The broad adaptation of cloud computing in healthcare has deeply aroused serious concerns about privacy and security, particularly in managing sensitive patient data. This paper proposes a state-ofthe-art architecture that integrates cloud infrastructure with edge computing for secure healthcare monitoring using Wireless Body Area Networks. The proposed architecture based on an edge node ensures the preservation of privacy through anonymization techniques, which is achieved through the integrated ARX toolkit with advanced techniques including l-diversity, t-closeness, and δ presence. In our framework, a matrix-based access control system was introduced to manage permissions precisely. It also employed the Attribute-Based Signcryption for secure data transmission. The system will process the vital medical parameters through specialized functions at the edge nodes. It adopts a computation approach that balances efficiency in processing with privacy requirements in a very effective way. It integrates fuzzy logic into the architecture for diagnosis in medicine while considering compliance to healthcare regulations. These include a multi-layered mechanism for the preservation of privacy, a lean edge-based processing system for the parameters of medicine, and an integrated access control strategy. This work has contributed to the field of secure healthcare systems by showing how edge computing can improve both privacy and efficiency in the processing of medical data while imposing strict security standards at the cloud.

Traditional tokenization methods in language models often encounter challenges in efficiently capturing semantic complexities and managing computational resources. The introduction of Hierarchical Memory-Based Adaptive Tokenization (HMAT) addresses these limitations through a dynamic, contextaware approach that constructs a multi-level semantic hierarchy, enabling models to process information at varying degrees of abstraction. Empirical evaluations demonstrate that HMAT enhances tokenization efficiency, reduces computational overhead, and improves semantic coherence in generated outputs. These findings demonstrate the potential of HMAT to advance tokenization strategies within language models, offering a pathway toward more sophisticated and adaptable natural language processing systems.

The rapid evolution of language models has significantly enhanced their ability to generate human-like text; however, challenges persist in achieving complex syntactic and contextual understanding. Introducing Dynamic Syntax Transfer (DST), a novel methodology that dynamically adjusts syntactic structures in response to contextual variations, thereby enabling large language models to produce text with improved grammatical accuracy and contextual relevance. The implementation of DST within an open-source language model architecture has demonstrated notable advancements in syntactic precision and contextual coherence, as evidenced by comprehensive quantitative and qualitative analyses. These findings demonstrate DST's potential to advance context-aware syntax modeling, offering a promising avenue for future research and applications in sophisticated language generation systems.

Electromyogram (EMG)-informed neuromusculoskeletal (NMS) models can predict physiologically plausible muscle forces and joint moments. However, calibrating model parameters (e.g., optimal fiber length, tendon slack length) to the individual is time-consuming, with the optimization often requiring hours to converge and typically not accounting for unrecorded muscle excitations. This study addresses these limitations by incorporating differentiable physics and muscle synergies into the calibration of NMS models. Methods: We implemented an NMS model with auto-differentiable Hill-type muscles, enabling the use of adaptive gradient descent optimizers. Two types of calibration were evaluated: a standard EMG-driven approach and a synergy-hybrid approach that also synthesized unrecorded excitations. These methods were evaluated using upper and lower limb data, each from a single participant. Results: The calibration time was reduced by up to 26 times while maintaining comparable accuracy in moment predictions. Compared to the EMG-driven calibration, the synergy-hybrid calibration improved the estimates of model parameters for reduced number of EMG channels. Conclusion: Autodifferentiable Hill-type muscle models greatly reduce NMS model calibration time and enables the synthesis of unrecorded muscle excitations through muscle synergies, facilitating the calibration of all muscle parameters. Significance: This new rapid calibration could support deployment of NMS models in time-sensitive applications, including real-time biomechanical analyses and personalized neurorehabilitation.

We introduce a guard time-free approach for simultaneous channel estimation and equalization, addressing both inter-carrier interference (ICI) and inter-symbol interference (ISI) in multi-carrier communication systems. The proposed method utilizes two maximum a-posteriori (MAP) equalizers based on the BCJR algorithm, operating in the time-frequency domain. These equalizers share soft-bit data among themselves and with a channel decoder. We derive the a-posteriori probabilities (APPs) and show that the equalizers exploit Dopplertime diversity. We also derive maximum-likelihood (ML) bounds on the system's performance, which also exhibit diversity gains. We initially simulate on Rayleigh fading doubly-dispersive (DD) channels with perfect receiver channel state information (CSI). We compare with an existing iterative maximum likelihood equalizer (IMLE) from the literature and show that the proposed system outperforms it. Next, we focus on an underwater acoustic channel (UAC), and develop a joint channel estimationequalization architecture. To estimate this sparse channel, we take a compressed sensing (CS) approach using the orthogonal matching pursuit (OMP) algorithm together with an overcomplete dictionary set. Compared to 1D equalization with full guard time, the guard time-free estimation-equalization system achieves bit error rate (BER) reductions between 24% and 78% for high-spread channels that can surpass the orthogonal timefrequency space (OTFS) crystallization condition.

