight directions. The findings provide important guidelines for the design of future haptic devices.

Keywords: Wearable Computing, Human-Computer Interaction, Assistive Technology
Abstract: Locomotive syndrome is a condition in which motor function during walking deteriorates due to motor unit disorders, increasing the risk of requiring nursing care. This is a serious problem that reduces the quality of life. In this study, we focused on cybernics treatment, a method in which a wearable cyborg facilitates voluntary movement in paralyzed body parts, thereby providing sensory feedback from the paralyzed periphery to induce neural plasticity. Facilitating movements that activate the muscles along the spiral line—a myofascial chain that spirals across the body—is expected to improve coordination between the lower limbs and the trunk in patients with impaired locomotion, thereby enhancing their walking function. This study aimed to develop a method for activating the muscle groups along the spiral line responsible for coordinated hip-trunk motion using a wearable cyborg hybrid assistive limbs (HAL) through cybernics treatment, and to confirm the feasibility of this method in assisting twisting movements through coordinated hip-trunk motion in a fundamental experiment. The system consists of a trunk twisting unit and a hip flexion unit. By mechanically linking these components and implementing a control system that synchronizes trunk twisting and hip flexion based on the wearer’s intended movement, the system enables coordinated hip-trunk motion assistance. We conducted a fundamental experiment on an able-bodied adult male. Our results confirmed the presence of assistive torque during trunk twisting and the synchronization between trunk rotation and lateral bending, thereby confirming the feasibility of assisting hip flexion and trunk twisting in accordance with the wearer’s movement intention.

Keywords: Wearable Computing, Human-Computer Interaction, Medical Informatics
Abstract: Since the invention of the electrocardiogram (ECG), significant advancements have been made in its acquisition methods. However, achieving both comfort during measurement process and standardization simultaneously remains a challenge. In this study, we present a novel signal routing method, by dividing the pathway into the on-body part and environmental part and having the connection of these two parts enabled by physical contact, we are able to move the rigid hardware away from the human body. Based on this signal routing method, we design and implement an ECG signal acquisition system capable of measuring standard leads ECG. The Pearson correlation coefficient between the ECG signals acquired by our system with dry electrodes and the standard system with silver/silver-chloride (Ag/AgCl) electrodes reaches up to 0.98 in limb leads and up to 0.94 in chest leads, which demonstrates that our system has the potential to acquire the standard leads ECG signal.

Keywords: Wearable Computing, User Interface Design, Human-Computer Interaction
Abstract: Typing on a smartwatch presents significant challenges due to several factors, including the small size of these devices, the lack of tactile feedback, and the fat finger problem. This study presents an empirical analysis of two distinct text-entry methods for a one-line QWERTY keyboard: a traditional touch-based interaction and an input technique that leverages the rotary-crown, a feature commonly found on a substantial number of smartwatch devices. Interaction via the rotating crown enables text entry even when touch input is impractical—such as when wearing gloves, in wet conditions, or in cases of reduced precision like the 'fat finger' problem. By rotating the rotary-crown, the user can navigate to and select the desired character, which is then input by pressing the corresponding key to confirm the selection. The method was evaluated in comparison to the touch-interaction: the results demonstrated that participants achieved higher accuracy with the rotary-crown input, with a Total Error Rate of 9%; conversely, participants exhibited faster performance with the touch-interaction, achieving a typing speed of 9.3 words per minute.

Keywords: Ethics of AI and Pervasive Systems, Information Systems for Design, Networking and Decision-Making
Abstract: Explainable Artificial Intelligence (XAI) plays a crucial role in high-stakes decision-making by ensuring that machine learning models provide clear and trustworthy explanations. However, many existing interpretability methods, such as SHAP and Partial Dependence Plot (PDP), struggle to differentiate between correlation and causality. To overcome this challenge, we introduce a causality-aware surrogate modeling framework that improves the global interpretability of complex models. Our approach combines Probability of Sufficiency (PS), Probability of Necessity (PN), and Probability of Causality (PoC) with decision tree-based rule extraction to identify rules and features that have a direct causal impact on the target outcome. Experiments on the German Credit dataset reveal that certain rules exhibit strong sufficiency while showing weak necessity, highlighting key causal factors in loan approval decisions. Among these, savings and duration stand out as critical features for counterfactual reasoning. By ensuring that extracted rules capture true causal relationships rather than misleading correlations, our method enhances model transparency, trustworthiness, and counterfactual fundamentals.

Keywords: Human-centered Learning, Human-Machine Interaction, Networking and Decision-Making
Abstract: The interaction between humans and AI in safety-critical systems presents a unique set of challenges that remain partially addressed by existing frameworks. These challenges stem from the complex interplay of requirements for transparency, trust, and explainability, coupled with the necessity for robust and safe decision-making. A framework that holistically integrates human and AI capabilities while addressing these concerns is notably required, bridging the critical gaps in designing, deploying, and maintaining safe and effective systems. This paper proposes a holistic conceptual framework for critical infrastructures by adopting an interdisciplinary approach. It integrates traditionally distinct fields such as mathematics, decision theory, computer science, philosophy, psychology, and cognitive engineering and draws on specialized engineering domains, particularly energy, mobility, and aeronautics. Its flexibility is further demonstrated through a case study on power grid management.

Keywords: Human-centered Learning, Human-Computer Interaction, Human-Collaborative Robotics
Abstract: Chess is a highly popular board game that is still practiced by individuals of different ages, despite the rise of digital games. Aiming to update chess education without compromising its classical charm, this paper introduces Roboknight, a low-cost, modular robotic platform designed to promote interactive and fun chess learning among students of all proficiency levels. The platform integrates a 4-degree-of-freedom (4-DOF) robotic arm and a dual-power control system to provide robust operation. It employs computer vision advances, including YOLO-based detection and FEN encoding, in real-time move detection and verification. A central logic module oversees the game flow, applying predefined rules and managing interactions between the vision module, control, and the chess engine. A web-based dashboard further adds to the potential for remote monitoring and performance analysis, thus enabling large-scale deployment in educational settings. Although the existing prototype delivers a functional and enjoyable learning experience, future upgrades will focus on multi-device synchronization, conversational tutoring using chatbot integration, and improved energy efficiency through smart power management.

Keywords: Telepresence, User Interface Design, Human-Computer Interaction
Abstract: Society and organizations are undergoing a profound digital transformation, driven by emerging technological advances and new paradigms that address the growing demands of productive, commercial, service-related, and social activities. This evolution has led to the emergence of so-called Next-Generation Collaborative Environments, where providing awareness remains a critical factor for usability, efficiency and a satisfactory user experience. In this work, we analyse some significant awareness frameworks from the literature and propose a new conceptualisation tailored to these emergent environments, aimed at groupware designers. Our proposal is novel in both its scope and its non-linear approach to the awareness areas and dimensions it encompasses. This conceptual schema has been preliminarily validated through its application to the modelling of a diverse set of case studies.

Keywords: Design Methods, Resilience Engineering, Human-Collaborative Robotics
Abstract: As the demand for disassembling end-of-life products grows, limitations in traditional disassembly line design, low efficiency, and high resource consumption become increasingly evident. Particularly in large-scale disassembly tasks, where the cost of conventional remanufacturing rises and the technologies fail to meet high-efficiency requirements. The integration of robots into disassembly lines is a promising solution to alleviate these issues. This work presents a multi-product hybrid disassembly line balancing problem that considers machine wear rates and establishes a mixed-integer programming model guided by profit maximization to address it. An improved fireworks algorithm is used in the proposed approach. The developed solution is compared with genetic and ant colony algorithms. Evaluation results and analysis demonstrated the competitive efficiency and stability of our approach.

Keywords: Multimedia Systems
Abstract: In multimedia applications such as films and video games, spatial audio techniques are widely employed to enhance user experiences by simulating 3D sound: transforming mono audio into binaural formats. However, this process is often complex and labor-intensive for sound designers, requiring precise synchronization of audio with the spatial positions of visual components. To address these challenges, we propose a visual-based spatial audio generation system - an automated system that integrates face detection YOLOv8 for object detection, monocular depth estimation, and spatial audio techniques. Notably, the system operates without requiring additional binaural dataset training. The proposed system is evaluated against existing Spatial Audio generation system using objective metrics. Experimental results demonstrate that our method significantly improves spatial consistency between audio and video, enhances speech quality, and performs robustly in multi-speaker scenarios. By streamlining the audio-visual alignment process, the proposed system enables sound engineers to achieve high-quality results efficiently, making it a valuable tool for professionals in multimedia production.

Keywords: Manufacturing Automation and Systems, Cyber-physical systems
Abstract: The disassembly and assembly line balancing problem (DALP) is a critical task in industrial production, involving the efficient organization of disassembly and assembly tasks to improve the productivity and flexibility of production lines. In practical applications, task allocation, robot movement, and workstation layout optimization are key factors affecting production efficiency. This study proposes an improved parallel advantage actor-critic algorithm to address DALP with space constraints due to robot movement. Considering the limitations of workstation space, this approach optimizes the robot’s movement paths between workstations, reducing the cost of opening workstations, and optimizing task allocation strategies. To enhance the convergence speed and stability of the conventional Parallel A2C algorithm, action space optimization and a greedy strategy are incorporated into the algorithm. Experimental results demonstrate that the improved parallel advantage actor-critic outperforms the A2C and AC algorithms in terms of efficiency and performance, particularly in handling disassembly tasks with space constraints, significantly improving the operational efficiency and economic benefits of the production line.

Keywords: Manufacturing Automation and Systems, Cyber-physical systems
Abstract: Product disassembly is significant for recycling scrapped products and reducing environmental pollution and resource waste. The recovery, reuse, and recycling of industrial products is crucial in modern industry. Manual disassembly efficiency significantly impacts the disassembly line’s overall effectiveness, especially workers’ skill level and learning efficiency. This paper proposes a multi-period personnel scheduling problem that considers worker learning effects. A mixed integer programming model for the disassembly balance problem was established to maximize disassembly profit. This problem is solved using a new reinforcement learning algorithm, the importanceweighted actor-learner architecture(IMPALA). The correctness and effectiveness of the proposed algorithm are verified through comparative experiments with the famous IBM optimizer CPLEX and some popular peer algorithms.

Keywords: Robotic Systems
Abstract: In recent years, the widespread adoption of unmanned aerial vehicles (UAVs) has increased the demand for advanced obstacle avoidance capabilities. Traditional path planning algorithms often exhibit low efficiency and poor adaptability, making them unsuitable for dynamic environments. To address these limitations, researchers have explored reinforcement learning (RL)-based approaches, with Proximal Policy Optimization (PPO) widely used for its stable policy updates and improved sample efficiency. However, PPO suffers from slow convergence, which limits its real-time applicability. To overcome this issue, this paper integrates the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) into PPO and proposes PPO-CMA, an enhanced algorithm for end-to-end UAV path planning. The PPO-CMA network, leveraging depth images and UAV pose feedback, generates continuous control actions, combining CMA-ES adaptive search with PPO policy optimization to improve convergence speed and learning efficiency. The proposed method is evaluated in the VisFly simulation environment, demonstrating significantly faster convergence compared to traditional PPO while ensuring accurate target-reaching and obstacle avoidance. The real-world experimental results validate the effectiveness, robustness, and practical applicability of PPO-CMA for UAV navigation and motion planning in real-world scenarios.

Keywords: Autonomous Vehicle, System Modeling and Control, Mechatronics
Abstract: This paper addresses the challenge of real-time control of multirotors subjected to partially known and unmodeled dynamics. A physics-informed Koopman operator framework is proposed, where the known physical dynamics and unknown residual effects are separated using a Strang splitting approach. The continuous-time Koopman operator is trained on physics-derived trajectories, while the discrete-time Koopman operator is learned from real-world trajectory data, enabling a data-efficient and globally linearizable model of the multirotor dynamics. The learned linear model is subsequently used to design a discrete-time Sequential Action Control (SAC) policy for real-time trajectory tracking. Experimental validation on a quadrotor platform tracking a lemniscate trajectory demonstrates that the proposed PI-EDMD-based SAC controller achieves superior tracking accuracy and up to 67% lower control energy consumption compared to baseline nonlinear SAC and LQR controllers. These results highlight the effectiveness of the proposed framework in enhancing both trajectory fidelity and actuation efficiency for multirotors.

Keywords: Fault Monitoring and Diagnosis, Cyber-physical systems, Quality and Reliability Engineering
Abstract: Multirotor unmanned aerial vehicles (UAVs) frequently experience degraded control authority due to partial actuator faults, compromising mission reliability and safety. Purely data-driven fault diagnostic methods, although effective, typically demand extensive labelled datasets and lack direct interpretability. Physics-informed neural networks (PINNs), which incorporate physical laws directly into their learning process, offer a promising alternative by enabling data-efficient training and interpretable results. This study proposes a PINN for actuator fault diagnosis in quadrotor UAVs by embedding discrete Newton–Euler residuals within its loss function, ensuring predictions remain consistent with rigid-body dynamics. A quadrotor UAV is modelled in a high-fidelity simulation environment, and flight data from these simulations are collected for analysis. A sensitivity study is subsequently conducted, testing the PINN against varied fault magnitudes, fault intervals, and different training dataset sizes (100%, 50%, and 30%).The proposed PINN outperforms a similarly sized multilayer perceptron (MLP), reducing fault detection delay by approximately 20%, consistently achieving macro F1 scores above 0.90, and improving prediction accuracy R^2 and RMSE, even with limited labelled data.

Keywords: Cyber-physical systems, System Modeling and Control, Mechatronics
Abstract: This study presents a data-driven and physics-informed learning framework for predicting the kinematics of an elastic-incorporated flapping-wing device. An experimental setup was developed where a flapping-wing device was designed and built to collect time-series kinematic data. A physics-informed neural network (PINN) was trained on the experimental dataset, achieving an RMSE of 0.3140 on the holdout set. In comparison, a multi-layer perceptron (MLP) of equivalent architecture yielded a higher RMSE of 0.426 and exhibited memorization. The incorporation of physics-based constraints into the PINN enhanced predictive accuracy and improved generalization, particularly in capturing the oscillatory behavior of the flapping system. These results demonstrate the effectiveness of PINNs in predicting the kinematics of an elastic-integrated flapping mechanism. Future directions include transforming the kinematic model to a full dynamic formulation, which would allow for precise control of flapping-wing prototypes with integrated elasticity.

Keywords: System Modeling and Control
Abstract: The growing adoption of drones for goods delivery has emerged as a potentially viable solution. By operating through aerial routes, drones significantly reduce delivery times and expand operational reach. However, covering large areas requires prolonged flights, leading to high battery consumption and an increased risk of collisions, particularly in densely populated regions. This study presents a Stochastic Petri Net model to evaluate drone performance, focusing on metrics such as utilization, delivery rate, mean mission time, and drop probability. Additionally, energy consumption and carbon footprint metrics were investigated to assess the environmental impact of drone operations. The model incorporates factors such as strategic recharging points and collision probability, providing insights into drone performance under high-demand scenarios.

Keywords: System Modeling and Control, Modeling of Autonomous Systems
Abstract: This paper presents the modeling and control of a tilt-birotor UAV, an emerging aerial platform that combines efficiency and agility. Departing from conventional multirotor configurations, the birotor drone utilizes two tiltable rotors to generate lift and directional control. This not only reduces the drone’s weight but also allows better maneuverability, at the expense of higher nonlinearity and couplings in the drone model. To improve tracking performance and robustness an attitude and altitude stabilization strategy, based on a nonlinear backstepping controller, is proposed in this work. To generate the low-level control inputs, the proposed controller is combined with both linear and nonlinear allocation strategies. Simulation results, conducted on the full non-linear model of a V-shaped birotor design, confirm the viability of the proposed control strategy.

