A modification to the Gyrator Capacitor (GC) magnetic model is proposed to correct errors that may arise when running the GC model of magnetic structures on common circuit simulators. The proposed method adds a conduction path to the DC component of the magnetic flux, due to a DC bias current, a path that is absent in the original model. This helps to distribute correctly the MMFs across the capacitors. The conduction path is realized by placing resistors in parallel to the capacitors to help distribute correctly the DC component of the MMFs across the capacitors. Theoretical considerations backed by circuit simulation validate the proposed approach.

The development of fifth-generation (5G) technology marks a significant milestone for digital communication systems, providing substantial improvements in data transmission speeds and enabling enhanced connectivity across a wider range of devices. However, this rapid increase in data volume also introduces new challenges related to transmission latency, reliability, and security. This paper introduces KyMLP-LDPC, a novel approach that integrates a multi-layer parallel LDPC (MLP-LDPC) algorithm with Kyber, a post-quantum cryptography scheme, to accelerate and enable reliable and secure transmission. MLP-LDPC partitions the LDPC parity check matrix into processing groups to streamline parallel decoding and minimize message collisions during transmission, thereby accelerating error correction operations. Kyber encrypts data preemptively to safeguard against potential attacks. The effectiveness of our proposed method is evaluated using both image data and signals transmitted through an additive white Gaussian noise communication channel. Evaluation results demonstrate that the proposed method achieves superior performance in terms of error correction capabilities and data security compared to existing approaches.

Structural Health Monitoring (SHM) is a critical aspect of ensuring the safety and longevity of infrastructure such as bridges, buildings, and dams. Traditional SHM techniques, while effective, often rely on manual inspections and can be labor-intensive, time-consuming, and prone to human error. Recent advancements in Artificial Intelligence (AI) have the potential to revolutionize SHM by automating data analysis, improving predictive capabilities, and enhancing overall system efficiency. This article explores the integration of AI in SHM, highlighting key innovations, challenges, and future directions for this transformative technology.

This paper presents a design approach for dual-band dual-polarized (DBDP) waveguide antenna array. A novel transmission network using orthogonal parallel plate waveguide (PPW) is proposed for the excitation of dual polarization. Dual-band bandstop structures are incorporated into the PPW feeding network. They suppress the leakage between the two polarizations and improve the transmission efficiency of the network. They also enable a wide range of frequency ratio from 1 to 2.64 between the two bands. A new orthogonal linear signal generator (LSG) enables ultra-wideband, dual-polarized quasi-TEM excitation, ensuring consistent and reliable performance across a broad bandwidth. Fed by four sequentially rotated single-ridge 1-to-4 power dividers, the LSG effectively generates dual-polarized signals. For demonstration, a K/Ka-band shared-aperture dual-polarized array antenna has been designed, analyzed and experimentally verified. The array features a wide impedance bandwidth (17.7-21.2 GHz and 27.5-31 GHz), high isolation (>50 dB), and excellent cross polarization discrimination (XPD) (>40 dB). The prototype demonstrates high aperture efficiency up to 80% and consistent radiation patterns at both bands, affirming its potential for SATCOM systems.

6G networks intend to be highly dynamic, automated, and distributed while enabling novel user-centric services, which has raised the attention to trustworthiness aspects. In this context, trustworthiness has been pointed out as a key driver for the development of 6G technology. However, a common understanding of the definition and extension of trustworthiness is still missing in the community. Therefore, this paper aims at providing critical definitions for trustworthiness to steer the development of 6G by building upon 3 pillars: trustworthiness attributes, context, and assessment. First, we provide a comparative study of attributes for trustworthiness and identify the set of self-contained attributes that allow for a quantitative characterization of trustworthiness in 6G systems. Then, we recognize and discuss the main misconceptions regarding trustworthiness in the context of 6G and propose an assessment framework of two components for trustworthiness evaluation. This framework includes a numerical scaling of relevance of attributes, a scoring system, and a labeling approach to admit flexibility to identify a degree of trustworthiness evaluation for different capabilities in 6G.

