Recently, in-car monitoring has emerged as a promising technology for detecting early-stage abnormal status of the driver and providing timely alerts to prevent traffic accidents. Although training models with multimodal data enhances the reliability of abnormal status detection, the scarcity of labeled data and the imbalance of class distribution impede the extraction of critical abnormal state features, significantly deteriorating training performance. Furthermore, missing modalities due to environment and hardware limitations further exacerbate the challenge of abnormal status identification. More importantly, monitoring abnormal health conditions of passengers, particularly in elderly care, is of paramount importance but remains underexplored. To address these challenges, we introduce our IC3M, an efficient camera-rotation-based multimodal framework for monitoring both driver and passengers in a car. Our IC3M comprises two key modules: an adaptive threshold pseudo-labeling strategy and a missing modality reconstruction. The former customizes pseudo-labeling thresholds for different classes based on the class distribution, generating class-balanced pseudo labels to guide model training effectively, while the latter leverages crossmodality relationships learned from limited labels to accurately recover missing modalities by distribution transferring from available modalities. Extensive experimental results demonstrate that IC3M outperforms state-of-the-art benchmarks in accuracy, precision, and recall while exhibiting superior robustness under limited labeled data and severe missing modality.

In recent years, increasing model complexity and improvements in self-attention mechanisms have enabled language models to perform remarkably well on tasks involving human language understanding and generation. However, limitations in preserving context over long sequences continue to present challenges, as models often struggle to retain coherence across extended discourse. The introduction of the Trans-Contextual Embedding (TCE) Mechanism provides a novel approach to addressing these challenges through a context-aware embedding layer integrated into transformer architectures, uniquely designed to maintain semantic depth and logical continuity. Through incorporating TCE into the initial embedding layer, the model achieves improved contextual coherence by dynamically weighting contextual relevance, leading to more accurate representation of complex, interdependent text elements. A series of experiments evaluated the TCE Mechanism on various opensource models, with results indicating a marked improvement in context continuity, semantic density, and reduced error rates, suggesting the applicability of TCE in a range of complex language tasks. Furthermore, the model achieved these improvements with minimal increase in computational overhead, presenting TCE as a scalable option for enhancing LLM capabilities in real-world applications. These findings lay foundational insights into context preservation mechanisms, positioning TCE as a significant advancement for the development of sophisticated and contextually robust language models.

The next generation of Wi-Fi is meant to achieve ultra-high reliability for wireless communication. Several approaches are available to this extent, some of which are being considered for inclusion in standards specifications, including coordination of access points to reduce interference. In this paper, we propose a centralized architecture based on digital twins, called WiTwin, with the aim of supporting wireless stations in selecting the optimal association according to a set of parameters. Unlike prior works, we assume that Wi-Fi 7 features like multi-link operation (MLO) are available. Moreover, one of the main goals of this architecture is to preserve communication quality in the presence of mobility, by helping stations to perform reassociation at the right time and in the best way.

Antenna gain is an important metric for most modern communication systems. It can be determined several ways, but most commonly by producing an incident plane wave using a reference antenna, and measuring the received power at the antenna of interest. By measuring the received power several wavelengths away in an isolated environment, like an anechoic chamber, the antenna gain can be determined. This and other similar methods work well for frequencies above 2 GHz, but lower frequency measurements can be logistically challenging and expensive due to the large facilities required, as well as the lack of readily available broadband absorber materials. This work presents a new method for determining antenna gain using a transmission line based near field S-parameter measurement in a waveguide. To provide evidence for the proposed method, two monopole antennas are modeled over an infinite ground plane using full wave electromagnetics, and both are experimentally measured within a waveguide. It was found that there is very good agreement between the model and measurement, providing evidence of the validity of the method.