Keywords: Robotic Systems, Modeling of Autonomous Systems, System Modeling and Control
Abstract: Dynamic environments present a significant challenge for mobile robot localization, as changes in object configuration and human movement can lead to incorrect position estimates. Conventional methods struggle with these variations, often resulting in mismatches and localization failures. To address this issue, we propose a Visual SLAM-based localization method that integrates semantic information using an RGB-D camera. Our approach utilizes instance segmentation to enhance localization robustness in 3D environments. During map creation, feature points extracted from images in a static environment are assigned label IDs corresponding to object regions, embedding semantic information into the 3D map. During localization, feature matching is constrained by label consistency, ensuring that only points with identical labels are matched. Additionally, feature points associated with high-mobility objects, such as chairs or people, are excluded to prevent errors caused by environmental changes. Furthermore, during pose optimization, the information matrix is weighted according to semantic labels, giving higher importance to static objects and reducing the influence of dynamic ones. We validate the proposed method through real-world experiments in cluttered indoor environments with varying furniture arrangements and human presence. The results demonstrate that our approach significantly reduces mismatches and improves localization accuracy compared to conventional ORB-SLAM2. Unlike existing methods that update maps online, our approach maintains a fixed pre-constructed map, reducing inconsistencies over time. This study enhances localization robustness in changing environments by incorporating semantic constraints into Visual SLAM.

Keywords: Conflict Resolution
Abstract: The inverse analysis of the graph model for conflict resolution (GMCR) determines the preferences required to ensure that a desired resolution is an equilibrium. However, existing studies have yet to examine the transition mechanism from the current state to the desired resolution. To address this issue, this study introduces an integrated minimum cost conflict mediation path (I-MCCMP) model within the GMCR framework. This model not only identifies the necessary preferences to establish the desired resolution as an equilibrium but also determines the optimal conflict mediation path to achieve it. To demonstrate its applicability, the proposed model is applied to the real-world Lake Gisborne Conflict.

Keywords: Conflict Resolution, Decision Support Systems
Abstract: This study proposes a multi-agent-based simulation and optimization framework to enhance phased evacuation strategies by integrating fire dynamics, evacuees’ characteristics, and evacuation processes. Unlike conventional models, this framework probabilistically assesses the integrated effects of heat, asphyxiant gases, and irritant gases on different evacuees’ incapacitation. The Non-dominated Sorting Genetic Algorithm III (NSGA-III), coupled with a trained neural network, is employed to find optimal phased evacuation strategies considering the Total Evacuation Time (TET), congestion, and fire impact. To assess its effectiveness, the framework is applied to a fire scenario in an educational building, comparing simultaneous and phased evacuation strategies. Results demonstrate that the selected phased evacuation strategy significantly enhances evacuation efficiency, reducing TET, congestion, and fire impact by 14.8%, 33.3%, and 13.1%, respectively. These findings underscore the framework’s potential for improving fire evacuation planning, providing a simulation-based approach to optimizing evacuation strategies and enhancing safety in fire.

Keywords: Conflict Resolution
Abstract: In this paper, we establish a mathematical representation of the Longest-Valid-Chain rule in the blockchain system based on the social choice correspondence. The "longest"-ness is defined by the resources used to create a new block, and the "valid"-ness is defined by the compatibility between protocols. Instead of a single individual or an entity making the decision, the Longest-Valid-Chain rule is an important method for decision-making in the blockchain-based decentralized organization. The Longest-Valid-Chain rule can transform a list of individual preferences into a social choice when the protocol fork occurs in the blockchain system. According to the analysis, we prove that the Longest-Valid-Chain rule is not Paretian, but is anonymous, neutral, and strongly monotonic.

Keywords: AI and Applications, Deep Learning, Machine Learning
Abstract: Test time adaptation (TTA) equips deep learning models to handle unseen test data that deviates from the training distribution, even when source data is inaccessible. While traditional TTA methods often rely on entropy as a confidence metric, its effectiveness can be limited, particularly in biased scenarios. Extending existing approaches like the Pseudo Label Probability Difference (PLPD), we introduce ETAGE, a refined TTA method that integrates entropy minimization with gradient norms and PLPD, to enhance sample selection and adaptation. Our method prioritizes samples that are less likely to cause instability by combining high entropy with high gradient norms out of adaptation, thus avoiding the overfitting to noise often observed in previous methods. Extensive experiments on CIFAR-10-C and CIFAR-100-C datasets demonstrate that our approach outperforms existing TTA techniques, particularly in challenging and biased scenarios, leading to more robust and consistent model performance across diverse test scenarios. The codebase for ETAGE is available on https: //github.com/afsharshamsi/ETAGE.

Keywords: AI and Applications, Computational Intelligence, Neural Networks and their Applications
Abstract: Depression is a serious mental health illness that significantly affects an individual’s well-being and quality of life, making early detection crucial for adequate care and treatment. Detecting depression is often difficult, as it is based primarily on subjective evaluations during clinical interviews. Hence, the early diagnosis of depression, thanks to the content of social networks, has become a prominent research area. The extensive and diverse nature of user-generated information poses a significant challenge, limiting the accurate extraction of relevant temporal information and the effective fusion of data across multiple modalities. This paper introduces MMFformer, a multimodal depression detection network designed to retrieve depressive spatio-temporal high-level patterns from multimodal social media information. The transformer network with residual connections captures spatial features from videos, and a transformer encoder is exploited to design important temporal dynamics in audio. Moreover, the fusion architecture fused the extracted features through late and intermediate fusion strategies to find out the most relevant intermodal correlations among them. Finally, the proposed network is assessed on two large-scale depression detection datasets, and the results clearly reveal that it surpasses existing state-of-the-art approaches, improving the F1-Score by 13.92% for D-Vlog dataset and 7.74% for LMVD dataset. The code is made available publicly at https://github.com/rezwanh001/Large-Scale-Multimodal-Depression-Detection.

Keywords: AI and Applications, Deep Learning, Multimedia Computation
Abstract: 目深度估计 （MDE） 对于自动驾驶至关重要，无需昂贵的 LiDAR 即可实现 3D 感知，但现有的针对 MDE 的 2D 白盒对抗攻击缺乏对视点和现实世界变化的鲁棒性。我们提出了 DepthHack，这是 MDE 的第一个 3D 黑盒对抗性攻击框架，它使用概率采样和基于分数的优化来制作强大的 3D 对抗性纹理。DepthHack 可确保在各种天气条件和视点下的稳健性，在 Monodepth2 （Carla） 上实现 8.88 m 的平均深度估计误差，比 HardBeat 高出 2.1%，仅 60k 个查询。这项工作揭示了 MDE 漏洞，增强了自治系统的安全性。在 4 个 MDE 模型和 2 个数据集上的实验验证了其卓越的性能、效率和跨数据集泛化能力。

Keywords: AI and Applications, Deep Learning, Representation Learning
Abstract: Knowledge tracing models have long grappled with the dual challenges of data sparsity and the limited ability to capture group learning patterns. Current contrastive learning paradigms in knowledge tracing (e.g., CL4KT framework) primarily employ sequence augmentation strategies to alleviate data scarcity constraints. However, such approaches frequently compromise semantic coherence during the augmentation process while failing to account for inherent similarity patterns within learner cohorts. To overcome these limitations, this paper introduces the GSCL-KT model (Group Similarity Contrastive Learning for Knowledge Tracing), which, for the first time, incorporates a group-similarity-aware contrastive learning mechanism into the knowledge tracing domain. Unlike traditional approaches that rely on manual data augmentation, GSCL-KT dynamically identifies positive and negative sample pairs from educationally homogeneous groups, enabling the discovery of group-level cognitive patterns while maintaining semantic coherence. The proposed model incorporates several advanced optimization strategies, including the Talking-Heads attention mechanism for fine-grained interaction modeling, the ContraNorm method for feature distribution regularization, and a correlation network enhanced by label dependencies. Experimental results on four real-world educational datasets demonstrate that GSCL-KT consistently outperforms existing baseline models, achieving the highest AUC and competitive performance across metrics.

Keywords: AI and Applications, Evolutionary Computation, Machine Learning
Abstract: With the stagnation of Moore's Law scaling, efficient hardware architectures employing compute-in-memory paradigms have become increasingly crucial to sustain AI innovations. This motivates the development of high-throughput architectures with balanced energy-latency profiles through machine learning algorithms. However, for human-in-the-loop optimization methods, human labor is involved in most of the iterations, whereas for automated methods, either expert domain knowledge is required in the design or the search space is relatively small. To address these challenges, we propose ArchERL, an evolutionary reinforcement learning (ERL) framework that represents the first application of ERL to general hardware architecture design. Specifically, ArchERL tightly couples population-based evolutionary algorithm for global exploration with an actor-critic reinforcement learning module for prior-free policy refinement, and ArchERL employs periodic weight synchronization and gradient feedback between the two modules to achieve efficient collaborative search and rapid convergence. To evaluate the proposed method, extensive experiments are conducted in multiple simulated hardware environments, including the DRAM controller and DNN mapping. The results demonstrate that ArchERL achieves state-of-the-art performance and outperforms widely used baselines in both efficiency and effectiveness.

Keywords: AI and Applications, Expert and Knowledge-Based Systems, Deep Learning
Abstract: Agriculture is a multidisciplinary domain that spans a wide range of sectors, including horticulture, animal husbandry, and water and land resource management. The intricate complexity of this field arises from the interconnectivity of its diverse sub-sectors, requiring advanced knowledge-specific expertise to address interdisciplinary challenges effectively. Traditional methods for integrating knowledge across sub-sectors are often limited in their ability to deliver precise and context-specific insights, particularly for complex, multi-sectors inquiries. In this paper, we present a multi-sectors framework for question-answering in agriculture. Our framework utilizes multiple Large Language Models, each of which is specialized in a sector of agriculture. By adopting the Mixture of Experts paradigm, our framework synthesizes the expertise of multiple Large Language Models to produce accurate and interdisciplinary answers to questions spanning multiple agricultural sectors. We have evaluated the performance of our approach using qualitative metrics. Our results demonstrate the effectiveness of our approach compared with a baseline Large Language Model question-answering strategy.

Keywords: Human Performance Modeling, Human-Collaborative Robotics, Shared Control
Abstract: Human and automation capabilities are the foundation of every human-autonomy interaction and interaction pattern. Therefore, machines need to understand the capacity and performance of human doing, and adapt their own behavior, accordingly. In this work, we address the concept of conjugated capabilities, i.e. capabilities that are dependent or interrelated and between which effort can be distributed. These may be used to overcome human limitations, by shifting effort from a deficient to a conjugated capability with performative resources. For example: A limited arm's reach may be compensated by tilting the torso forward. We analyze the interrelation between elementary capabilities within the IMBA standard to uncover potential conjugation, and show evidence in data of post-rehabilitation patients. From the conjugated capabilities, within the example application of stationary manufacturing, we create a network of interrelations. With this graph, a manifold of potential uses is enabled. We showcase the graph's usage in optimizing IMBA test design to accelerate data recordings, and discuss implications of conjugated capabilities on task allocation between the human and an autonomy.

Keywords: Human-Machine Interaction, Human-Collaborative Robotics, Human Factors
Abstract: In this paper, we investigate the impact of high-level semantics (evaluation of the environment) on Human-Robot Team (HRT) and Human-Robot Interaction (HRI) in the context of mobile robot deployments. Although semantics has been widely researched in AI, how high-level semantics can benefit the HRT paradigm is underexplored, often fuzzy, and intractable. We applied a semantics-based framework that could reveal different indicators of the environment (i.e. how much semantic information exists) in a mock-up disaster response mission. In such missions, semantics are crucial as the HRT should handle complex situations and respond quickly with correct decisions, where humans might have a high workload. Especially when human operators need to shift their attention between robots and other tasks, they will struggle to build Situational Awareness (SA) quickly. The experiment suggests that the presented semantics: 1) alleviate the perceived human operator's workload; 2) increase the operator's trust in the SA; and 3) help to reduce the reaction time in switching the Level of Autonomy (LoA) when needed. Additionally, we find that participants with higher trust in the system are encouraged by high-level semantics to use teleoperation mode more.

Keywords: Haptic Systems, Shared Control, Human-Machine Cooperation and Systems
Abstract: Haptic shared control systems that support drivers by means of added torques on the steering wheel are often tuned heuristically. To allow for more systematic design, this paper focuses on the Four Design Choices Architecture (FDCA) and systematically analyzes its tuning with an offline simulation model for the driver's control behavior and neuromuscular system. These analyses indicated that within the FDCA architecture the Level of Haptic Support (LoHS), which is a feedforward channel supporting negotiation of upcoming curves, is a main contributor to joint system performance. In a driving simulator experiment, the adaptation to and acceptance of different LoHS levels was investigated. Driver acceptance was found to increase with increasing LoHS values up to 1. Objective metrics, including torque conflict (70% reduction), steering effort (81% reduction), steering wheel reversal rate, and lateral deviation all improved, indicating that with the FDCA a high LoHS is both acceptable and, in fact, preferred.

Keywords: Human Performance Modeling, Human-Machine Cooperation and Systems, Shared Control
Abstract: InceptionTime neural network models were trained to detect distractions in manual control tasks with pursuit and preview displays. Training and test data were collected in an experiment where ten participants were deliberately distracted from the primary control task using the Surrogate Reference Task (SuRT). Overall, distractions are easier to detect in pursuit tasks, with test accuracies of around 80% and 60% for pursuit and preview data, respectively. With preview, human controllers see the future target trajectory, which enables them to mitigate distraction effects. Unexpectedly, data with longer distractions from 'hard' SuRT tasks are more difficult to classify than 'easy' distractions; an effect attributed to differences in human behavior between the training and test data collection conditions. These results show clear opportunities for neural network models to detect distractions, in real-time, for increasing safety of human-operated vehicles.

Keywords: Shared Control, Human Performance Modeling, Human-Machine Interface
Abstract: Haptic Shared Control (HSC) systems offer a means to support human drivers in the transition to fully-automated driving. Matching HSC systems settings with drivers' time-varying neuromuscular system (NMS) dynamics requires real-time HSC adaptations. This paper presents an experimental validation of a previously proposed method for predicting drivers' time-varying neuromuscular admittance using an 'average' grip force scheduled linear parameter varying (LPV) model. The quality of LPV model predictions is compared to that of Recursive Least Squares (RLS) fits of an admittance model on the same data. Ten participants performed steering wheel manipulation tasks with steering wheel perturbations that needed to be kept within a certain displacement boundary by adapting their grip force. Time-invariant (TI) and time-varying (TV) boundary levels were used to, respectively, construct and validate the LPV model. Results show that the average relation between admittance and grip force that underlies the current LPV method varies too much between TV and TI tasks, hampering accurate admittance predictions. Compared to the quality-of-fit of 80-90% obtained with RLS on the TV data, the LPV model's predictions are insufficiently accurate and do not exceed 55% on average. An approach that enables individual instead of average LPV models to be constructed directly from TV experiment data needs to be pursued for HSC implementations.

Keywords: Systems Safety and Security,
Abstract: The traditional intelligent models developed based on the source device fault diagnosis (SDFD) framework have achieved high accuracy, but may lead to unreliable diagnostic results for unseen individual diagnostic scenarios, as they ignore potential individual differences (such as assembly changes, noise interference). This paper proposes a refined trigonometric activation representation transformer (RTART), which is a signal processing informed neural network (SPINN) tailored for cross individual fault diagnosis (CIFD). By integrating the theory of Sparse Short Time Fourier Transform (DSTFT) and Kolmogorov Arnold Representation Theorem (KART) into structural design of the Transformer, RTART enhances the extraction of periodic vibration features and frequency modulation diversity. Specifically, the original vibration signals are first segmented into regional patches based on one-dimensional convolutional patch (1-DCP) tokenizer, and then input into a multi-scale region pruning (MSRP) module for global feature refinement, which aims to refine decision features for reducing the risk of overfitting due to individual differences. A trigonometric activation representation (TrigAR) module is developed to enhance the reliable feature expression of periodic vibration signals and improve the model generalization. The proposed method is validated on a well-known public dataset using the CIFD benchmark. Compared with the most advanced methods, RTART has better generalization ability and achieves a cross individual accuracy rate of over 95%.