This paper presents an innovative system for dynamic 3D scene perception tailored for battlefield environments, leveraging unmanned intelligent agents equipped with binocular vision and inertial measurement units (IMUs). The system processes binocular video streams and IMU data to deploy advanced deep learning techniques, including instance segmentation and dense optical flow prediction, facilitated by a specially curated target dataset. The integration of a ResNet101+FPN backbone for model training results in a combat unit type recognition accuracy of 91.8%, a mean Intersection over Union (mIoU) of 0.808, and a mean Average Precision (mAP) of 0.6064. A dynamic scene localization and perception module utilizes these deep learning outputs to refine pose estimations and enhance localization accuracy by overcoming environmental complexities and motioninduced errors, typically associated with SLAM methodologies. Application tests conducted within a simulated battlefield metaverse environment demonstrate a 44.2% improvement in self-localization accuracy over traditional ORB-SLAM2 Stereo methods. The system efficiently tracks and annotates dynamic and static battlefield elements, continuously updating global maps with precise data on agent poses and target movements. This work not only addresses the dynamic complexity and potential information loss in battlefield scenarios but also lays a foundational framework for future enhancements in network capabilities and environment reconstruction methodologies. Future developments will focus on precise identification of combat unit models, multiagent collaboration, and the application of 3D scene perception to advance real-time decision-making and tactical planning in joint combat scenarios. This approach holds significant potential for enriching the battlefield metaverse, fostering deep human-machine interaction, and guiding practical military applications.

Biodiversity plays a crucial role in maintaining the balance of the ecosystem. However, poaching and unintentional human activities contribute to the decline in the population of many species. Hence, active monitoring is required to preserve these endangered species. Current human-led monitoring techniques are prone to errors and are labor-intensive. Therefore, we study the application of deep learning methods like Convolutional Neural Networks (CNNs) and transfer learning, which can aid in automating the process of monitoring endangered species. For this, we create our custom dataset utilizing trustworthy online databases like iNaturalist and ZooChat. To choose the best model for our use case, we compare the performance of different architectures like DenseNet, ResNet, VGGNet, and YOLOv8 on the custom wildlife dataset. Transfer learning reduces training time by freezing the pre-trained weights and replacing only the output layer with custom, fully connected layers designed for our dataset. Our results indicate that YOLOv8 performs better, achieving a training accuracy of 97.39 % and an F1 score of 96.50 %, surpassing other models. Our findings suggest that integrating YOLOv8 into conservation efforts could revolutionize wildlife monitoring with its high accuracy and efficiency, potentially transforming how endangered species are monitored and protected worldwide.

Conventional optical networks are limited by static operational methods, hindering scalability and effectiveness. As networks operate with reduced margins to maximize resource utilization, the risk of hard failures increases, necessitating efficient failure prediction systems and accurate quality of transmission (QoT) estimation. Effective management requires the detection of soft failures, accurate bit error rate (BER) predictions, and dynamic network operations to maintain minimal margins. Machine learning (ML) offers promising solutions for automating these tasks, significantly enhancing failure management and network reliability. This article provides an extensive overview of ML techniques applied to optical networks, specifically focusing on failure management. Key ML techniques discussed include network kriging (NK) for performance estimation and failure localization, support vector machine (SVM) for classification tasks, convolutional neural networks (CNNs) for signal analysis and soft failure identification, and generative adversarial networks (GANs) for synthetic data generation and soft failure detection. It also explores the application of artificial neural networks (ANNs), autoencoders (AEs), Gaussian process (GP), long shortterm memory (LSTM), and gated recurrent units (GRUs) in optical networks. The paper surveys ML techniques for earlywarning and failure prediction, failure detection, identification, localization, magnitude estimation, and soft failure detection and prediction. Emphasizing automations, it discusses how ML algorithms can streamline failure management processes, reducing manual intervention and service disruptions. The potential of large language models (LLMs) and digital twins (DTs) for further advancements in automating failure management, optimizing performance, and network optimization in optical networks is also examined. LLMs significantly advance network management by improving network design, diagnosis, security, and autonomous optimization through the integration comprehensive domain resources and intelligent agents. These advancements are paving the way towards achieving artificial general intelligence and fully automated optical network management.

This preliminary study examines the temporal dynamics of multisensory integration in early to mid-adulthood, identifying five Regions of Interest (ROIs) and analyzing integration times from 0-500ms. Age-related differences in multisensory processing, especially in the middle-aged group, are noted. The use of ERP-based channel selection played a crucial role in enhancing age group classification accuracy, achieving 96.6% accuracy with the Random Forest classifier using a small set of selected 13 channels. Results show early integration consistently occurs within 200-325ms across age groups, with audio stimuli integrating slower than visual stimuli, and AV integration times falling in between. The study finds delayed early integration in audio-leading conditions (A50V) and faster integration in visualleading conditions (V50A). This optimized channel selection improves the ergonomics of EEG-based age group classification and simplifies the clustering process. It demonstrates the effectiveness of using minimal electrodes and straightforward features for multisensory integration tasks during early to mid-adulthood.