The escalating complexity and adaptability of ransomware attacks have exposed critical gaps in traditional detection models, which struggle to effectively identify evolving threat behaviors without a significant margin of error. The Adaptive Pattern Scoring Mechanism (APSM) offers a novel approach to this challenge, leveraging a dynamic scoring framework that recalibrates based on real-time behavioral anomalies, enabling enhanced precision in distinguishing ransomware activities from benign events. APSM addresses inherent limitations in signaturebased and heuristic methods by integrating an adaptive scoring algorithm, which continuously adjusts detection thresholds and assigns differential weights across key behavioral dimensions, achieving a robust balance between detection accuracy and resource efficiency. Empirical evaluation reveals APSM's high accuracy and precision rates, with lower false-positive rates than conventional approaches, indicating its significant capability to detect ransomware with minimal operational disruption. The integration of dynamic scoring not only optimizes APSM's performance in real-time environments but also enables scalability across high-load conditions, marking a substantial advancement in automated cybersecurity infrastructures. Ultimately, APSM provides a sophisticated, resource-efficient solution that aligns closely with the requirements of modern ransomware detection, showing its value as a responsive tool in proactive threat management.

This article presents a novel design methodology for bandpass filters applied to rectangular waveguide (RW) technology using TE 201 singlets with inductive irises. The conventional circuit model for an inductive iris interconnecting two RW sections, which typically handles only the TE 10 mode, is extended to include the propagation of the TE 20 mode through a wider RW section. The impact of the reactances at the source and load terminals of the circuit model is analyzed to develop a unique solution for the synthesis of the singlet coupling matrix. The proposed methodology improves design flexibility, allowing for filters with both cascaded singlet and cascaded triplet topologies. This is demonstrated through two illustrative designs. The first design employs propagating RW sections as phase shifters to interconnect TE 201 singlets, which is the most common approach. The second design, however, replaces these sections with TE 101 resonant cavities. This substitution offers advantages in spuriousfree range, volume, and mass. Both prototypes were successfully manufactured and measured to validate the approach experimentally. In addition, the proposed method simplifies electromagnetic simulations by focusing on the irises at a single frequency point, thus avoiding the need to analyze entire singlet structures over full frequency sweeps.

Given a picture classified as a Persian cat by AI, users may ask questions such as, "What are the contributions of the eyes and ears to the classification result?" or "Which features contribute the most?". Existing computational XAI methods have achieved success in explaining the AI output, but they cannot directly help users find the answers to such questions. In this paper, we propose a new approach that addresses the challenge by visually exploring XAI explanation through intuitive features. Our method computes the contributions of predefined semantic features (e.g., eyes, ears, body) to individual images and represents them with a quantitative representation. This approach provides an easily interpretable explanation of model classification and enables convenient visual exploration over multiple images. We also have developed a visualization prototype that allows users to efficiently explore, filter, and compare image groups based on the quantitative representation of the semantic features. We demonstrate the effectiveness of this approach in providing an intuitive and scalable way to interpret XAI results through a set of example scenarios and expert feedback.

We present BALDUR, a novel Bayesian algorithm designed to deal with multi-modal datasets and small sample sizes in high-dimensional settings while providing explainable solutions. To do so, the proposed model combines within a common latent space the different data views to extract the relevant information to solve the classification task and prune out the irrelevant/redundant features/data views. Furthermore, to provide generalizable solutions in small sample size scenarios, BALDUR efficiently integrates dual kernels over the views with a small sample-to-feature ratio. Finally, its linear nature ensures the explainability of the model outcomes, allowing its use for biomarker identification. This model was tested over two different neurodegeneration datasets, outperforming the state-of-the-art models and detecting features aligned with markers already described in the scientific literature.

Fleets of autonomous vehicles offer an innovative solution to enhance logistic operations in manufacturing systems by increasing efficiency and safety. However, navigating in unstructured environments with moving obstacles presents significant challenges. This paper addresses this problem by proposing a novel control architecture that combines grid based mapping with receding horizon control techniques. Specifically, unicycle models are used to describe the dynamics of the robots, and sequences of feasible and safe way-points are computed by leveraging the real-time grid map of the entire operating scenario. A distributed set-theoretic model predictive control scheme leverages this information to ensure safe navigation. The effectiveness of the proposed approach is demonstrated through experimental results in a real-world context, where a team of three autonomous robots assists human workers along a steel production line manufacturing angle pipes, block-stages, and caps.