Keywords: Smart Buildings, Smart Cities and Infrastructures, Digital Twin, Intelligent Transportation Systems
Abstract: The rapid expansion of outdoor traffic camera systems requires efficient methods to accurately estimate the geolocation of objects within their scenes. We present an innovative and scalable framework that completely automates this process by combining easy-to-build 3D world modeling with real-world traffic camera imagery. First, using the Cesium plugin for Unreal Engine, we create detailed and scalable 3D representations of urban environments, leveraging publicly available, highly accurate 3D data. This results in the creation of globally curated 3D content, including terrain, imagery, and photogrammetry. The real-world traffic camera imagery is then matched within our model using state-of-the-art feature matching techniques. By estimating the homography between synthetic images from the 3D model and the real images from traffic cameras, we accurately determine the geolocation of observed objects within the scene.

This approach not only enhances geolocation accuracy but also enables seamless scalability across diverse urban settings and camera deployments worldwide. Our method significantly reduces the manual effort required for traffic camera calibration, thus streamlining the deployment of intelligent transportation systems at scale. We demonstrate the high performance of our approach by experimenting in an urban trial site with multiple smart city cameras and publicly available cameras around the world. Additionally, we highlight the adaptability of our framework for a wide range of computer vision-based traffic analytics applications, including its potential for drone-based localization.

Keywords: Smart Buildings, Smart Cities and Infrastructures, Distributed Intelligent Systems, Cyber-physical systems
Abstract: With the widespread deployment of intelligent building management systems (BMS) in smart buildings, security threats posed by false data injection (FDI) attacks have become increasingly prominent. To address the proven vulnerability of Multi-Agent Deep Reinforcement Learning (MADRL)-driven BMS to FDI intrusions, this study proposes a QMIX framework enhanced by adversarial training. By incorporating adversarial samples crafted via the Fast Gradient Sign Method (FGSM) during training and introducing Noisy Network layers, the framework's robustness against black-box attacks is substantially improved. Experimental results demonstrate that compared to the baseline model without defense mechanisms, the proposed framework achieves significantly smaller performance degradation under black-box attacks across varying intensity levels, particularly under high-intensity attack scenarios, while maintaining stable control performance. Furthermore, we conduct white-box attack experiments to systematically assess model robustness, along with ablation studies quantifying the individual contributions of adversarial training and Noisy Networks to defense efficacy.

Keywords: Smart Buildings, Smart Cities and Infrastructures, Distributed Intelligent Systems, Robotic Systems
Abstract: Current indoor positioning technologies face significant limitations in dynamic industrial environments, where metallic structures and signal interference degrade accuracy. While Bluetooth Low Energy (BLE) RSSI-based systems offer a cost-effective and energy-efficient solution, their performance is often hampered by path loss variability in dynamic environments. Usually, to improve BLE RSSI-based positioning accuracy, the number of infrastructure nodes (IN) may be increased. However, increasing the number of static infrastructure nodes (SIN) leads to increased complexity, cost, and network load.

To address these challenges, we propose a novel collaborative indoor positioning method integrating autonomous mobile robots (AMRs) as mobile infrastructure nodes (MINs). Unlike traditional static deployments, our system leverages AMRs equipped with high-accuracy onboard positioning and BLE receivers to dynamically collect and correlate RSSI data while navigating the environment. This approach enables real-time adaptive localization by fusing MIN data with fixed SIN measurements using a machine learning-based fusion model (combining Neural Networks and Kernel Density Estimation).

Keywords: Smart Buildings, Smart Cities and Infrastructures, System Architecture
Abstract: Collaborative indoor positioning requires maps with explicitly defined spatial connectivity. This paper presents a framework for constructing indoor map layouts using spatial entities and reachability structures. We introduce the Floor-Group-Area (FGA) Matrix Problem to visualize and optimize layout connectivity through column reordering. The framework exports 1bpp OGMs and spatial metadata in JSON. In user evaluations, trained participants completed full layouts of a four-story museum in 20 minutes, with all computationally intensive algorithms running in few seconds on a standard PC.

Keywords: Active BMIs, Other Neurotechnology and Brain-Related Topics, BMI Emerging Applications
Abstract: Brain-machine interfaces (BMIs) have significantly advanced neuro-rehabilitation by enhancing motor control. However, accurately decoding continuous grasp force remains a challenge, limiting the effectiveness of BMI applications for fine motor tasks. Current models tend to prioritise algorithmic complexity rather than incorporating neurophysiological insights into force control, which is essential for developing effective neural engineering solutions. To address this, we propose EEGForceMap, an EEG-based methodology that isolates signals from the premotor-parietal region and extracts task-specific components. We construct three distinct time-frequency feature sets, which are validated by comparing them with prior studies, and use them for force prediction with linear, non-linear, and deep learning-based regressors. The performance of these regressors was evaluated on the WAY-EEG-GAL dataset that includes 12 subjects. Our results show that integrating EEGForceMap approach with regressor models yields a 61.7% improvement in subject-specific conditions (R² = 0.815) and a 55.7% improvement in subject-independent conditions (R² = 0.785) over the state-of-the-art kinematic decoder models. Furthermore, an ablation study confirms that each preprocessing step significantly enhances decoding accuracy. This work contributes to the advancement of responsive BMIs for stroke rehabilitation and assistive robotics by improving EEGbased decoding of dynamic grasp force.

Keywords: BMI Emerging Applications, Other Neurotechnology and Brain-Related Topics
Abstract: Electroencephalography (EEG) emerges as a highly accessible, affordable, and pervasive method for early detection and continuous monitoring of Alzheimer's Disease (AD). Existing studies usually employ all the avialiable EEG channels, despite the importance of each EEG channel in AD detection is largely undeveloped yet. This study presents a systematic evaluation of EEG channel selection through exhaustive analysis of all possible channel combinations (1–19 channels) to determine optimal configurations for AD detection. We performed 570 experiments (190 combinations × 3 seeds) using iterative forward selection. Our results demonstrate that: (1) temporal channels (particularly T5) show the highest single-channel predictive value (64.23% F1 for T5 alone); (2) an optimal 9-channel subset {T5, T6, F8, F4, Fp1, O1, P3, P4, F3} achieves peak performance (74.8% F1), outperforming full 19-channel setups (71.42%); (3) a 5-channel subset {T5, T6, F8, F4, Fp1} obtains competitive F1 of 73.8%, achieving the trade-off between less channels and higher performance. We also provide a new concept, Marginal Utility (MU), that quantifies the performance change in AD detection when adding a specific EEG channel to the model's input, providing a metric for individual electrode significance. Our data-driven, exhaustive analysis ensures unbiased identification of the most diagnostically relevant channels for Alzheimer’s disease. These findings also enable the development of efficient, clinically viable EEG systems for AD detection using 70% fewer channels without sacrificing diagnostic accuracy.

Keywords: Active BMIs, BMI Emerging Applications, Passive BMIs
Abstract: This study proposes a Brain-Computer Interface (BCI) based on Motor Imagery (MI) for ankle-foot dorsiflexion training, providing Functional Electrical Stimulation (FES) as a way of NeuroFeedback (NF), in the tibialis anterior muscle. Riemannian geometry is utilized as a feature extraction technique for a more reliable MI-based electroencephalography discrimination, also considering Passive Movement (PM) data. A complete Spinal Cord Injury (SCI) individual tested the BCI during five sessions, each one on a different day, achieving an accuracy around 0.33, greater than the chance level of 0.25, considering that this is a four-class classification system. The mean BCI latency was lower than 240 ms. Significant Relative power changes (R) were observed in the mu (8-12 Hz) band for MI of dorsiflexion of the feet as well as negative R values centered around Cz, suggesting a better MI performance after observing PM as a visual real guide. As a highlight, our BCI calibrated with more reliable MI data, greatly enhanced the mu (8-12 Hz) rhythm modulation in the operation stage more than the high-beta (18-24 Hz) band modulation. These findings are relevant for advancing NF and BCI approaches to restore lower-limb motor functions, as well as enhance neuroplasticity.

Keywords: Other Neurotechnology and Brain-Related Topics, BMI Emerging Applications, Passive BMIs
Abstract: Immersive virtual reality (VR) applications rapidly expand across domains, including training, gaming, and healthcare. More recently, multisensory immersive experiences, including olfactory and haptic stimulation, have emerged and shown great promise, especially for interventions in well-being and mental health management. Multisensory experiences, however, are very subjective (e.g., one subject may like certain smells, while others do not), and recent results have suggested that some participants may not respond positively to the treatment. As multisensory VR interventions can be costly and time-consuming for both patients and clinicians, being able to find neuromarkers that predict intervention outcomes would be invaluable. Here, we aim to take the first steps in the development of a neuromarker to predict the response to a multisensory nature immersion VR intervention. A pilot experiment was performed with twenty patients diagnosed with post-traumatic stress disorder. Potential neuromarkers are extracted from electroencephalography (EEG) signals measured from an instrumented VR headset. We show that some EEG patterns start to differ between responders and non-responders as early as the fourth session, i.e., one-third of the way into the entire intervention. This suggests that neuromarkers to predict the outcomes of a multisensory VR immersion intervention may exist. These markers could be used not only to save time and resources for clinicians and patients but also to promote precision treatment where interventions are adjusted to each patient, maximizing success rates.

Keywords: Other Neurotechnology and Brain-Related Topics, Passive BMIs, BMI Emerging Applications
Abstract: Cognitive decline in aging is a critical issue that often leads to long-term functional impairments, including dementia. Cognitive training aims to maintain or improve abilities such as working memory, attention, and motor skills. Working memory, in particular, significantly influences daily life activities and is known to gradually deteriorate with age. In this study, we propose a cognitive training paradigm that actively engages working memory within a virtual reality (VR) environment. We evaluated the effectiveness of this paradigm through a user study involving 20 older adults, specifically assessing its impact on short-term working memory. Experimental results revealed that, based on a standardized working memory test administered before and after the VR training, 60% of participants improved their performance, 20% maintained their initial performance, and 20% experienced a decline. Additionally, EEG analysis demonstrated that participants who improved, exhibited significant differences in alpha-theta band power spectral density compared to those whose performance declined; these frequency bands are known indicators of working memory performance. Our findings suggest that VR-based cognitive training has the potential to enhance short-term working memory in older adults.

Keywords: BMI Emerging Applications, Passive BMIs, Active BMIs
Abstract: This study presents a cross-subject electroencephalography (EEG) decoding framework for imagined finger tapping of the affected hand in chronic stroke patients. EEG from 11 participants was recorded using 27 channels during motor imagery of individual fingers. To capture functional connectivity, the weighted phase lag index was computed across six frequency bands. Due to high inter-subject variability, Kruskal–Wallis and Mann–Whitney U tests were used for two-stage feature selection. Selected features were classified using Random Forest, Support Vector Machine, and AdaBoost, with leave-one-subject-out validation. AdaBoost achieved the best performance (39.6% accuracy, 38.7% F1-score), exceeding chance level. Discriminative features were prominent in alpha and delta bands, reflecting activity in prefrontal and parietal-occipital areas. These results indicate that stroke patients retain task-relevant neural patterns, and the proposed method supports interpretable, scalable classification for motor imagery-based rehabilitation.

Keywords: Manufacturing Automation and Systems
Abstract: This paper presents a lightweight and flexible artificial neural network (ANN) model for predicting the transfer characteristics of amorphous Indium Gallium Zinc Oxide (a-IGZO) thin-film transistors (TFTs). Traditional TCAD simulations, while accurate, are computationally intensive and inflexible to rapid design variations. Prior efforts using variational autoencoders (VAEs) showed promise but were limited by rigid input formats that required retraining for any changes in curve dimension or voltage range. To overcome this, we propose an ANN architecture that treats gate voltage as a dynamic input, enabling continuous and accurate predictions across a wide voltage spectrum without the need for retraining. Experimental evaluation through 5-fold cross-validation confirms the ANN’s competitive performance, achieving an average R² score of 0.9898, outperforming VAE models in flexibility and robustness. This approach offers a fast, accurate, and hardware-efficient alternative for a-IGZO TFT modeling and optimization, facilitating the monolithic 3D (M3D) integration and the next generation of display.

Keywords: Intelligent Transportation Systems, Consumer and Industrial Applications, Quality and Reliability Engineering
Abstract: As Level 2–3 autonomous vehicles become more widespread, the reliable validation of Driver Monitoring Systems (DMS) has become essential to assess driver attention and readiness for control takeover. Conventional evaluation methods require real drivers to operate actual vehicles under test conditions, resulting in high costs and limited ability to replicate diverse real-world driving scenarios. In this study, we propose a simulation-based DMS evaluation framework that utilizes virtual drivers and synthetic driving scenarios. We aim to construct digital human models representing diverse demographics across race, gender, and age as well as driving behavior animation assets, using Unreal Engine and camera-based skeleton data. Our focus is on replicating realistic gaze, facial expressions, and driver behaviors in accordance with international DMS evaluation standards. To date, we have developed a pipeline for skeleton data collection and refinement and are in the process of building a driver behavior animation library. This approach is expected to enhance the efficiency and scalability of DMS evaluation, while improving realism and ensuring regulatory compliance.

Keywords: Digital Twin, System Modeling and Control, Smart Buildings, Smart Cities and Infrastructures
Abstract: Biogas plants have been the subject of modeling and research in recent years. Such a plant is a system consisting of several components, the most important of which are the combined heat and power (CHP) unit that produces electricity and heat, the organic feed receiving area, the digester and the gas storage. All of these components have separate and detailed models developed for different purposes. A missing system approach that provides a sufficient level of detail and correlation between feeding, biogas production rate, storage level and CHP output was identified. Therefore, this work presents the concept of an approach that uses a deeper system understanding combined with processing of big amount of data rather than the popular fully machine learning based models. It outlines the methodology for establishing a relationship between CHP power and gas storage level in both forward and backward coupling. Possible physics based models are presented, which should serve as a predictor between the storage level of the biogas power plant and the feedstock. Finally, the individual steps are linked to a system-level approach that captures the fact that the power generated affects the state of the system and vice versa. The models presented are under development and use a sophisticated mix of statistical, data-driven and physical modelling methods. As input, they require recorded data that is minimally in detail and generally available for all biogas power plants. Together with the presented approach, the developed model should be able to serve as a digital twin of an arbitrary power plant that can be integrated into energy system models. Additionally, it should deliver quality indicators for the biogas produced at the respective plant. As such, the proposed methodology offers added value over existing approaches, which are often unidirectional or focused solely on individual components.

Keywords: Manufacturing Automation and Systems, Consumer and Industrial Applications
Abstract: Job scheduling is essential for enhancing productivity in manufacturing environments, with traditional methods primarily focusing on maximizing throughput by minimizing tardiness or completion times. However, in societies facing population decline and aging workforces, there is a growing need to consider workers' well-being alongside productivity goals. Factory workers are increasingly affected by demographic shifts and technological innovations, which leads to diverse lifestyle needs that require flexible working arrangements. This paper proposes a well-being oriented integer optimization approach to parallel machine scheduling that supports flexible working styles while maintaining production efficiency. Our method introduces job processing priorities and employs dummy jobs to represent non-working periods, creating constraints that respect workers' diverse availability patterns. Numerical experiments demonstrate that our proposed method successfully creates schedules that minimize tardiness while accommodating workers' diverse working hour preferences, contributing to both productivity and factory worker well-being in manufacturing operations.

Keywords: Decision Support Systems
Abstract: In multi-criteria sorting (MCS) problems, decision makers often express indirect and potentially conflicting preferences, especially when these preferences come from multiple sources or take different forms. Effectively handling such inconsistency is essential for generating reliable sorting results in MCS problems. This paper introduces a novel approach tailored to resolving inconsistency in MCS problems with heterogeneous preferences. The proposed framework begins with a consistency checking model to detect conflicts in the provided heterogeneous preferences. If the inconsistency is detected, a two-stage adjustment model will be applied: the first stage minimizes the number of adjustments, while the second stage preserves high-confidence preferences wherever possible. Once the adjusted preferences are consistent, a sorting result determination model is used to assign alternatives to predefined categories, with an emphasis on maximizing discriminative power between categories.