Ensuring the correct positioning of the electrode array during cochlear implant surgery is crucial for achieving optimal results. Traditionally, the locations of the electrodes are assessed through radiological imaging after the surgery. However, electrical impedance measurements have recently emerged as a promising alternative for electrode localization. The aim of this study is to assess the performance of various machine learning algorithms to regress electrode locations using impedance telemetry alongside other data sources such as morphological parameters. We conducted a comprehensive performance analysis on a selection of different models and features in an evaluation dataset of 118 cases. This included 17 cases with up to two extracochlear electrodes. A final evaluation was performed on a hold-out dataset consisting of 13 cases. All cases used the same lateral wall electrode array with a length of 28 mm. Model performance was benchmarked against existing models, emphasizing those previously published. The best-performing model for predicting linear insertion depth (Extremely Randomized Trees) achieved a mean absolute error of 0.8 mm ± 0.6 mm (mean ± standard deviation) using leave-one-out cross-validation. We further reviewed the models in terms of feature importance and sensitivity to improve their interpretability and reliability. The gradient direction of the impedance matrix was found as one of the most important features. Our results demonstrate that our machine learning approach is superior to previous models and has potential for use in routine clinical practice. In future studies, it needs to be confirmed that the models can generalize to other, i.e., shorter or longer, electrode arrays.

Named Entity Recognition (NER) is a crucial component in extracting structured information from unstructured text across various domains. A novel approach has been developed to address the variability in domain-specific annotations through the integration of a unified label schema, significantly enhancing cross-domain NER performance. The study involved comprehensive modifications to the Mistral Large model, including adjustments to its architecture, output layer, and loss function, to incorporate the aligned label schema effectively. The methodology encompassed rigorous data collection, preprocessing, and evaluation processes, ensuring robust model training and validation. Evaluation metrics such as precision, recall, F1-score, and accuracy demonstrated substantial improvements, validating the efficacy of the label alignment algorithm. The research highlights the model's ability to generalize entity recognition capabilities across diverse domains, making it adaptable to various linguistic and contextual details. The implications extend to numerous applications reliant on accurate entity recognition, including information retrieval, question answering, and knowledge base population, demonstrating the broader impact of the findings. Through these significant advancements, the study contributes to the development of more intelligent and adaptive NER systems capable of handling the complexities of diverse and evolving textual environments.

Wearable sensors enable to monitor patients in their everyday living and working environments. However, since they generate highly personal data, their protection from misuse or unauthorized use is crucial. Under certain conditions, it is important to examine whether the data they generate uniquely describe their deployment environment. This can be used as an additional measure to certify the authenticity of the data. In this paper, we uniquely identify persons based on the measurements of wireless electrocardiograms and motion sensors. We let 34 subjects perform 7 different activities while synchronously sampling a 5-lead wireless electrocardiogram, two 3D accelerometers, and a 3D gyroscope. Using both time-and frequency-domain features, we train a multi-layer CNN to achieve a classification accuracy exceeding 98%. Our model is able to uniquely classify 88% of the subjects with 100% accuracy, even though more than 82% of them have similar physical constitution (they are young people between the ages of 21 and 24, with a mean age of 22 years and standard deviation of 1.9 years). In order to demonstrate the variability of the data, we perform activity recognition using the same datasets; the model is able to classify the activities with an average accuracy of 92%.

Proper classification of documents is of tremendous importance for an organization. As digital copies of documents are now available due to technological advances, it has become convenient to classify them automatically using machine learning or deep learning algorithms. Deep CNNs have been widely applied for document image classification, whereas newly introduced transformer-based models have also presented favorable outcomes. In this paper, one of the advanced and very highperforming deep CNN models, ConvNext V2, developed very recently, has been adapted for document image classification task, that leverages the imitation of the self-attention mechanism of transformers through the use of masked autoencoders. Prior studies have suggested that models pre-trained on ImageNet may not perform optimally when directly applied to document classification tasks. Nevertheless, our findings reveal substantial enhancements in performance using ConvNext V2, indicating that further domain-specific pre-training (for instance, on the RVL-CDIP dataset) might not be essential for attaining high accuracy. Our results demonstrate that effective, direct application of ImageNet pre-trained models can yield significant benefits. The model has been applied to one of the most standard document image classification datasets Tobacco-3482. The results show a high overall accuracy of 92.25% with fast convergence, outperforming several other state-of-the-art methods. The source code for this work can be accessed by using the following link: https: //github.com/MdSaifulIslamSajol/document-tobacco-convnextv2

Fisheye cameras got their name from fisheye lenses, used by them to capture extremely wide-angled images, often providing a hemispherical or panoramic view. However, the fisheye lens is characterized by its distinctive visual distortion, where straight lines appear curved, and objects at the edges of the frame are elongated or stretched. Fisheye cameras have played an irreplaceable component in vision-first detection approaches in autonomous driving due to their ability to capture a complete view of objects that are extremely close-range to the ego vehicle, making them most critical to detect from the safety point of view. Fisheye cameras are also very specialized, with limited but critical usage in robots like autonomous vehicles (AVs). This review paper fills the gap in our literature by walking across how to leverage the pros of fisheye cameras by minimizing their cons. This paper also provides a good one-stop destination of literature for anyone willing to deploy a fisheye camera in the autonomy stack of a robot. Finally, we also discuss future research directions to enlighten the progress of the fisheye camera perception domain.