Quantitative capillary-free zone (CFZ) analysis in optical coherence tomography angiography (OCTA) holds promise for the early detection of retinal diseases. However, its clinical deployment is hindered by time-consuming manual segmentation procedures for periarterial and perivenous CFZs, which often focus solely on large vessels. In this study, we introduce deep learning for the automated segmentation of periarterial and perivenous CFZs in OCTA. Both convolutional neural networks (CNNs) and vision transformers (ViTs) were evaluated for automated CFZ segmentation. Nine quantitative features were derived to characterize CFZ changes associated with diabetic retinopathy (DR). Quantitative analysis revealed significant changes in CFZ ratios, counts, and mean sizes that can reliably differentiate control subjects, diabetic patients without DR (NoDR), and those with mild DR, underscoring their potential as sensitive biomarkers for early disease detection, progression monitoring, and treatment assessment.

Vehicular Knowledge Networking (VKN) is a paradigm where vehicles exchange knowledge instead of data. Named Data Networking (NDN) is a resilient architecture for sharing data between vehicles without information about the hosting vehicle. NDN might not be adapted for sharing knowledge, as the peculiar complexity of knowledge requires knowledge-driven NDN functions as well as a specific simulation environment enabling joint knowledge perception, inference and reasoning. In this paper, we present an open-source cosimulation framework connecting NDN, traffic, dynamic control and perception simulators for knowledge perception & inference, and with an AI-as-a-Service (AIaaS) platform for knowledge reasoning. This paper notably describes new interfaces and functions between the NDN daemon and the AIaaS micro-services for handling interest naming, caching and forwarding between knowledge producers and consumers. We demonstrate the benefit of the proposed framework and knowledge-specific extensions through an AI-driven vehicular intersection management.

Course evaluation is a critical component in higher education pedagogy. It not only serves to identify limitations in existing course designs and provide a basis for curricular innovation, but also to offer quantitative insights for university administrative decision-making. Traditional evaluation methods, primarily comprising student surveys, instructor self-assessments, and expert reviews, often encounter challenges, including inherent subjectivity, feedback delays, inefficiencies, and limitations in addressing innovative teaching approaches. Recent advancements in large language models (LLMs) within artificial intelligence (AI) present promising new avenues for enhancing course evaluation processes. This study explores the application of LLMs in automated course evaluation from multiple perspectives and conducts rigorous experiments across 100 courses at a major university in China. The findings indicate that: (1) LLMs can be an effective tool for course evaluation; (2) their effectiveness is contingent upon appropriate fine-tuning and prompt engineering; and (3) LLM-generated evaluation results demonstrate a notable level of rationality and interpretability.

This paper focuses on Ambimorphic prototype research in the field of information engineering. Firstly, it elaborates the research background, purpose and significance, and analyzes the concept of Ambimorphic prototype and related theoretical foundations. It then introduces in detail the application of public key-private key simulation in information security and communication systems, as well as the design and implementation of telephone semantic tools, including requirements analysis, architecture design and key technologies. Through experiments, performance tests and result analyses are conducted on both, including individual tests and comprehensive application tests. Finally, it summarizes the research results, limitations, and looks forward to future research directions, providing new ideas and methods for the development of information engineering.

Developing channel models typically requires aggregating channel measurements and the corresponding extracted propagation parameters from different research institutions to form a sufficiently large data basis. However, uncertainties arising from limitations of the sounding hardware and algorithms may greatly impact the comparability between sounding results. Especially, (sub-) THz channel sounding is challenged by a potentially low SNR. At the same time, high Doppler shifts may occur due to the high carrier frequencies, limiting the time spans for coherent or incoherent data processing. Hence, the channel dynamics additionally restrict the processing gain for the SNR. In this paper, we address these challenges metrologically from several perspectives. First, we discuss methods of baseband waveform precoding to adapt it to the sounder hardware or the employed high-resolution estimation algorithms. This allows using available transmit power in an optimal sense. Second, the assessment of the sounder performance requires a traceable reference allowing tracing back measurements (or estimated propagation parameters) to a physical ground truth. Therefore, we propose and discuss an over-the-air artifact allowing a joint verification of delay and Doppler parameters in a multipath scenario. The evaluations of exemplary sub-THz measurements with a multicarrier-based sounder highlight the strong interplay between sounder hardware and estimation algorithms, especially when coping with the mutual interference of parameters from multiple propagation paths. Hence, a metrological assessment always requires considering the full processing pipeline from the unprocessed measurements up to the extracted propagation parameters.