Keywords: Digital Twin, Cyber-physical systems, Robotic Systems
Abstract: The rapid development of humanoid robots to perform human-like tasks has introduced new possibilities in automation. In the context of advanced autonomous robotic operations, the potential damage due to unanticipated actions is a significant concern. To address this challenge, this paper presents a digital twin designed to replicate and monitor robotic operations in a virtual environment. Key features include the integration of manufacturer-provided kinematic models, real-time sensor fusion, and a user-friendly graphical interface to support reinforcement learning applications. The digital twin successfully mirrors the movements of a physical robot and records visual data for downstream machine learning tasks. Further, we implemented integration of procedurally generated virtual environments that allows for dynamic scenario creation, enabling robots to train and adapt to diverse and unpredictable conditions without the risks associated with real-world testing. This paves the way toward transitioning our entire laboratory into a digital ecosystem, providing a safe and controlled environment where robots can operate, learn, and improve autonomously.

Keywords: System Modeling and Control, Decision Support Systems, Cyber-physical systems
Abstract: This paper presents a metaethical framework, that potentially allows to conceptually grasp current arms races in artificial intelligence and the ethical problems, such as cyber influence operations, connected to these arms races. AI arms races do not slow down, but enhance the weaponisation of AI. Rather than solving ethical problems, the metaethical framework thus first and foremost can be considered as a map, helping to navigate through the complex reality underlying the arms races and ethical problems. In this sense, the metaethical framework should in the future be capable of evaluating different countermeasures such as deplatforming and watermarking.

Keywords: Cyber-physical systems, Smart Buildings, Smart Cities and Infrastructures, Intelligent Transportation Systems
Abstract: This paper presents a novel method for transforming large-scale historical expressway route search records into a three-dimensional (3D) Origin-Destination (OD) map, enabling data compression, efficient spatiotemporal sampling and statistical analysis. The study analyzed over 380 million expressway route search logs to investigate online search behavior related to tourist destinations. Several expressway interchanges (ICs) near popular attractions, such as those associated with spring flower viewing, autumn foliage and winter skiing, are examined and visualized. The results reveal strong correlations between search volume trends and the duration of peak tourism seasons. This approach leverages cyberspace behavioral data as a leading indicator of physical movement, providing a proactive tool for traffic management and tourism planning.

Keywords: Robotic Systems, Soft Robotics, Consumer and Industrial Applications
Abstract: Grasping and manipulation remain fundamental challenges in the effective deployment of robotic systems in real-world applications. In retail and supermarket scenarios, soft robotic grippers enable safe and efficient material handling. However, existing grasp planners are designed for rigid or suction-based grippers. Soft grasping is more challenging in terms of planning, estimation and sensing due to deflections in the gripper material on contact with the target. We present a system for soft robotic grasping using a custom gripper with a reconfigurable wrist that leverages reinforcement learning to augment existing vision-based techniques to adapt to the target object's geometry. This system includes a hidden superquadrics module to guide the adaptation of the gripper's palm configuration. We evaluate this system in a PyBullet simulation environment and compare it with a baseline synergy-based grasp strategy. Ongoing and future work involves transferring this planner to our physical robotic platform and evaluation in retail stores.

Keywords: System Modeling and Control, Consumer and Industrial Applications, Mechatronics
Abstract: Model-Based Development (MBD) has been widely used as an efficient product development method in industry.　A controller designed using MBD may not achieve the desired control performance due to the effects of disturbances and model errors in the actual plant. A hierarchical-type control system has been proposed as the control system architecture for products developed through MBD. In this control structure, a compensator is introduced to suppress the effects of disturbances and model errors in the actual plant. This paper proposes a PID-type compensator (GMV-PID compensator) based on Generalized Minimum Variance Control (GMVC) as a method for designing compensators in the hierarchical-type control system. The proposed method is intended for application to plastic processing machinery such as injection molding machines and film production machines. As part of ongoing work, this paper presents preliminary experimental results obtained by applying the proposed method to a pilot-scale slider-crank system, which simulates the toggle-type clamping mechanism of an actual injection molding machine.

Keywords: System Modeling and Control, Decision Support Systems, Modeling of Autonomous Systems
Abstract: Water-fertilizer integrated regulation system aims to to improve crop yield, nutrient use efficiency (NUE), and water use efficiency (WUE). This study develops an intelligent control system with a cloud server network to centralize data management and process. The system contains of the monitoring module, the intelligent cloud platform module, and the control terminal. In the monitoring module, the sensor network and phenotypic monitoring method were utilized to collect real-time crop and environment data. In the intelligent cloud platform module, a perception-computing-control integrated computing optimization framework was constructed to achieve the localization of control tasks by analyzing perceived data, training predictive control models, and optimizing deep reinforcement learning algorithms. The control terminal contains of task scheduling, equipment control, and data collection drive, as well as a reasonable deployment pipeline in the cloud. Our developed control model has increased crop productivity nearly 9% and achieved water resource savings nearly 15.6%. Multi-modal large models will be applied to adjust model parameters in different environments.

Keywords: Cooperative Systems and Control
Abstract: Recent technological advances have introduced autonomous mobile robots into public areas that were previously occupied only by pedestrians. When pedestrians and mobile robots coexist, the collision avoidance algorithm in which the robots can reach their goals without causing collisions or significant delays is important. One approach is to incorporate human sensation into the avoidance algorithm, and the braking index representing human braking sensation in two-dimensional environments was developed. The goal of this study was to verify whether the collision avoidance algorithm based on the braking index could work efficiently in a congested environment using agent simulations. The results showed that the agents with proposed algorithm achieved smoother and safer movement than the agents without it in a congested environment. Additionally, we found a traffic density-flow relationship between congestion and movement smoothness. This finding implies that this relationship could be applied to two-dimensional environments, such as shopping malls or airports.

Keywords: Smart Metering, Decision Support Systems, Intelligent Power Grid
Abstract: The evolution of metering technologies has enabled the collection of a vast amount of energy consumption data, offering new opportunities for more efficient energy management. While utility providers increasingly leverage machine learning and data visualization to simplify and optimize data analysis, current systems often present barriers in understanding collected data. This paper introduces a novel multi-agent architecture that integrates Large Language Models (LLMs) to enhance the interpretability of smart meter data. Developed within the Digital Enterprise initiative by Lutech S.p.A., the proposed framework enables natural language interactions and energy-related reports. An early evaluation has been conducted through a series of basic interaction tests, demonstrating the feasibility of the approach and its potential to improve data-driven decision-making.

Keywords: Decision Support Systems, Cyber-physical systems, System Modeling and Control
Abstract: This work-in-progress paper presents a concept for a holistic cybersecurity analysis. The proposed approach is designed to enable a quantitative assessment of the cybersecurity posture of an energy system, based on knowledge of its assets and their interconnections. Identified vulnerabilities are reported and categorized, while corresponding mitigation suggestions are collected and validated. At present, cybersecurity assessments in this domain still rely heavily on expert judgment, leading to subjective evaluations that hinder comparability and objectivity. To address this, the proposed approach aims to systematize and automate the analysis process. This paper is published at an early stage to actively invite feedback and foster community involvement, as broad expert input is essential to improve the quality, robustness, and acceptance of the approach. This paper documents a research project aimed at providing grid operators with an analysis tool to support decision-making in the future.

Keywords: Homeland Security
Abstract: Large Language Models (LLMs) have revolutionised the field of Natural Language Processing (NLP) and are currently being integrated into more critical domains, raising concerns about the possibility of hidden backdoors that could potentially allow collecting user data or manipulate output. This paper investigates the possibility of hidden backdoors by analysing network traffic during local LLM usage. Two models, DeepSeek-R1 and Mistral, were tested in experiments to have a comparison of LLMs from different geopolitical and regulatory environments. Using Ollama, a software that allows to run LLMs locally, three experiments were performed: 1) Monitoring TCP Connections on a per process level, 2) running the local LLM in a Docker container with full network isolation, and 3) monitoring all network traffic using Wireshark on a monitored Docker bridge. The results showed that there was no external network communication during the experiments. Anomalies due to other means than influence via a hidden backdoor were found such as DeepSeek's language output, which was in Chinese for certain prompts, even though the prompt was in English. In conclusion, our findings indicate that it is possible to locally isolate LLMs for critical usage, and that Docker-based network isolation could be a practical approach for detecting hidden backdoors in LLMs.

Keywords: Distributed Intelligent Systems, Large-Scale System of Systems, System Architecture
Abstract: Hyperautomation aims to digitize end-to-end business processes, but most of the current solutions and platforms still rely on pre-defined workflows to carry out complex tasks. However, business process automation can greatly benefit from recent technological innovations at the intersection of Large Language Models (LLMs) and Multi-Agent systems (MAS). In this paper, we present an Agentic Artificial Intelligence (AI) framework where a central LLM-driven orchestrator dynamically plans and delegates tasks to specialized LLM agents, which in turn exploit tools to act on business platforms and other external systems. A case study based on document management illustrates how the approach is able to deal with complex requests involving source file parsing and report generation, leveraging an approach based on Retrieval Augmented Generation (RAG) to enable knowledge sharing among specialized agents. Finally, the proposed framework introduces a practical blueprint for scalable, explainable Agentic AI in enterprise hyperautomation environments.

Keywords: AI and Applications
Abstract: The IEEE Lotfi A. Zadeh Award for Emerging Technologies recognizes individuals or teams for exceptional contributions to emerging technologies that have demonstrated significant impact, originality, and importance in recent years. This prestigious honor reflects IEEE’s commitment to advancing innovative technologies that shape the future and embodies the pioneering spirit of Lotfi A. Zadeh, the father of fuzzy logic. The award, which includes a bronze medal, certificate, and honorarium, will be presented by 2023 IEEE President Saifur Rahman and conferred to the awardee, Dimitar Filiev, during this session.

Keywords: AI and Applications, Application of Artificial Intelligence
Abstract: Artificial Intelligence (AI) is rapidly transforming the automotive industry. From enhancing vehicle performance and safety to streamlining manufacturing processes, AI-powered systems are driving significant advancements. This presentation, drawing on the speaker's 30+ years of experience in R&D, discusses the transformative role of AI in various automotive applications. We will explore the evolution of AI in the automotive sector, from early AI to deep learning and generative AI. We will examine the challenges and opportunities associated with integrating AI solutions. The presentation will highlight specific examples of AI applications in powertrain, vehicle systems, and autonomous driving. We will examine the lessons learned from applying data-driven AI techniques and discuss the potential of combining knowledge-based approaches with data-driven methods to further enhance AI's impact on the automotive industry.

Keywords: Trust in Autonomous Systems, System Architecture
Abstract: With autonomous systems playing critical roles across society and industry, their trustworthiness has become a foundational requirement—not only for technical reliability, but to ensure ethical, safe, and responsible deployment. As these systems increasingly make decisions that affect humans, trustworthiness is essential to earning acceptance, managing risk, and complying with evolving legal and societal expectations. This talk focuses on standards and frameworks for integrating trust into both smart system architectures and the broader cross-domain ecosystems that support their safe adoption and positive impact.

Keywords: Deep Learning, Neural Networks and their Applications, Machine Vision
Abstract: Recent advances in deepfake technology have created increasingly convincing synthetic media that poses significant challenges to information integrity and social trust. While current detection methods show promise, their underlying mechanisms remain poorly understood, and the large sizes of their models make them challenging to deploy in resource-limited environments. This study investigates the application of the Lottery Ticket Hypothesis (LTH) to deepfake detection, aiming to identify the key features crucial for recognizing deepfakes. We examine how neural networks can be efficiently pruned while maintaining high detection accuracy. Through extensive experiments with MesoNet, CNN-5, and ResNet-18 architectures on the OpenForensic and FaceForensics++ datasets, we find that deepfake detection networks contain winning tickets, i.e., subnetworks, that preserve performance even at substantial sparsity levels. Our results indicate that MesoNet retains accuracy at 80% sparsity on the OpenForensic dataset, with only 3,000 parameters. The results also show that our proposed LTH-based iterative magnitude pruning approach consistently outperforms one-shot pruning methods. Using Grad-CAM visualization, we analyze how pruned networks maintain their focus on critical facial regions for deepfake detection. Additionally, we demonstrate the transferability of winning tickets across datasets, suggesting potential for efficient, deployable deepfake detection systems.

Keywords: Deep Learning, Machine Learning, Neural Networks and their Applications
Abstract: Deep neural networks (DNNs) are widely used in perception systems for safety-critical applications, such as autonomous driving and robotics. However, DNNs remain vulnerable to various safety concerns, including generalization errors, out-of-distribution (OOD) inputs, and adversarial attacks, which can lead to hazardous failures. This survey provides a comprehensive overview of runtime safety monitoring approaches, which operate in parallel to DNNs during inference to detect these safety concerns without modifying the DNN itself. We categorize existing methods into three main groups: Monitoring inputs, internal representations, and outputs. We analyze the state-of-the-art for each category, identify strengths and limitations, and map methods to the safety concerns they address. In addition, we highlight open challenges and future research directions.

Keywords: Deep Learning, Image Processing and Pattern Recognition, Machine Learning
Abstract: 超声弹性成像 （USE）、射频 （RF） 数据中 保留高频内容，而 B 模 （BM） 数据 通过射频信号变换获得，遭受 大量信息丢失，导致次优 BM 散斑跟踪中的位移场。然而，射频 临床超声中通常无法获得数据 系统，从 BM 进行可靠的位移估计 序列对实用弹性成像至关重要。虽然 现有的基于深度学习的BM运动跟踪方法 提供可用的位移预测、其准确性和 应变图像质量仍然有限，尤其是在真实 体内数据。为了解决这个限制，我们建议将 无监督光流跟踪神经 网络——KDCLPWC-Net——融合知识 蒸馏和课程学习以增强斑点 跟踪 BM 数据的性能。具体来说，一个 引入跨模态知识蒸馏框架， 具有并行特征对齐模块 （FAM） 将多尺度特征表示从 改善轴向位移的教师网络 学生网络中的估计。此

Keywords: Deep Learning, Representation Learning, Machine Learning
Abstract: By utilizing the large amount of unlabeled data among distributed clients, federated semi-supervised learning (FSSL) has become a new research topic. However, the knowledge discrepancies among diverse clients and non-independent and identically distributed (Non-IID) data pose significant challenges to the model performance and generalization in FSSL. To tackle these challenges, we propose a FSSL framework with the dynamic aggregation and prototype retraining method (FedDAPR). FedDAPR improves the performance and robustness of the global model by developing a dynamic aggregation scheme. Moreover, to enhance the degraded training performance caused by the Non-IID data, we propose a prototype-based retraining mechanism for the classifier of the global model. Experimental results on two benchmark datasets demonstrate the effectiveness of the proposed FedDAPR.

Keywords: Assistive Technology, Biometrics and Applications,, Human Factors
Abstract: In recent decades, the detection of fatigue has been largely studied to improve safety and performance in various domains such as healthcare, transportation, and manufacturing. In fact, fatigue significantly affects cognitive and physical performance. There are numerous studies on the detection of fatigue, but most of them focus on binary classification (i.e., fatigue vs. resting state) or different levels of fatigue intensity, without distinguishing between specific fatigue types. This work aims to discriminate between cognitive and physical fatigue, and proposes the use of physiological signals to classify four distinct conditions: rest, cognitive fatigue, physical fatigue, and combined cognitive and physical fatigue. We analyze data from cardiac, eye, electrodermal, and electromyographic activity. We consider different feature selection methods, including correlation analysis, principal component analysis and sequential forward floating search method. Ultimately, we classify them using state-of-the-art machine learning methods. The highest classification accuracy (86.823%) is obtained from a support vector machine method using a selection of the features extracted from cardiac and electromyographic sensors. This study highlights the potential for real-time fatigue classification, which can enhance the adaptability of automated systems to human needs, particularly in high risk environments like transportation and healthcare. Furthermore, the findings suggest that fatigue monitoring can be effectively conducted with minimal sensor requirements, contributing to the design of more efficient wearable sensor systems.