In this paper, we propose a conceptual framework and prove "The AI Incompleteness Theorem" within that framework. The theorem presented here states that it is not possible to guarantee an AI/ML inferencing system that is hallucination-free, even with an infinite amount of model training data and with an infinite amount of compute resources to train AI/ML models with that data. We prove this theorem using the properties of infinite sets, Cantor's theorem and Hilbert's Grand Hotel paradox. We also provide a new definition of AGI (artificial general intelligence) within this conceptual framework. We define a system having attained AGI as a system that does not produce infinitely many hallucinations when presented with infinitely many input datasets. We present a corollary that states AGI is not possible even with an infinite amount of training data and an infinite amount of compute resources.

This paper proposes a hybrid deep actor-critic framework for the optimal operation of a phase-changing soft open point (PCSOP) in an unbalanced distribution network. The framework combines algorithmic features of off-policy reinforcement learning and imitation learning. The proposed method comprises a policy-guiding module based on the PCSOP physics and an adaptive dynamic experience replay buffer. The policy-guiding module facilitates the agent’s navigation of the complex action space of the PCSOP. The dynamic experience replay accelerates agent training by leveraging expert demonstrations. As part of the design process, the paper also proposes a data-driven linearization of operational power losses in PCSOPs to enhance the convergence of nonlinear AC optimal power flow calculations without compromising accuracy. The proposed framework was trained and tested on a modified three-phase IEEE-33 bus test feeder. Results demonstrate the superiority of our framework compared to three different methods, including the conventional nonlinear AC optimal power flow.

Federated learning (FL) is a distributed machine learning (ML) technique that enables multiple, decentralized clients to collaboratively learn a shared ML model without sharing training data. However, deploying FL presents challenges related to energy efficiency, heterogeneous devices, and fault tolerance. This paper introduces novel techniques that address all these challenges almost for the first time. We propose an Adaptive Client Scheduling (ACS) algorithm to efficiently manage heterogeneous clients clustered by characteristics like processing power. Our proposed energy-efficient CNN and ACS methods are designed to reduce the computation overhead and energy consumption associated with techniques such as hyperparameter optimization and Adaptive Weight Pruning (AWP) algorithm. The results obtained from the MNIST and Cifar-10 datasets demonstrate that our proposed CNN and ACS algorithms can reduce the computation overhead and energy consumption by up to 10% and 20%, respectively, compared to baseline methods while achieving similar or better accuracy compared to baseline CNNs. Furthermore, the AWP algorithm on our CNN model could achieve higher energy-efficiency in FL, resulting in additional energy savings of up to 5% in fault-tolerant fog-edge environments. To create a fault-tolerant fog-edge environment, we utilize the task replication technique (TR) to improve reliability by 12%. TR is used to create multiple instances of a task and distribute them to different fog nodes. The combination of our proposed methods for energy-efficient CNN and ACS, along with AWP and task replication, enables highly energy-efficient and fault-tolerant FL processes.

We demonstrate here efficient teleoperation by both valid and arm-amputated participants on Reachy2, an opensource humanoid robot made by Pollen Robotics that ranked 2nd at ANA Avatar XPRIZE. The 3D parallel joint Orbita included at neck and wrist levels enables reproducing natural head and arm movements. This translates into an accurate egocentric view, and provides the opportunity to test a novel prosthesis control based on natural arm movement reconstructed by an Artificial Neural Network (ANN) that receives movement goal and stump motion as inputs. Direct first-person teleoperation by able-bodied participants (n=14) elicited a 100% success rate for grasping objects at various locations, with best possible usability scores and small workload scores. High success rates (>92%) were also obtained when all distal joints from the elbow onward were operated with the novel prosthesis control, using a movement goal identified through gaze-guided computer vision. While usability and workload scores were slightly degraded when able-bodied participants used the prosthesis control as compared to direct teleoperation, both were scored similarly well by a sample from our target population (n=8 participants with transhumeral limb loss). This platform and control schemes offer broad perspectives for prosthesis control, but also for teleoperation robotics and human-robot interactions.