Cell tracking is crucial for understanding the complex patterns of cellular migrations that underlie many physiological, pathological, and therapeutic processes. Positron emission particle tracking (PEPT) is a method that uses list-mode positron emission tomography (PET) data to localize moving particles non-invasively inside opaque systems. However, while the application of this method to in vivo cell tracking has previously been evoked, its implementation has been limited to tracking one cell at a time. This study investigates the feasibility of tracking multiple cells simultaneously using a recently developed expectation maximization (EM) algorithm called PEPT-EM. The primary challenge to the translation of this algorithm towards biomedical applications is the low radioactivity of the cells being tracked. We experimentally demonstrated the performance of the PEPT-EM algorithm using a preclinical PET scanner for tracking <100 Bq droplets and cells, in phantoms and in vivo. We found that while background count and multiplexing effects impact static source tracking, sensitivity remains critical for dynamic tracking of moving sources. We successfully localized multiple single cells in vivo, moving at speeds up to 25 mm/s, marking the first use of PEPT-EM for such applications. Our findings highlight the exciting potential of PEPT for real-time, high throughput tracking of multiple single cells in vivo, paving the way for studying cell tracking in biological systems.

A document by Jun Ueda. Click on the document to view its contents.

The trend towards sensor miniaturization has heightened interest in optimizing sensor array configurations across various scientific and industrial applications that imply multichannel measurements. In this article, we present an extended version of our previous approach, adapting it to the realworld conditions in magnetoencephalography (MEG). At each step the method selects the sensor location that maximizes the ROI-related signal-to-noise ratio (SNR), followed by the projection recursively applied to the gain matrix to orthogonalize the subsequent stereotypic steps with respect to the source subspace served by the already selected sensors. This algorithm offers high computational efficiency and flexibility with respect to physical constraints. We dub our approach as Optimal Layout Design via Recursively Applied Leadfield Elimination (OLD RALFE). Our analysis of the optimized MEG sensor layouts obtained with OLD RALFE shows that the proposed algorithm performs similarly to several other significantly more computationally demanding sensor layout optimization approaches in terms of both SNR and the Cramér-Rao lower bound (CRLB) on the spatial resolution of the inverse problem solution. OLD RALFE demonstrated in application to MEG, is broadly applicable to many other multichannel measurement challenges.

The exponential rise in cyber threats has escalated the urgency for robust, automated defenses capable of identifying and mitigating sophisticated ransomware attacks. Current detection systems, predominantly reliant on static or rule-based methodologies, often fall short in recognizing rapidly evolving ransomware variants, necessitating adaptive solutions that address both accuracy and computational efficiency. In response, an innovative framework combining Adaptive Anomaly Clustering (AAC) with Threat Signature Prediction (TSP) is proposed to deliver a dynamically adaptive ransomware detection model. The AAC component isolates malicious activities through behavior-specific clustering, achieving high specificity in identifying ransomware, while TSP predicts potential ransomware signatures in real-time, allowing proactive defense mechanisms even for previously unseen ransomware. Together, AAC and TSP offer a comprehensive, layered detection approach that not only improves detection accuracy but also enhances response speed and scalability across diverse network conditions. Through rigorous evaluation, the framework demonstrates significant improvements over traditional methods in detection precision, false-positive reduction, and adaptability to novel ransomware threats, thus showing its potential for integration into advanced cybersecurity infrastructures.

In this article, we introduce a novel magnetometer suitable for millisecond-range magnetorelaxometry. The magnetometer exhibits high sensitivity to weak magnetic fields, approximately 40 pT / √ Hz. It is compact, operates at room temperature, and possesses a wide dynamic range. We also propose a regularized basis decomposition algorithm for processing multiexponential relaxation data. Our research includes a thorough investigation of the sensor's characteristics and feasibility for magnetorelaxometry, encompassing test tube experiments with magnetic micro-and nanoparticles, among other applications.