Keywords: Assistive Technology, Haptic Systems, Wearable Computing
Abstract: Auditory directionality remains challenging for d/Deaf and hard of hearing (DHH) people. We evaluated a shoulder-mounted tactile system that vibrates bilaterally to signal sound direction and urgency in multi-person meetings. Eight DHH participants watched subtitled VR meetings with and without the device, pressing a trigger when their name or an alarm appeared and selecting the speaker afterward. The device increased reaction count for name calls and shortened reaction time for both name and alarm alerts (all p < 0.05), while speaker identification accuracy stayed high and unchanged. Post-experiment surveys confirmed strong demand for directional cues (p < 0.01) and alarm sound alerts (p < 0.05). Reflective Thematic Analysis of open-ended responses highlighted (i) the need to customize target sounds, (ii) meeting-focused use cases, and (iii) requests for refined vibration patterns and lighter design. These findings demonstrate the system’s value for timely notifications and emphasize directional feedback and customizability as critical to practical adoption.

Keywords: Brain-Computer Interfaces, Haptic Systems, Human-Computer Interaction
Abstract: Understanding the neural correlates of sensory imagery is crucial for advancing cognitive neuroscience and developing novel Brain-Computer Interface (BCI) paradigms. This study investigated the influence of imagined temperature sensations (ITS) on neural activity within the sensorimotor cortex. The experimental study involved the evaluation of neural activity using electroencephalography (EEG) during both real thermal stimulation (TS: 40 °C Hot, 20 °C Cold) applied to the participants' hand, and the mental temperature imagination (ITS) of the corresponding hot and cold sensations. The analysis focused on quantifying the event-related desynchronization (ERD) of the sensorimotor mu-rhythm (8-13 Hz). The experimental results revealed a characteristic mu-ERD localized over central scalp regions (e.g., C3) during both TS and ITS conditions. Although the magnitude of mu-ERD during ITS was slightly lower than during TS, this difference was not statistically significant (p>.05). However, ERD during both ITS and TS was statistically significantly different from the resting baseline (p<.001). These findings demonstrate that imagining temperature sensations engages sensorimotor cortical mechanisms in a manner comparable to actual thermal perception. This insight expands our understanding of the neurophysiological basis of sensory imagery and suggests the potential utility of ITS for non-motor BCI control and neurorehabilitation technologies.

Keywords: Image Processing and Pattern Recognition, Deep Learning, Neural Networks and their Applications
Abstract: Abstract— Ensuring operational safety in power grids is critical, particularly in substations where occlusion and small-object challenges are prevalent. We propose PowerOpsNet, a unified end-to-end framework for real-time unsafe behavior recognition. It combines an optimized YOLOv5-A detector—enhanced via Mosaic-8 augmentation, multi-scale features, and a dynamic focal loss—with a CNN-LSTM-GCN-based behavior recognition module. A benchmark dataset comprising 11 object types and 18 actions was collected from real substations. Experiments demonstrate 91.3% detection precision and 87.6% recognition accuracy, with inference speeds of 370ms and 200ms, respectively. PowerOpsNet supports real-time safety monitoring and is extensible to other high-risk domains.

Keywords: Image Processing and Pattern Recognition, Machine Vision
Abstract: Due to the complex structures of many characters, automatic font generation remains a challenging research task. Although existing methods achieve satisfactory performance, the generated characters still suffer from significant detail loss, such as stroke loss, stroke connection issues, or breakage, among others. To address this issue, we propose an approach based on Adaptive Feature Extraction (AFE) and the Detail Enhancement Module (DEM). AFE adaptively fuses local and global channel attention to enhance content feature extraction, enabling the network to focus on structurally important features, thereby improving character clarity and consistency. DEM first employs multi-scale dilated depthwise convolutions with feature aggregation to effectively enhance both local details and global contours in font structures. It then uses deformable convolutions to predict displacement maps and applies these maps to perform geometric feature deformation on the low-level feature maps from the content encoder. This approach improves style transfer accuracy while preserving structural consistency in fine-grained font generation. Experimental results demonstrate the superiority of our method, which can generate Chinese characters with rich details. The generated characters outperform those produced by existing methods. Furthermore, our approach can be extended to cross-lingual generation.

Keywords: Image Processing and Pattern Recognition, Machine Vision, AI and Applications
Abstract: The bird's eye view (BEV) representation is essential for accurate 3D perception tasks (e.g., 3D object detection) in autonomous driving for its precise localization, scale consistency, and modality independence. However, traditional depth-prediction-based methods, which rely on predicted depth distributions derived from semantic image features for BEV transformation, face challenges such as ambiguous depth-prior information and dependence on predefined depth distribution types. To address these limitations, we propose a novel multi-modal 3D object detection method, Depth Completed BEV (DC-BEV), which leverages ground-truth sparse LiDAR depth to guide BEV transformations, significantly enhancing depth estimation accuracy. Specifically, we introduce a Multimodal-Depth Completion (MDC) mechanism, which enriches sparse LiDAR depth into dense depth maps by integrating semantic and geometric cues from images. Additionally, to mitigate gradient instability caused by inadequate implicit supervision, we present an Explicit Depth Supervision (EDS) mechanism that directly supervises depth predictions using a dedicated depth loss. Comprehensive experiments conducted on the nuScenes dataset demonstrate that DC-BEV achieves superior performance, notably improving detection accuracy through enhanced depth estimation quality and robust BEV representation.

Keywords: Image Processing and Pattern Recognition, Machine Vision, Application of Artificial Intelligence
Abstract: Multiple person tracking is generally achieved by analyzing the color texture, shape, motion, and other information in RGB video images. However, this approach is challenging in low-light or dark environments without adequate lighting, leading to reduced accuracy. Monocular depth estimation methods have limitations in accurately estimating depth and cannot effectively distinguish between front and rear figures during occlusion. In this paper, we focused on the motion information instead of the appearance information of the detected persons. Moreover, we utilize additional estimated depth information instead of only the detected two-dimensional position in the image. Through experimental results, we show that the accuracy of tracking heat in an image can be improved by processing, such as Kalman filtering of three-dimensional position information, to compensate for tracking breaks due to occlusion.

Keywords—thermal images, monocular depth estimation, occlusion, person tracking, depth map correction

Keywords: Transfer Learning, Machine Vision, Image Processing and Pattern Recognition
Abstract: Source-free Unsupervised Domain Adaptation (SFUDA) aims to utilize a pre-trained source model on unlabeled target domains without accessing source data. The emergence of SFUDA has improved the robustness of UDA methods and enables domain adaptation to still perform well in fields with security and privacy considerations. Self-training is one approach to solve the domain gap problem in the SFUDA field, which iteratively selects high-confidence target samples as pseudo-labeled samples to guide target model learning. However, previous works only use the source domain model to generate the first round of pseudo labels before performing denoising tasks. This preprocessing is undoubtedly rough, because it ignores the domain gap between the source domain and the target domain. Given the existence of domain gap, some incorrect pseudo labels may never be corrected by subsequent denoising tasks. Inspired by the teacher-student model, this work proposes a method called Positive Sample Encouragement (PSE) that effectively addresses this issue in a coarse-to-fine manner. Specifically, we first conduct a contrastive learning task that enables the target model to effectively inherit from the source model, while allowing the source model to adapt to target data. Furthermore, we incorporate mutual information-based class balancing, that assigns high weights to high-confidence classes by learning pseudo-label feature distributions. Through extensive experiments on three widely-used benchmarks, we demonstrate that our proposed method achieves competitive performance compared with the state-of-the-art approaches.

Keywords: Machine Learning, Deep Learning, Neural Networks and their Applications
Abstract: In order to better dispatch police patrols, it is crucial to accurately predict crime hotspots. Considering crime prediction as a typical time series forecasting task (i.e., using crime history to predict future occurrences) has been shown to be effective. However, contextual features (e.g., demographics, economics, etc.) can also be introduced to improve the accuracy of the predictions as they relate to the formation of crime. Since such contextual features are usually considered to be static relative to the time scales of crime records, we first propose a parallel branch model with dedicated branches for each type of data so that temporal crime history and time-invariant contextual features can be processed coherently. Then, we incorporate both spatial and social dependencies into the model, considering that criminals may flee to neighboring areas and similar crime patterns may occur in areas with similar social functions. The experimental results confirm the effectiveness of our proposed model. In particular, our model achieves state-of-the-art performance with an accuracy of 75.3% and an AUC-ROC (area under the receiver operating characteristic curve) of 0.79.

Keywords: Machine Learning, Machine Vision, Hybrid Models of Computational Intelligence
Abstract: Motion control of human musculoskeletal models has been playing an essential role in the field of biomechanics and computer animation simulation for the last decade. However, challenges remain, including the high computational complexity of the new task and the difficulty of ensuring the authenticity of the motion data mapping. In this paper, we propose LI-AMCU, a novel modeling framework for motion control. To achieve more efficient control in the deployment of various tasks, we combine reinforcement learning with a GAN-like framework and propose an update module in the control policy based on the latent space. To guarantee a more natural and realistic movement of the data-driven model, we propose a geometric transformation learning module based on isomorphism. Furthermore, experiments on the Human3.6M and LaFAN1 datasets demonstrate that LI-AMCU achieves state-of-the-art performance and has the robustness and generalizability to enable complex musculoskeletal models to perform different motion tasks. We also visualize the various motions of physics-based simulated characters. The implementation code will be released after acceptance.

Keywords: Machine Learning, Metaheuristic Algorithms, Evolutionary Computation
Abstract: Black-box optimization (BBO) is a proven approach for solving complex real-world problems, such as hyperparameter optimization (HPO) in machine learning. HPO often involves expensive objective function evaluations, which limit the number of feasible evaluations. Accordingly, the demand for optimization methods with high parallel efficiency and fast convergence has increased. Sequential Uniform Design (SeqUD) satisfies these properties by iteratively applying uniform sampling and search space contraction (zooming) around the best solutions. However, in many HPO problems, only a few parameters strongly influence the objective function value, while others have little effect—a property known as low effective dimensionality (LED). Therefore, the uniform contraction utilized in SeqUD can lead to inefficient search for HPO problems. In this paper, we propose a new zooming strategy, referred to as Dynamic Importance-guided and Volume Enforcement (DIVE), for efficiently performing HPO with SeqUD. Numerical results utilizing benchmark functions assuming HPO characteristics show that the SeqUD with DIVE outperforms CMA-ES, a major evolutionary algorithm, in terms of both evaluation efficiency and parallel scalability under limited function evaluations.

Keywords: Machine Learning, Neural Networks and their Applications, AI and Applications
Abstract: Robustness against noise and input perturbations is a crucial requirement for trustworthy machine learning (ML). We introduce Robustness Assessment and Regularized Enhancement (RARE), a model-agnostic framework that evaluates and improves robustness in regression tasks, including neural and symbolic approaches. RARE centers on a user-specified noise model (e.g., Gaussian or Laplace) to differentiate intended from unintended input variations, yet in practice generalizes to unseen noise. Its four components are: (i) a significant-pattern operator that per feature synthesizes a single representative noise, i.e., worst-case in our tests, with best- or average-case also supported; (ii) a robustness assessor that measures the ratio of input deviation induced by significant patterns to output variation, analogous to estimating the inverse of local Lipschitz constants of the learned function; (iii) a data-augmentation module that increases training exposure to noise using the significant patterns; and (iv) a regularizer that embeds the metric in the loss to penalize low robustness. Because RARE targets the Lipschitz behaviour of the function under learning rather than the model's, it neither requires white-box access to model parameters nor assumes continuity or differentiability of the learning method. Consequently, RARE applies to off-the-shelf models such as Random Forests (RF) and even End-to-End Transformer-based Symbolic Models (E2E). Unlike adversarial methods, RARE does not require generating extensive synthetic data to improve robustness. Experiments showed MLPs trained with the regularized loss exhibit 7-33% higher robustness under Gaussian noise and up to a 32.3% improvement under high-variance Laplace noise. Regularized CNNs outperform unregularized CNNs in low-noise conditions under Gaussian noise. A regularized symbolic equation learner converges in as few as 3,000 epochs for some cases, while the non-regularized one converged in 20,000 epochs. Finally, experimental results show that compared to Meta's E2E, we achieve an average 66.83% reduction in fitting time, corresponding to an expected 3x speedup across all tested equations.

Keywords: Human-Machine Cooperation and Systems, Human Factors, Interactive Design Science and Engineering
Abstract: Maintaining superiority on the battlefield is vital in ensuring mission success. The use of advanced technologies such as Autonomy and Artificial Intelligence offers the ability for the human to let the machines do the heavy lifting of tasks; especially those that require rapid processing of complex information and within a dynamic environment. This paper outlines the MASTER SOUP project that demonstrates the use of AI to manage multiple sensors, whilst also introducing a new method to assist the multidisciplinary design team better understand how the system deals with decisions that occur during the mission. The use of a method utilizing decision strings is outlined and discussed. By adopting this method, it was found that it was beneficial to both Human Factors and AI Engineers in terms of designing the Human-Machine Teaming concept. The use of decision strings provides an intuitive methodology for identifying and deconstructing the nature of decisions within the Human-Machine Team. Further to this it can be used to the benefit of all members of the design team to facilitate the elicitation of design requirements that provide benefits across multidisciplinary fields.

Keywords: Human-Machine Cooperation and Systems, Human-Collaborative Robotics, Intelligence Interaction
Abstract: Fusing information from human observations can help robots overcome sensing limitations in collaborative tasks. However, an uncertainty-aware fusion framework requires a grounded likelihood representing the uncertainty of human inputs. This paper presents a Feature Pyramid Likelihood Grounding Network (FP-LGN) that grounds spatial language by learning relevant map image features and their relationships with spatial relation semantics. The model is trained as a probability estimator to capture aleatoric uncertainty in human language using three-stage curriculum learning. Results showed that FP-LGN matched expert-designed rules in mean Negative Log-Likelihood (NLL) and demonstrated greater robustness with lower standard deviation. Collaborative sensing results demonstrated that the grounded likelihood successfully enabled uncertainty-aware fusion of heterogeneous human language observations and robot sensor measurements, achieving significant improvements in human–robot collaborative task performance.

Keywords: Cooperative Work in Design, Human-Machine Cooperation and Systems, Human Factors
Abstract: Traditional methods for Technology Foresight often struggle to keep up with the fast changes in technology. This work looks at how crowdsourcing can improve Technology Foresight by using collective intelligence to spot trends and tackle complex problems. The goal of this research is to overcome the limits of traditional approaches by using online platforms and new technologies like big data and machine learning to make innovation more accessible. We examine how to apply crowdsourcing effectively in Technology Foresight, discussing its benefits, challenges, and real-world uses. We used a Rapid Review (RR) approach, combined with Large Language Models (LLMs), to refine how we select and analyze research. This method helped us find important insights and categorize applications in different areas. The findings show that crowdsourcing can speed up technology forecasting by bringing in diverse viewpoints and encouraging collaboration across fields. Examples include its use in urban planning, disaster management, and open innovation platforms. Furthermore, combining crowdsourcing with advanced technologies enhances its effectiveness in foresight processes. This work offers practical advice for organizations that want to use crowdsourcing in their innovation strategies and suggests future ways to integrate new technologies into Technology Foresight.