The rapid evolution and convergence of wireless technologies and vertical applications have fundamentally reshaped our lifestyles and industries. Future wireless networks, especially 6G, are poised to support a wide range of applications enabled by heterogeneous services, leveraging both traditional connectivity-centric functions and emerging beyond-communication capabilities, particularly localization, sensing, and synchronization. However, integrating these new capabilities into a unified 6G paradigm presents significant challenges. This article provides an in-depth analysis of these technical challenges for integrative 6G design and proposes three strategies for concurrent heterogeneous service provisioning, with the aggregated goal of maximizing integration gains while minimizing service provisioning overhead. First, we adopt multi-dimensional multiple access (MDMA) as an inclusive enabling platform to flexibly integrate various capabilities by shared access to multidimensional radio resources. Next, we propose value-oriented heterogeneous service provisioning to maximize the integration gain through situation-aware MDMA. To enhance scalability, we optimize control and user planes by eliminating redundant control information and enabling service-oriented prioritization. Finally, we evaluate the proposed framework with a case study on integrated synchronization and communication, demonstrating its potential for concurrent heterogeneous service provisioning.

The rise of eHealth technologies has transformed cardiac disease diagnosis, leveraging edge computing, AI, and IoT to offer critical insights into heart health. Data privacy constraints in centralized systems hinder access to large-scale ECG datasets, posing challenges for early diagnosis. While advancements in quantization and compression enable neural networks to run on edge devices, effective solutions for efficient training and inference on constrained devices are limited. To address challenges to training on the edge, we propose a mix-precision quantized DNN FPGA accelerator designed for multi-class cardiac diagnosis. Our solution achieves a top-1 test accuracy of up to 93.26% while enhancing computational efficiency, optimizing resource usage, and reducing transmission power. Our Mix-Precision Quantized FPGA Accelerator achieves up to 136x and 7.2x faster inference compared to state-of-theart Split-CNN and DCNN-Convolutional FPGA Accelerators, respectively. The accelerators offer a throughput of up to 1439.36 samples per second, latency of only 695µs, and programming logic power consumption staying below 600mW. Using hardwaresoftware co-design, our FPGA-based "Training on the Edge" approach combines software flexibility with hardware speed and improved diagnostic Top-1 test accuracy by up to 2.8% within just five training cycles, making the model more diverse to the dataset. The proposed approach also accelerates the development and reduces hardware rebuild time by a factor of (Training Cycles-1)x, ensuring efficient, sustainable ML solutions on edge devices. The source code is made available on https://github.com/shakeelakram00/Continual-Learningon-FPGAs-using-FINN

This short paper describes the distributed recognition of targets. The sensor data fusion of features arising from multiple sensors is considered for the purpose of target recognition/classification. This is performed in a scenario wherein the underlying distributions are not Gaussian (i.e., the distributions do not obey Normality). Furthermore, there is 'correlation' between the separate sensor features. The separate (sensor) features are not statistically independent. The data fusion procedure pursued here does not find itself in the object identification sensor data fusion paradigm. It is an intermediate step between the two levels of data fusion for target recognition. In a departure from the (class) conditional assumptions typically made, another factorization of the joint conditional distribution is evaluated. This factorization requires the conditioning on previous feature vectors. A novel adaptive procedure is suggested to address that alternate factorization. A non-standard nonparametric classification procedure is detailed in providing the classification results. The classification/recognition results are for multiple classes. Results are compared against the centralized method and the statistically independent method.

Polyspectral models are a class of nonlinear behavioral models that were applied to power amplifiers over 20 years ago but have not seen subsequent development. Polyspectral models have a multiple-branch topology that is a special case of the Volterra series. They provide closed-form, optimum solutions to their constituting filters. Here we bring the model construction measurements up to date using modern test equipment. The two-branch version presented here is simple to implement and provides good accuracy and computational efficiency for system simulations. We present results for a GaN amplifier subject to orthogonal frequency-division multiplexing for signals of bandwidth up to 200 MHz.