In this paper, bikeshare data from Chicago on weather-friendly days in 2019 and 2020 were analyzed to investigate the variation in bikeshare travel before and during the pandemic. Our results show that bikeshare trips during the pandemic were much longer than prior to the pandemic. The increased rate of bikeshare usage was unbalanced spatially and varied significantly for different user types. Bikeshare was used remarkably more by casual users than by subscribers, and the increase occurred much more in the outskirts of the city. The increase in bikeshare travel was associated with a reduction in travel by ridehailing and public transit, especially in the urban periphery. The correlation of bikeshare use with the bus system was much less significant than with the rail system. Bike lanes/facilities had a mixed effect on bikeshare travel. Weekend bike trips increased in areas where there was no bike lane. Weekday trips, on the contrary, increased in the vicinity of bike greenways.

Temporal perception, the ability to detect and track objects over time, is critical in autonomous driving for maintaining a comprehensive understanding of dynamic environments. However, this task is hindered by significant challenges, including incomplete perception caused by occluded objects and observational blind spots, which are common in singlevehicle perception systems. To address these issues, we introduce LET-VIC, a LiDAR-based End-to-End Tracking framework for Vehicle-Infrastructure Cooperation (VIC). LET-VIC leverages Vehicle-to-Everything (V2X) communication to enhance temporal perception by fusing spatial and temporal data from both vehicle and infrastructure sensors. First, it spatially integrates Bird's Eye View (BEV) features from vehicle-side and infrastructureside LiDAR data, creating a comprehensive view that mitigates occlusions and compensates for blind spots. Second, LET-VIC incorporates temporal context across frames, allowing the model to leverage historical data for enhanced tracking stability and accuracy. To further improve robustness, LET-VIC includes a Calibration Error Compensation (CEC) module to address sensor misalignments and ensure precise feature alignment. Experiments on the V2X-Seq-SPD dataset demonstrate that LET-VIC significantly outperforms baseline models, achieving at least a 13.7% improvement in mAP and a 13.1% improvement in AMOTA without considering communication delays. This work offers a practical solution and a new research direction for advancing temporal perception in autonomous driving through vehicle-infrastructure cooperation.

This paper investigates the probability distribution of the signal-to-interference noise ratio (SINR) for a 6G communication system comprising a multi-antenna transmitter, an intelligent reflecting surface (IRS) and a remote receiver station. A common assumption in the literature is that the density distribution function for SINR and signal-to-noise ratio (SNR) of an IRS-to-ground communication follows a Rayleigh and Rician distribution. This assumption is essential as it influences the derivation of the properties of the communication system such as the physical layer security models and the designs of IRS controller units. Therefore, in this paper, we present an analytical derivation for the density distribution functions of the SINR for an IRS-to-6G ground communication ameliorating the typical assumptions in the literature. We demonstrated that the SINR density function of an IRS-to-6G ground communication contains a hypergeometric function which defines the solution to a second-order linear ordinary differential equation (ODE) with 3 points. We further applied the derived density distribution function to determine the average secrecy rate for passive eavesdropping.

In recent years, much of the research on speech enhancement (SE) has focused on using deep neural network models to improve performance. While these models often excel on specific datasets, they tend to have high memory and computational demands. This makes it crucial to evaluate their performance across different datasets in real-world applications. To tackle this challenge, we propose a novel speech enhancement method that integrates a diffusion model with Actor-Critic techniques. Our approach involves processing noisy speech through a diffusion model and a pre-trained large-scale language model, which acts as the Actor to label the noisy input. These labels then serve as conditions for the SE model. The enhanced output is evaluated by the Critic, which scores the voice quality. During training, the SE model's enhanced outputs are reintroduced as noisy inputs for the Actor and SE models, fostering further improvements. We conducted experiments using two different sizes of the TIMIT and VCTK-DM datasets. In the test results, although CDiffSE-AC demonstrates limited score improvements within the same dataset, it significantly outperforms in crossdataset tests, even surpassing the baseline MOSE method on the TIMIT and VCTK-DM test sets. Specifically, CDiffSE-AC achieves PESQ improvements of 0.15 (7.2%) and 0.18 (7.5%), respectively.