Keywords: Assistive Technology, Medical Informatics, Human-Machine Cooperation and Systems
Abstract: Cybernics treatment using the Hybrid Assistive Limb (HAL) can improve gait abilities. Contrary to conventional evaluations that compare 2-min walk distances without wearing HAL between pre- and post-intervention, we assess gait data measured by HAL during gait assistance, which enables the observation of gait changes that accompany the intervention. To establish this novel evaluation approach, it is essential to examine the relationship between changes in gait ability without HAL and gait patterns during HAL-assisted walking. Focusing on one individual with a neuromuscular disease, this study analyzed and evaluated the relationship between changes in the 2-min walk distance and changes in gait patterns during HAL-assisted walking. Principal component analysis (PCA) was employed to characterize the changes in gait patterns during HAL wear, followed by the creation of an individual model that predicts the rate of change in the 2-min walk distance without HAL from those features using the eXtreme Gradient Boosting (XGBoost). The model was then interpreted using SHapley Additive exPlanations (SHAP) to analyze the contributions of each feature to the prediction. Analysis of nine HAL-assisted walking trials and corresponding 2-min walk distance measurements revealed that changes in the 2-min walk distance were associated with alterations in specific gait patterns during HAL-assisted walking: knee joint angles, knee joint torques generated by HAL, and trunk pitch angles representing anterior-posterior trunk tilting. These findings clarified important features related to changes in gait ability within longitudinal gait pattern changes during cybernics treatment and demonstrated the utility of our analysis method and HAL-measured data for evaluating individual gait pattern changes.

Keywords: Human Performance Modeling, Human Factors, Human-Machine Interaction
Abstract: Cognitive impairment due to hypoxia significantly affects human performance, particularly during tasks that require sustained mental effort and engagement. Electroencephalography (EEG) has emerged as a valuable non-invasive tool for analyzing brain activity and cognitive performance. However, traditional EEG-based methods for human performance modeling, such as the Engagement Index and wavelet entropy, are limited in their ability to detect rapid and localized changes in brain dynamics that are critical for real-time modeling. This paper investigates Spectral Stability, an entropy-based method that captures fast, localized changes in brain activity by analyzing the hierarchical rankings of spectral intensities across EEG frequency bands. Using a data set collected under normoxic and hypoxic dual-task scenarios, we evaluate the ability of Spectral Stability to provide granular insight into oscillatory cognitive states in different regions of the brain. Our findings demonstrate that Spectral Stability reliably distinguishes between normoxic and hypoxic conditions across multiple brain regions, outperforming the Engagement Index in spatial specificity, while also correlating significantly with the Engagement Index, indicating shared underlying neural dynamics. These results highlight that Spectral Stability may provide additional key information within neuronal dynamics as a tool to advance cognitive state modeling and understanding real-time engagement variability under impairment.

Keywords: Autonomous Vehicle, Intelligent Transportation Systems, Trust in Autonomous Systems
Abstract: As autonomous vehicles (AVs) become an essential component of modern transportation, they are increasingly vulnerable to threats such as GPS spoofing attacks. This study presents an adaptive detection approach utilizing a dynamically tuned Density Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, designed to adjust the detection threshold (varepsilon) in real-time. The threshold is updated based on the recursive mean and standard deviation of displacement errors between GPS and in-vehicle sensors data, but only at instances classified as non-anomalous. Furthermore, an initial threshold, determined from 120,000 clean data samples, ensures the capability to identify even subtle and gradual GPS spoofing attempts from the beginning. To assess the performance of the proposed method, five different subsets from the real-world Honda Research Institute Driving Dataset (HDD) are selected to simulate both large and small magnitude GPS spoofing attacks. The modified algorithm effectively identifies turn-by-turn, stop, overshoot, and multiple small biased spoofing attacks, achieving detection accuracies of 98.62pm1%, 99.96pm0.1%, 99.88pm0.1%, and 98.38pm0.1%, respectively. This work provides a substantial advancement in enhancing the security and safety of AVs against GPS spoofing threats.

Keywords: Autonomous Vehicle, Trust in Autonomous Systems, Robotic Systems
Abstract: This study investigates the vulnerabilities of autonomous navigation and landing systems in Urban Air Mobility (UAM) vehicles. Specifically, it focuses on Trojan attacks that target deep learning models, such as Convolutional Neural Networks (CNNs). Trojan attacks work by embedding covert triggers within a model’s training data. These triggers cause specific failures under certain conditions, while the model continues to perform normally in other situations. We assessed the vulnerability of Urban Autonomous Aerial Vehicles (UAAVs) using the DroNet framework. Our experiments showed a significant drop in accuracy, from 96.4% on clean data to 73.3% on data triggered by Trojan attacks. To conduct this study, we collected a custom dataset and trained models to simulate real-world conditions. We also developed an evaluation framework designed to identify Trojan-infected models. This work demonstrates the potential security risks posed by Trojan attacks and lays the groundwork for future research on enhancing the resilience of UAM systems.

Keywords: System Modeling and Control, Conflict Resolution, Cooperative Systems and Control
Abstract: In this paper, we propose a conflict-free multi-agent flight scheduling that ensures robust separation in constrained airspace for Urban Air Mobility (UAM) operations application. First, we introduce Pairwise Conflict Avoidance (PCA) based on delayed departures, leveraging kinematic principles to maintain safe distances. Next, we expand PCA to multi-agent scenarios, formulating an optimization approach that systematically determines departure times under increasing traffic densities. Performance metrics, such as average delay, assess the effectiveness of our solution. Through numerical simulations across diverse multi-agent environments and real-world UAM use cases, our method demonstrates a significant reduction in total delay while ensuring collision-free operations. This approach provides a scalable framework for emerging urban air mobility systems.

Keywords: Autonomous Vehicle, Modeling of Autonomous Systems, Trust in Autonomous Systems
Abstract: The increasing demand for autonomous systems in complex and dynamic environments has driven significant research into intelligent path planning methodologies. For decades, graph-based search algorithms, linear programming techniques, and evolutionary computation methods have served as foundational approaches in this domain. Recently, deep reinforcement learning (DRL) has emerged as a powerful method for enabling autonomous agents to learn optimal navigation strategies through interaction with their environments. This survey provides a comprehensive overview of traditional approaches as well as the recent advancements in DRL applied to path planning tasks, focusing on autonomous vehicles, drones, and robotic platforms. Key algorithms across both conventional and learning-based paradigms are categorized, with their innovations and practical implementations highlighted. This is followed by a thorough discussion of their respective strengths and limitations in terms of computational efficiency, scalability, adaptability, and robustness. The survey concludes by identifying key open challenges and outlining promising avenues for future research. Special attention is given to hybrid approaches that integrate DRL with classical planning techniques to leverage the benefits of both learning-based adaptability and deterministic reliability, offering promising directions for robust and resilient autonomous navigation.

Keywords: Modeling of Autonomous Systems, Trust in Autonomous Systems, System Modeling and Control
Abstract: The integration of Unmanned Aircraft Systems (UAS) into the National Airspace System (NAS) requires robust encounter modeling tools to evaluate Detect-and-Avoid (DAA) systems. However, existing tools often lack the ability to model the unique dynamics of large UAS (lUAS) and small UAS (sUAS), and few are available as open-source solutions. In this paper, we introduce novel flight modeling concepts for lUAS and sUAS, developed within an open-source framework. For lUAS, we propose a hybrid modeling strategy that combines probabilistic manned aircraft models with UAS-specific performance constraints. A machine learning-based surrogate model is employed to streamline feasibility evaluation and enable the generation of realistic trajectories. The sUAS flight modeling technique enables mission-aware trajectory generation by incorporating geospatial data and customized control architecture for fixed-wing and multirotor configurations. These models capture a wide range of aircraft dynamics, offering the potential to generate versatile encounter datasets for DAA evaluation. By releasing them as open-source resources, we aim to encourage broader collaboration, inform regulatory development, and drive innovation in encounter modeling, all in support of the safe and effective integration of UAS into the NAS.

Keywords: Systems Safety and Security
Abstract: Sparkplug is an emergent open-source software specification for Industrial Internet of Things (IIoT) systems, designed to favor data integration and device interoperability in an MQTT infrastructure. Although the security issues of IIoT systems can have relevant safety implications, Sparkplug only provides basic security features and essential, coarse-grained access control (AC) mechanisms. Effective AC solutions for Sparkplug-based IIoT systems still need to be designed, and, due to the Sparkplug's increasing popularity and its recent definition as an ISO Standard, this has become a crucial need. To fill this void, this paper proposes an approach to efficiently enforcing fine-grained AC in Sparkplug-based IIoT systems. In particular, we define a fine-grained discretionary AC model and a related reference monitor implementing an efficient enforcement mechanism. Early performance evaluations show a reasonably low time overhead.

Keywords: Systems Safety and Security,, Human Factors, Networking and Decision-Making
Abstract: Urban gas systems, as critical components of urban infrastructure, are increasingly exposed to complex safety risks due to ongoing urban expansion and the resulting rise in accident frequency, posing significant threats to public welfare. Traditional cause analysis methods based on expert judgment often lack objectivity and reproducibility. This paper proposes a data-driven framework for identifying and evaluating the causes of urban gas accidents using information extracted from accident reports. A TF-H text mining technique is employed to extract causal factors, while the Apriori algorithm reveals their interrelationships. A Bayesian network is then constructed to infer accident severity and assess the relative importance of each factor. Validation with real-world cases confirms the effectiveness of the proposed approach. Based on the results, targeted management strategies are recommended to support the prevention and mitigation of urban gas accidents.

Keywords: Systems Safety and Security,, Resilience Engineering, Information Visualization
Abstract: The increasing integration of UAVs into civilian airspace has amplified the urgency for resilient and intelligent intrusion detection system (IDS) frameworks, as traditional anomaly detection methods often struggle to detect novel threats. A common strategy is to treat the unfamiliar attacks as out-of-distribution (OOD) samples; hence, inadequate mitigation responses can leave systems vulnerable, granting adversaries the capability to cause potential damage. Furthermore, conventional OOD detectors frequently fail to discriminate the stealthy adversarial attacks from OOD samples. This paper proposes a conditional generative adversarial network (cGAN)-based framework specifically designed to craft stealthy adversarial attacks that effectively evade IDS mechanisms. Initially, we construct a robust multi-class classifier as IDS which classifies the benign UAV telemetry data from known cyber-attack types, including Denial of Service (DoS), false data injection (FDI), man-in-the-middle (MiTM), and replay attacks. Leveraging this classifier, our proposed cGAN strategically perturbs known attack features, generating sophisticated adversarial samples engineered to evade detection through benign misclassification. Then, the generative stealthy adversarial samples is iteratively refined to maintain statistical similarity with out-of-distribution (OOD) samples while achieving a high attack success rate. To effectively detect these stealthy adversarial perturbations, a conditional variational autoencoder (CVAE) is implemented, using negative log-likelihood as a metric to distinguish adversarial samples from genuine OOD samples. Comparative analyses between CVAE-based regret analysis and traditional Mahalanobis distance-based detectors demonstrate that the CVAE’s negative log-likelihood significantly outperforms in detecting stealthy adversarial attacks from OOD samples. Our findings highlight the necessity of advanced probabilistic modeling techniques to reliably detect and adapt the existing IDS against novel, generative-model-based stealthy cyber threats.

Keywords: Systems Safety and Security, Environmental Sensing,, Human Perception in Multimedia
Abstract: The security and privacy protection of mobile devices have been critical issues. This study explores a novel acoustic side-channel attack method that infers the category of applications (apps) being used by analyzing the inaudible sounds emitted from chargers during the smartphone charging process. Unlike traditional methods that require direct contact or modification to the device, this approach demonstrates widespread availability and enhanced concealment. Through preprocessing, feature extraction, and model training, we conducted experiments using various smartphones, apps and chargers. The results indicate that both Random Forest and Transformer models achieve nearly perfect performance in app type inference under random splits cross-validation test, while the Transformer with end-to-end feature learning shows higher stability and generalizability under strict leave-one-out cross-validation, with an average accuracy of 63.52%. Finally, we give further analysis on effectiveness and efficiency of the proposed method, highlighting the potential threat of acoustic side-channel attacks and emphasizing the necessity of enhancing privacy protection and improved charging technologies.

Keywords: Biometrics and Applications,, Systems Safety and Security
Abstract: Face Anti-Spoofing (FAS) is essential for securing face recognition systems, but its practical deployment is severely hindered by the domain shift problem, where models trained on source domains perform poorly on unseen target domains. To address this critical challenge, this paper proposes VF-Mix, a framework whose novelty lies in the synergistic integration of multiple techniques to effectively balance feature invariance and discriminability. VF-Mix first utilizes a variational encoder to learn a regularized latent space. Within this space, cross-domain feature mixing and domain adversarial training work in concert to align distributions and learn generalized, domain-invariant cues. Crucially, supervised contrastive learning is simultaneously applied to enhance the class-separability of these features, ensuring that the push for invariance does not weaken discriminative power. This entire process is stabilized by a gradient alignment mechanism that harmonizes optimization signals from diverse source domains. Extensive experiments on standard benchmarks validate our approach. Under the challenging leave-one-out O&C&M→I protocol, VF-Mix achieves state-of-the-art performance, reducing the HTER to 3.75% and increasing the AUC to 99.34%, significantly outperforming prior methods. Ablation studies confirm that this synergy between components is critical for learning robust and discriminative features, presenting VF-Mix as an effective solution for advancing cross-domain face anti-spoofing.

Keywords: Visual Analytics/Communication
Abstract: In this work, we investigate the challenging task of Generalized Category Discovery (GCD). Given datasets collected from open-world scenarios comprising both labeled and unlabeled images, GCD aims to classify all unlabeled images while simultaneously identifying unlabeled novel categories. The fundamental challenge in GCD tasks stems from inherent annotation discrepancies between seen and novel classes within the dataset. The lack of reliable label supervision for novel classes in unlabeled data leads to significant disparities in the model’s learning between old and novel classes, which is termed the bias risk. Recent advancements in GCD have employed the entropy maximization algorithm to alleviate the bias risk. However, they fail to provide debiased optimization for unlabeled data, leading to models that struggle with extracting discriminative features from such data. To address these challenges, we have created an Open-world pseudo-contrastive learning framework named OpcGCD. Our OpcGCD framework implements a dynamic category-wise threshold mechanism, which employs a parametric prototype classifiers to generate debiased pseudo-labels for unlabeled samples. To facilitate the learning of discriminative feature representations, our proposed OpcGCD employs debiased pseudo-labels in the formulation of a contrastive learning loss. Extensive evaluations conducted on multiple GCD benchmark datasets demonstrate the robustness and effectiveness of the approach.

Keywords: Human-Machine Interaction, Human-Collaborative Robotics, Intelligence Interaction
Abstract: Unmanned Aerial Vehicles (UAVs) are increasingly important in dynamic environments such as logistics transportation and disaster response. However, current tasks often rely on human operators to monitor aerial videos and make operational decisions. This mode of human-machine collaboration suffers from significant limitations in efficiency and adaptability. In this paper, we present AirVista-II—an end-to-end agentic system for embodied UAVs, designed to enable general-purpose semantic understanding and reasoning in dynamic scenes. The system integrates agent-based task identification and scheduling, multimodal perception mechanisms, and differentiated keyframe extraction strategies tailored for various temporal scenarios, enabling the efficient capture of critical scene information. Experimental results demonstrate that the proposed system achieves high-quality semantic understanding across diverse UAV-based dynamic scenarios under a zero-shot setting.

Keywords: Cognitive Computing, Human-Computer Interaction, Multimedia Systems
Abstract: Current multi-modal large language models (MMLMs) primarily rely on instance-level feature statistics for cross-modal alignment. However, they commonly suffer three inherent limitations including vulnerability to outlier perturbations, neglect of inter-feature covariance structures, and local optimum trapping. Such limitations stem from a critical oversight—existing approaches disregard the global statistical structure of multi-modal data, treating cross-modal alignment as isolated feature-level alignment rather than systematic distribution-level alignment. To address these issues, this paper proposes Layer-wise Covariance Alignment (LCA), which firstly leverages the distribution-level alignment for cross-modal alignment. The effectiveness of LCA is validated through the use of parameter-efficient Low-Rank Adaptation (LoRA) on CLIP architectures. Experimental validation across 8 benchmarks demonstrates state-of-the-art performance, confirming the critical role of distribution-level alignment in overcoming sample-level optimization constraints for cross-modal learning.

Keywords: Human-Computer Interaction, Medical Informatics, Human-Machine Interaction
Abstract: Accurate prediction of intraoperative hazardous events and generation of effective intervention plans are critical to surgical safety, but face multiple challenges of real-time, accuracy, and interpretability. Large-scale language models have potential, but their high cost and potential ‘illusion’ problems limit their application in real-time clinical environments. Traditional multitask learning models are efficient but knowledge-constrained, making it difficult to capture complex reasoning processes. To bridge this gap, this paper proposes a multi-objective distillation knowledge enhancement model-KEMO, which innovatively adopts a multi-objective chain-of-thought distillation framework to not only mimic the prediction results of the instructor's LLM, but also explicitly migrate its structured reasoning process to the lightweight student model, which improves the answerability of the model by synergistically optimising the three objectives of event prediction, reasoning alignment and scenario generation. Interpretability. Meanwhile, combined with the Knowledge Graph-based Retrieval Augmented Generation mechanism, validated medical knowledge is dynamically injected to enhance the accuracy and reliability of decision-making and reduce model illusion. The experimental results show that the KEMO model significantly outperforms traditional models of the same magnitude in intraoperative hazardous event prediction and prognostic proposal generation, and achieves a performance comparable to that of a large faculty model.The KEMO model effectively bridges the gap between the large language model and the actual clinical application, and facilitates the transformation of the large model knowledge to the actual clinical deployment.

Keywords: AI and Applications, Application of Artificial Intelligence, Expert and Knowledge-Based Systems
Abstract: While current machine-learning-based AI techniques have been spectacularly successful, their present applications still leaves many important open questions – for example, how to make their results more reliable or, at least, how to gauge how reliable is each AI recommendation. In this paper, we argue that to fully answer these questions, we need to go beyond the current AI techniques, and that in this development, systems-, human-, and cybernetics-based ideas not only naturally appear, they seem to provide a way to the desired answers.

Keywords: AI and Applications, Cloud, IoT, and Robotics Integration
Abstract: The authors hope that this overview and suggestions will stimulate and contribute to further ongoing discussions and interesting research work and industrial applications in some fields of Systems, Man and Cybernetics.

Keywords: Agent-Based Modeling, Machine Learning
Abstract: Integrating artificial intelligence into educational technology presents great opportunities for automated educational systems. These systems could relieve teacher resources and support underperforming students. However, creating systems that are adaptive, scalable, and factually correct is resource intensive. Furthermore, there are many technologies that are prevalent, but lack systematic ways to integrate them into existing educational technologies. Building on reinforcement learning and large language models (LLMs), this paper introduces a multi-agent framework for adding both a reinforcement learning-based tutor and an LLM-driven peer to educational systems. The integrated architecture is unified with a central ontology, acting as a symbolic knowledge base and facilitating data transformation. We also detail a novel windowed experience sharing method for improving reinforcement learning training efficiency when dealing with similar environments and low-data situations. We present our architecture and simulated results to verify the reinforcement learning algorithm as an adaptive tutor, as well as the integration of an LLM-driven peer and educational outcomes from said integration.

Keywords: Neural Networks and their Applications, Computational Intelligence, Agent-Based Modeling
Abstract: Embodiment is a key aspect of human intelligence, related to our ability of to identify the context of the individual experiences at a given time, corresponding to the natural constraints represented by our body. Learning from higher cognitive functions and social coordination between humans can support building intelligent robot systems and facilitates harmonious human-machine interactions. This position paper provides an overview of neural structures and neural dynamics contributing to human cognitive functions, including multisensory integration, Gestalt formation, perception, and building sensory associations. Embodied cognitive principles are illustrated through the intentional action-perception cycle. The results are applied to the design novel algorithms for brain-inspired cognitive robotics. Example scenarios include imitation learning, and the emergence of dialogue patterns in social robotics settings.

Index Terms—Neurodynamics; Intentionality; Cognitive robotics; Embodied intelligence; Social Coordination; Neurodynamics.

Keywords: AI and Applications, Hybrid Models of Computational Intelligence, Deep Learning
Abstract: Large language models (LLMs) have demonstrated state-of-the-art performance across a wide range of natural language processing tasks. However, high-performing models are typically accessible only via APIs, incurring substantial inference costs. Cascade methods address this by initially employing a cheaper model and escalating to a stronger one only when necessary. Nevertheless, existing cascade approaches struggle to select a reliable representative response and assess the overall reliability of free-form outputs, as they rely on exact text matching. To overcome these limitations, we propose Keyword-inspired Cascade (KiC), a novel framework for cost-efficient free-form text generation. KiC identifies the most representative answer among multiple outputs from a weaker model and evaluates the semantic alignment of other responses with it. Based on the degree of alignment, KiC determines whether to accept the weaker model’s output or escalate to a stronger model. Experiments on three free-form text generation benchmarks show that KiC achieves 97.53% of GPT-4’s accuracy while reducing API costs by 28.81% on average, and even outperforms GPT-4 in a specific benchmark.

Keywords: AI and Applications, Image Processing and Pattern Recognition, Deep Learning
Abstract: With recent advancements in text-to-image (T2I) models, effectively generating multiple instances within a single image prompt has become a crucial challenge. Existing methods, while successful in generating positions of individual instances, often struggle to account for relationship discrepancy and multiple attributes leakage. To address these limitations, this paper proposes the relation-aware disentangled learning (RaDL) framework. RaDL enhances instance-specific attributes through learnable parameters and generates relation-aware image features via Relation Attention, utilizing action verbs extracted from the global prompt. Through extensive evaluations on benchmarks such as COCO-Position, COCO-MIG, and DrawBench, we demonstrate that RaDL outperforms existing methods, showing significant improvements in positional accuracy, multiple attributes consideration, and the relationships between instances. Our results present RaDL as the solution for generating images that consider both the relationships and multiple attributes of each instance within the multi-instance image.

Keywords: AI and Applications, Machine Learning, Optimization and Self-Organization Approaches
Abstract: While Large Language Models (LLMs) show promise for task planning, their efficacy diminishes in complex, long-horizon tasks within dynamic, partially observable environments, primarily due to challenges in long-term reasoning and effective adaptation from experience. A key limitation of current approaches is the insufficient utilization of historical trajectory information. To overcome these challenges in the context of Partially Observable Markov Decision Process (POMDP) planning, this paper introduces DMA-MCTS (Dynamic Memory-Augmented Monte-Carlo Tree Search), a framework that integrates Monte Carlo Tree Search (MCTS) with LLMs, augmented by a novel dynamic memory and reflection system. The core technical contributions include: (1) a dual-layer semantic memory repository enabling efficient context-aware retrieval of past experiences; (2) a memory-enhanced UCT selection strategy biased by historical Q-values to guide search; and (3) a differentiated reflection mechanism employing LLMs to extract generalizable knowledge from both successful and failed trajectories. Comprehensive evaluations conducted on complex object rearrangement tasks within the VirtualHome simulator demonstrate that DMA-MCTS significantly outperforms relevant baselines, including standard LLM-MCTS approaches, in terms of task success rate, generalization capabilities, and planning efficiency. These results underscore the critical importance of integrating structured dynamic memory and systematic reflection mechanisms for developing highly adaptive and effective LLM-based agents capable of tackling long-horizon planning problems.

Keywords: AI and Applications, Machine Learning, Transfer Learning
Abstract: The reward model is a critical component in training powerful large language models. However, current methods largely overlook the inherent subjectivity and variability in human preferences. Typically, this issue is indirectly addressed through model-side approaches, such as ensemble methods to estimate uncertainty from multiple predictions or uncertainty-aware regression to predict mean and variance. These indirect approaches fail to capture the intrinsic distributional characteristics and inter-rater disagreements present in human judgments. In contrast, we propose a direct, label-side solution by explicitly modeling human preference distributions. We recover missing information from scalar ratings to construct meaningful distribution labels. By employing Label Distribution Learning (LDL), each dimension of the resulting multidimensional discrete distribution explicitly corresponds to a specific preference score, naturally reflecting the subjective and multidimensional nature of human evaluations. Our approach improves explainability and confidence estimation, while also enabling more effective data selection and sample-efficient test-time adaptation. Empirical results demonstrate that our method not only achieves state-of-the-art performance but also provides a robust framework for uncertainty quantification and nuanced preference modeling.

Keywords: AI and Applications, Machine Vision, Deep Learning
Abstract: Robotic grasping, the ability of robots to reliably secure and manipulate objects of varying shapes, sizes and orientations, is a complex task that requires precise perception and control. Deep neural networks have shown remarkable success in grasp synthesis by learning rich and abstract representations of objects. When deployed at the edge, these models can enable low-latency, low-power inference, making real-time grasping feasible in resource-constrained environments. This work implements Heatmap-Guided Grasp Detection, an end-to-end framework for the detection of 6-Dof grasp poses, on the GAP9 RISC-V System-on-Chip. The model is optimised using hardware-aware techniques, including input dimensionality reduction, model partitioning, and quantisation. Experimental evaluation on the GraspNet-1Billion benchmark validates the feasibility of fully on-chip inference, highlighting the potential of low-power MCUs for real-time, autonomous manipulation.

Keywords: Human-Collaborative Robotics, Systems Safety and Security, Supervisory Control
Abstract: 3D anomaly detection is critical in industrial quality inspection. While existing methods achieve notable progress, their performance degrades in high-precision 3D anomaly detection due to insufficient global information. To address this, we propose Multi-View Reconstruction (MVR), a method that losslessly converts high-resolution point clouds into multi-view images and employs a reconstruction-based anomaly detection framework to enhance global information learning. Extensive experiments demonstrate the effectiveness of MVR, achieving 89.6% object-wise AU-ROC and 95.7% point-wise AU-ROC on the Real3D-AD benchmark.

Keywords: Human-Collaborative Robotics, Systems Safety and Security, Supervisory Control
Abstract: This paper addresses the challenge of fully unsupervised image anomaly detection (FUIAD), where training data may contain unlabeled anomalies. Conventional methods assume anomaly-free training data, but real-world contamination leads models to absorb anomalies as normal, degrading detection performance. To mitigate this, we propose a two-stage framework that systematically exploits inherent learning bias in models. The learning bias stems from: (1) the statistical dominance of normal samples, driving models to prioritize learning stable normal patterns over sparse anomalies, and (2) feature-space divergence, where normal data exhibit low intra-class consistency while anomalies display high diversity, leading to unstable model responses. Leveraging the learning bias, stage 1 partitions the training set into subsets, trains sub-models, and aggregates cross-model anomaly scores to filter a purified dataset. Stage 2 trains the final detector on this dataset. Experiments on the Real-IAD benchmark demonstrate superior anomaly detection and localization performance under different noise conditions. Ablation studies further validate the framework’s contamination resilience, emphasizing the critical role of learning bias exploitation. The model-agnostic design ensures compatibility with diverse unsupervised backbones, offering a practical solution for real-world scenarios with imperfect training data.

Keywords: Haptic Systems, Human-Machine Interface, Shared Control
Abstract: Haptic shared control (HSC) is a control approach where both a human operator and an autonomous controller exert force on a shared, physical control terminal of a robot. HSC allows the fluid combination of human intelligence and machine precision, improving task performance and reducing the human workload. When the autonomous controller is unreliable, however, the frequent disagreement between the human and the machine input adversely affects the task performance and the operator workload. Previous research showed that allowing the human operator to adjust the strength of the haptic guidance can alleviate the problem. However, the decision of when and how much to adjust the haptic guidance strength remained a challenging task for a human operator. This study proposed an approach that utilized a grip mechanism for adjusting and recommending an appropriate strength. In the proposed approach, the human operator was in charge of deciding the haptic strength via the grip angle. The autonomous controller suggested the appropriate haptic strength based on its reliability by applying proportional control to the grip angle. An experiment with a simulated remotely operated vehicle (ROV) teleoperation with HSC guidance showed that the proposed method is effective in aiding the adjustment of autonomous controller input strength, as well as in a marginal reduction in workload.

Keywords: Human-Machine Interface, Human-Machine Interaction, Human-Machine Cooperation and Systems
Abstract: This paper applies model-based biodynamic feedthrough (BDFT) cancellation to a touchscreen dragging task during realistic vertical (heave) and lateral (sway) aircraft turbulence, to mitigate erroneous turbulence-induced inputs. One-size-fits-all (OSFA) BDFT models were used to model the influence of turbulence accelerations on finger position, achieving average quality-of-fits of 61% and 69% in the vertical and horizontal screen directions, respectively. On average, 27% of the touch input error variance was mitigated these OSFA models, with individualized models providing only a marginal improvement (+4%). The application of OSFA models identified from a condition with equally-scaled turbulence in heave and sway (adjusted intensity) to the realistic turbulence condition did not significantly affect cancellation performance, indicating that BDFT models may not need to be adaptive to varying motion intensity. However, consistent with earlier work, BDFT dynamics were found to vary between vertical and horizontal finger movements, with BDFT dynamics exhibiting lower stiffness and a higher static gain for vertical BDFT. On average, the linear BDFT-related component of touch input errors contributed 41% of the overall error variance, indicating that current linear BDFT model may need to be extended to include nonlinear effects, such as varying finger friction.

Keywords: Telepresence, Shared Control, Human-Machine Interface
Abstract: Task performance in terms of task completion time in teleoperation is still far behind compared to humans conducting tasks directly. One large identified impact on this is the human capability to perform transformations and alignments, which is directly influenced by the point of view and the motion retargeting strategy. In modern teleoperation systems, motion retargeting is usually implemented through a one time calibration or switching modes.

Complex tasks, like concatenated screwing, might be difficult, because the operator has to align (e.g. mirror) rotational and translational input commands. Recent research has shown, that the separation of translation and rotation leads to increased task performance. This work proposes a formal motion retargeting method, which separates translational and rotational input commands. This method is then included in a optimal control based trajectory planner and shown to work on a UR5e manipulator.

Keywords: Deep Learning, Representation Learning, Machine Vision
Abstract: Perception systems in autonomous driving rely on sensors such as LiDAR and cameras to perceive the 3D environment. However, due to occlusions and data sparsity, these sensors often fail to capture complete information. Semantic Occupancy Prediction (SOP) addresses this challenge by inferring both occupancy and semantics of unobserved regions. Existing transformer-based SOP methods lack explicit modeling of spatial structure in attention computation, resulting in limited geometric awareness and poor performance in sparse or occluded areas. To this end, we propose Spatially-aware Window Attention (SWA), a novel mechanism that incorporates local spatial context into attention. SWA significantly improves scene completion and achieves state-of-the-art results on LiDAR-based SOP benchmarks. We further validate its generality by integrating SWA into a camera-based SOP pipeline, where it also yields consistent gains across modalities.

Keywords: Representation Learning, Computational Intelligence in Information, Expert and Knowledge-Based Systems
Abstract: The efficiency of Intelligent Transport Systems (ITS) runs on high-quality traffic data, however in real-world deployments, sensors failures, communication interruptions or other issues often lead to missing data, which affects the performance of ITS. Aiming at traffic data’s complex spatio-temporal characteristics, although the latent factorization of tensors (LFT) model has been widely used for missing-value completion, its non-convex objective function makes it difficult for first-order optimization methods to approximate high-quality second-order stationary points, therefore limiting the improvement of the completion accuracy. To address the issues, this paper proposes an incomplete tensor complementation model combining Tucker decomposition and second-order optimization strategy to improve the complementation accuracy and convergence stability. To address the issues, this paper proposes a Second-order Latent Factorization of Tensors based on Tucker Decomposition (SLTD), and efficiently solves it via Gauss-Newton approximation, so that it can significantly improve the model performance while keeping the computational cost low. Experimental results on real traffic datasets (in terms of average vehicle speed) from four cities verify the effectiveness of SLTD. Results show that the proposed model outperforms existing prevailing methods in terms of accuracy and provides a better solution for traffic data completion.