Joint sensing and communication (JSC) research direction has been envisioned to be a key technology in 6G communications, which has the potential to equip the traditional base station with sensing capability by fully exploiting the existing wireless communication infrastructures. To further enhance wireless communication performance in JSC, sensing-assisted communication has attracted attention from both industry and academia. However, most existing papers focused on sensingassisted communication under line-of-sight (LoS) scenarios due to sensing limitations, where the sensing target and communication user remain the same. In this paper, we propose a general sensing-assisted channel estimation framework, where the channel estimation accuracy for both Los and non-line-of-sight (NLoS) are considered. Simulation results demonstrate that our proposed sensing-assisted communication framework achieves higher channel estimation accuracy and throughput compared to the traditional least-square (LS) estimation. More importantly, the feasibility of the proposed framework has been validated to solve the complex channel estimation challenge in distributed multiple input and multiple output (MIMO) network.

In low-or middle-income countries, road traffic safety becomes challenging because of the increasing number of vehicles, limited road infrastructure, and inability to capture a variety of traffic violations. Traffic Management and Information Control Centres (TMICCs) collect and analyse traffic data to improve road traffic safety. State-of-the-art technologies, such as the Internet of drones and computer vision, could help to automatically extract traffic data and provide a hands-free solution for road traffic monitoring. Hence, we propose a "UAV-based Urban Traffic Monitoring (U-UTM)" system to automatically extract traffic violations based on the movement of vehicles and traffic flow parameters, which helps to improve road traffic safety. Along with proposing a framework, this paper implemented three U-UTM operations-estimating Ground Sample Distance (GSD), detecting lanes using vehicles' tracks, and post-processing to reduce broken tracks. This paper derived GSD mapping from empirical data, improved it by 25.09% over theoretical GSD mapping, accurately detected lanes/zones, and reduced broken tracks of vehicles by 48.19%. Videos, data, and codes are available at https://github.com/ERYAGNIK003/U-UTM.

The widespread use of resource-constrained Internet of Things (IoT) devices has expanded the attack surface for networks, particularly making them more vulnerable to Distributed Denial of Service (DDoS) attacks. Due to weak security, these devices can be exploited to form botnets that overwhelm networks with traffic or consume network resources, disrupting legitimate operations. This paper presents DDoShield, a twostage defense framework designed for real-time detection and attribution of various types of DDoS attacks, including flooding attacks and those exploiting vulnerabilities in protocols. Leveraging advancements in programmable networks, DDoShield detects malicious DDoS traffic and classifies these flows at the network controller, allowing for timely and accurate updates to the data plane’s defense mechanisms. Using the CICIoT2023 dataset, we implement a lightweight Decision Tree-based model in the data plane, achieving 80-99% accuracy in detecting DDoS traffic. The classification of the malicious flows is offloaded to the control plane, where a ResNet classifier demonstrates over 96% accuracy for DDoS attacks driven by flooding and transport layer vulnerabilities, though slightly lower accuracy (75-83%) for attacks exploiting application layer vulnerabilities. Compared to traditional intrusion detection systems, DDoShield reduces malicious packet detection delay by 3 milliseconds and significantly lowers bandwidth overhead, with a 20% increase in switch memory usage as a trade-off.

Intensive rehabilitation is essential for stroke survivors, facilitating motor recovery, improving activitiesof-daily-life (ADL) performance, and overcoming limitations in traditional practices by offering accessible, effective, and tailored therapy options. The integration of wearable technologies and machine learning is advancing homebased rehabilitation. However, a significant research gap remains in simulating realistic home environments to evaluate ADL measurement techniques effectively. In this paper, we present a methodology for identifying three isolated arm movements (reaching, lifting, and pronation/supination) using data from 12 healthy participants and two different sensor configurations. The first configuration involved four Inertial Measurement Units (IMUs) on the dominant arm, while the second used a single IMU on the wrist. In addition to comparing sensor configurations, we evaluated the generalization of two arm movement identification models by training them on structured trials and testing on semistructured trial, involving fourteen kitchen-related activities, simulating a home-based rehabilitation scenario. We employed a Random Forest (RF) classifier and a new variant of the previously proposed hybrid deep learning model, combining convolutional neural network and recurrent neural network architectures. The RF classifier could generalize the prediction for activities in the semi-structured trial with 86.54% balanced accuracy, while the hybrid model reached 87.96% in identifying those activities. Performance declined when using a single wrist IMU, with the RF classifier showing a smaller decrease in accuracy (10.57%) compared to the hybrid model (26.22%). Our findings demonstrate that the generalization of key arm movement identification is accurate and robust, therefore, indicating the potential for future application in home-based stroke rehabilitation.