Keywords: Representation Learning, Deep Learning, AI and Applications
Abstract: Self-supervised contrastive learning has demonstrated effectiveness in time series representation learning. However, existing methods still exhibit three major limitations: limited robustness to noise, missing values, and distribution shifts; reliance on a binary contrastive objective that overlooks similarity information between time series instances; and coarse-grained contrast on raw observations that fails to capture true underlying relationships between time series instances. To address these limitations, we propose a novel framework that achieves robust representation learning for time series through decomposition and fine-grained similarity-guided contrast. Specifically, we apply decomposition at the input stage to smooth noise and missing values, and the resulting disentangled trend-seasonal representations provide adaptability to distribution shifts. Furthermore, we introduce a similarity-guided contrastive loss that incorporates similarity information between instances. Additionally, our method enables fine-grained contrast through separate trend contrasting and seasonal contrasting. Extensive experiments on forecasting, anomaly detection, and classification tasks demonstrate that our framework achieves state-of-the-art performance. Further analyses validate its robustness and the effectiveness of its design.

Keywords: Representation Learning, Deep Learning, Transfer Learning
Abstract: ChronosRL is a hybrid zero-shot trading framework that plugs transformer-derived Chronos embeddings into deep reinforcement learning (DQN, PPO, A2C, RPPO), eliminating the handcrafted indicators that dominate crypto trading. For every daily bar, we pass a two-channel price–volume series through Chronos-small to obtain a 768-dimensional embedding capturing temporal patterns; the RL agent then decides Buy / Sell / Hold under single-position and 5% stop-loss. Across six USDT pairs from July 1–Nov 30 2024, ChronosRL delivers a mean ROI of 102.8% (std 79.7%), surpassing Buy-and-Hold (98.4%) and a Bollinger-MA baseline (46.4%). The best configuration — DQN + Chronos — achieves a Sharpe 0.78 while operating zero-shot, demonstrating the power of pretrained temporal representations to generalise in volatile markets. These results suggest that merging foundation models with RL is a promising path toward data-driven trading strategies.

Keywords: Representation Learning, Transfer Learning
Abstract: This paper introduces DRANet-SWD as a novel complete pipeline for disentangling content and style representations of images for unsupervised domain adaptation (UDA). The approach builds upon DRANet by incorporating the sliced Wasserstein discrepancy (SWD) as a style loss instead of the traditional Gram matrix loss. The potential advantages of SWD over the Gram matrix loss for capturing style variations in domain adaptation are investigated. Experiments using digit classification datasets and driving scenario segmentation validate the method, demonstrating that DRANet-SWD enhances performance. Results indicate that SWD provides a more robust statistical comparison of feature distributions, leading to better style adaptation. These findings highlight the effectiveness of SWD in refining feature alignment and improving domain adaptation tasks across these benchmarks. Our code can be found here.

Keywords: Neural Networks and their Applications, Deep Learning, Computational Intelligence
Abstract: We present SmoothRot, a novel post-training quantization technique to enhance the efficiency of 4-bit quantization in Large Language Models (LLMs). SmoothRot addresses the critical challenge of massive activation outliers, by integrating channel-wise scaling with Hadamard transformations. Our technique effectively transforms extreme outliers into quantization-friendly activations, significantly improving quantization accuracy. Experiments conducted on popular LLMs (LLaMA2 7B, LLaMA3.1 8B, and Mistral 7B) demonstrate that SmoothRot consistently reduces the performance gap between quantized and FP16 models by approximately 10-30% across language generation and zero-shot reasoning tasks, without introducing additional inference latency.

Keywords: Neural Networks and their Applications, Hybrid Models of Computational Intelligence, Application of Artificial Intelligence
Abstract: Texture roughness perception is crucial for autonomous robots to perform manipulation, quality inspection, and material discrimination in unknown environments. This work proposes an approach to combine vibration and force data using the VibroTact sensor for texture roughness classification. Vibration and force data are first processed by CNN and ANN models, and then combined using a Bayesian framework. This approach is evaluated by recognizing 15 textures with different roughness (7 soft and 8 hard textures) using individual ANN and CNN models, and is compared against the Bayesian combination of both methods. Texture data is collected by mounting the VibroTact sensor on a robotic arm and using three sliding exploratory procedures (vertical, diagonal, and circular sliding). The texture roughness recognition results achieve 100% accuracy using the combined approach, which improves the performance of individual ANN and CNN models which range from 87.50% to 100% accuracy. The results also show that diagonal and vertical sliding are optimal for recognizing hard and soft textures, respectively. This approach demonstrates its potential for industrial robotics applications that require texture discrimination.

Keywords: Neural Networks and their Applications, Image Processing and Pattern Recognition
Abstract: Accurate segmentation of brain tumors is crucial for clinical diagnosis and treatment. In recent years, state space modeling (SSM) demonstrates its potential in long-range dependency modeling. However, it still has limitations in 3D medical image applications. Moreover, segmentation accuracy is often affected by high-frequency noise and the complex details inherent in brain tumor images. To overcome these limitations, we propose a novel frequency mamba U-Net model, i.e., FMambaU-Net, which breaks through the bottleneck of SSM in 3D images, reduces noise and improves the robustness and accuracy of brain tumor segmentation by integrating improved Mamba and frequency domain operations in U-Net. Specifically, we embed a dynamic weighted (DW) Mamba module into the bottleneck of 3D U-Net, which enhances the processing of complex tumor morphology by prioritizing critical regions. Additionally, we design and integrate a 3D frequency domain fusion module (FDFM) into the skip connections, which uses 3D fast Fourier transform (FFT) to separate frequency components and assign lower weights to high frequencies, thereby reducing high-frequency noise and improving segmentation accuracy. Evaluations on the BraTS 2020 and 2021 datasets show that FMambaU-Net outperforms both the baseline and state-of-theart methods. On BraTS 2020, the DSCs for enhanced tumor (ET), tumor core (TC), and whole tumor (WT) are 79.41%, 84.18%, and 91.62%, respectively, while on BraTS 2021, the DSCs are 88.59%, 91.30%, and 91.88%. These results demonstrate the effectiveness and competitiveness of FMambaU-Net in brain tumor segmentation.

Keywords: Neural Networks and their Applications, Image Processing and Pattern Recognition
Abstract: This paper introduces C2-YOLOv8-obb, a novel QR code localization model that enhances the YOLOv8-obb model through two key innovations: (1) the integration of a Convolutional Block Attention Module (CBAM) into the backbone feature extraction network, enabling the model to dynamically focus on critical QR code features under chal- lenging conditions such as strong lighting or background clutter; and (2) the replacement of traditional convolution layers in both the backbone and neck with CMUNeXt Blocks, improving the global extraction capability of model. Exper- imental results demonstrate that the proposed C2-YOLOv8- obb model outperforms traditional contour detection methods under strong lighting conditions and complex environments, achieving a 2.28 times improvement in the effective recognition range. Compared to state-of-the-art models such as YOLOv7, YOLOv7-tiny, and YOLOv8-obb, C2-YOLOv8-obb achieves significant improvements across all key metrics, including a recall and mAP@50 exceeding 99%, precision reaching 98%, and mAP@50:95 achieving 92.1%. These results validate the effectiveness of the proposed method.

Keywords: Neural Networks and their Applications, Image Processing and Pattern Recognition, AIoT
Abstract: From the perspective of real-time processing, such as autonomous driving, sudden failures could potentially lead to severe and even life-threatening accidents. In particular, hardware faults in AI models used for real-time decision-making can compromise safety. To prevent these accidents before they happen, it is crucial to detect failures promptly. Moreover, for small and power-constrained devices such as drones, it is essential to develop a fault-tolerant AI that is computationally efficient and minimizes memory usage and power consumption. Approximated Triple Modular Redundancy (TMR) with quantization has been proposed for convolutional layers, which enables fault-tolerant inference with reduced computational cost. However, sufficiently efficient and reliable fault-tolerant methods for quantized Vision Transformers based on Transformer blocks have not yet been developed. In this paper, we propose Approximated Dual Modular Redundancy (DMR) for fault detection and Approximated TMR for fault recovery, specifically designed for Vision Transformers. In our evaluation, the proposed Approximate TMR was applied to an 8-bit quantized Swin Transformer. The results show that it reduces the computational cost significantly compared to conventional TMR. Furthermore, it successfully detects faults when single-bit flips occur with a probability exceeding 1%. We also demonstrate that the proposed Approximated TMR maintains higher accuracy than existing methods, such as Ranger and Clipper, even under single-bit flip faults.

Keywords: Cooperative Systems and Control
Abstract: We consider Byzantine-resilient distributed optimization with event-triggered communication. The proposed algorithm is designed to handle non-convex optimization problems in a network of agents where some agents may exhibit Byzantine behavior. Each normal agent has an estimate of a critical point of the global cost function and transmits the estimate to neighbors when the difference between the current and previously communicated values exceeds a predefined threshold. Normal agents then update their estimates by averaging the received values from a trusted set that is determined through the Iterative Outlier Scissor (IOS) procedure. By combining the event-triggered communication and the IOS filtering procedure, the proposed approach ensures resilience to Byzantine behavior and guarantees convergence even in adversarial settings.

Keywords: Cooperative Systems and Control
Abstract: This paper addresses decentralized event-triggered reinforcement learning with safety constraints. Each agent has an individual reward function and safety constraints that depend on the joint actions of agents. The objective is to maximize the team's long-term return while satisfying the safety constraints. We formulate the reinforcement learning problem as a nonconvex-concave min-max optimization problem and propose a decentralized policy gradient algorithm. Each agent has estimations for the optimal primal and dual solutions of the min-max optimization problem. Different from the existing decentralized reinforcement learning, the agents share these estimates only when the error exceeds a predefined threshold. We show that the estimates of agents converge to a neighborhood of a locally optimal solution while effectively reducing the communication overhead.

Keywords: Cooperative Systems and Control
Abstract: This study addresses risk-averse distributed reinforcement learning in multi-agent systems, where agents collaboratively optimize their policies with a risk index in the objective function. Unlike conventional reinforcement learning, which maximizes the expected cumulative reward, the proposed risk-averse reinforcement learning incorporates Conditional Value at Risk (CVaR) to account for rare but significant adverse events. We propose a risk-averse distributed reinforcement learning algorithm based on the policy gradient method, which enables agents to collaboratively update their policies by sharing information through a communication network. Through a theoretical analysis, we show that the proposed algorithm ensures agreement among agents on their estimated policies. Moreover, we show that the consensus value converges in a neighborhood of a locally optimal solution, which ensures that each agent's learned policy remains aligned with risk-aware optimization criteria.

Keywords: Cooperative Systems and Control, Adaptive Systems, Robotic Systems
Abstract: This paper addresses the robust time-constrained consensus control of a class of multi-agent systems (MASs) subject to nonlinear dynamics and external disturbances. Adaptive compensation technique is utilized to eliminate the negative effects of nonlinearities and disturbances. New distributed finite-time consensus control strategies are designed to ensure bounded consensus in MASs by leveraging adaptive compensation signals. The finite-time stability of the MAS is proved by utilizing the Lyapunov theorem. Finally, simulation outcomes involving multiple unmanned marine systems within a multiagent framework confirm the efficacy of the proposed approach.

Keywords: Cooperative Systems and Control, Autonomous Vehicle
Abstract: This study investigates the use of Multi-Objective Particle Swarm Optimization (MOPSO) to optimize the hyper- parameters of Cooperative Adaptive Cruise Control (CACC) controllers, aiming to enhance performance in dynamic traffic scenarios. The proposed approach, named CACCpo, identifies hyperparameter combinations on Pareto fronts for a CACC controller based on the reduction of its accumulated error and overshoot, which are related to the system’s response time and stability, respectively. The proposed CACC controller, tuned with MOPSO, achieves an accumulated error of 21.21 and an overshoot of 6.38%, outperforming other controllers in speed and accuracy when responding to variations in the leader vehicle’s speed. In addition, it provides smoother acceleration and deceleration with minimal oscillations, ensuring a more stable and efficient system. Moreover, it maintains a constant safe distance between vehicles, which is essential for passenger safety and comfort.

Keywords: Intelligent Power Grid, Control of Uncertain Systems
Abstract: Recent years have witnessed the increasing adoption of unmanned aerial vehicles (UAVs) for power grid inspection, as they gradually replace conventional hazardous manual operations. However, conventional metaheuristics and operations research-based path planning algorithms suffer from long computation times for large-scale problems, thus making them unsuitable for real-time multi-UAV scheduling under flight time and energy consumption uncertainty. To solve the multi-UAV online path planning problem, this paper proposes a multi-agent deep reinforcement learning (MADRL) algorithm that performs online path planning based on real-time environmental and UAV state information, with the goal of minimizing the total travel distance. We employ resource preservation and decision sharing to handle real-time cooperation under travel time uncertainty while preventing resource conflicts. We design a Safety Mask mechanism that constrains dangerous UAV actions to address energy consumption uncertainty. Experiments on instances with 40-400 towers show that our algorithm requires minimal computation time and generates higher-quality solutions compared to other baseline algorithms in large-scale tower scenarios.

Keywords: Intelligent Power Grid, Control of Uncertain Systems, Cyber-physical systems
Abstract: With the large-scale integration of renewable energy sources, the Optimal Power Flow (OPF) problem in modern power systems has become increasingly uncertain and complex, making it difficult for conventional methods to strike a balance between computational efficiency and optimization quality. While reinforcement learning (RL) holds potential in such high-dimensional and dynamic optimization scenarios, it still encounters major hurdles, including slow convergence, insufficient exploration, and a high risk of converging to local optima under uncertain conditions caused by renewable energy integration. To address these issues, this paper models and decouples the total uncertainty into the environmental-noise-induced uncertainty and the lack-of-knowledge-induced uncertainty. Then, based on the decomposition, this paper proposes (1) a novel uncertainty-driven exploration formula that converges better and (2) an uncertainty-aware experience replay buffer, which dynamically identifies and focuses on decision regions with high uncertainty to improve RL’s performance. Experimental results on a real 179-bus power system confirm that the proposed method improves exploration efficiency and yields higher overall returns.

Keywords: Intelligent Power Grid, Fault Monitoring and Diagnosis, System Architecture
Abstract: With the development of artificial intelligence, many visual detection methods have been applied to power systems. However, diverse tasks cause existing solutions to face low detection accuracy and resource constraints in edge power scenarios. In this paper, we propose a low-rank enhanced lightweight multimodal LLM framework for efficient edge power visual detection. The framework can effectively balance the model performance with the resource constraints of edge deployment by introducing multimodal LLM and fusing data augmentation strategies and low-rank optimization. First, to enhance the adaptability of diverse visual detection tasks, we design a diverse data augmentation strategy for multimodal LLM to solve the problem of insufficient power scene data. Then, we reduce the number of training parameters by a low-rank optimization technique to enable the model to run efficiently on resource-limited edge devices. Experimental results demonstrate that our proposed framework can improve the accuracy by more than 3% on average under different power visual detection tasks. It can also save 76.49% of graphics memory consumption and accelerate training by 81.88%.

Keywords: Intelligent Power Grid, Intelligent Green Production Systems
Abstract: This paper examines the effectiveness of various synthetic data generation methods for deterministic wind power forecasting. Specifically, this work evaluates four approaches—Gaussian Mixture Models (GMMs), t-Copula, DoppelGANger, and FCPFlow—by comparing the forecasting performance, measured using Mean Absolute Error and Root Mean Squared Error, of models trained on synthetic versus real datasets. Our findings indicate that statistical methods (such as GMM and t-Copula) achieve notably better performance under limited data availability. However, the deep generative model FCPFlow yields superior results when sufficient training data is available. These findings suggest that the choice of synthetic data generation method should be informed by the specific data availability context.