Early detection and identification of risk markers is extremely important in managing certain neurodegenerative diseases. Identifying risk-genes is an important research trend in this regard, along with the field of imaging genomics-to understand the relationship between brain activity and genetics [3] Drawing from these themes, this paper explores the possibility of a connection between individuals' resting brain activity & the presence of the G allele of the Phosphatidylinositol-binding Clathrin Assembly Protein (PICALM) gene, towards early detection of Alzheimer's Disease. The G allele is a known risk factor associated with Alzheimer's Disease [AD] progression. To measure resting brain activity, I used the beta frequency values of the electroencephalogram (EEG) readings of sample patients. A dataset of 79 patients' EEG recordings and PICALM allele data from OpenNEURO was used in this research. From the EEG recordings, each patients' beta frequency attributes were determined using spectral analysis, using Minimum Norm Estimation (MNE) and other related computational libraries. Subsequently, I explored statistical correlation between beta wave characteristics from frontal channels and the presence of PICALM G allele, and observed strong positive correlation between the presence of the G allele and peak resting beta frequencies, specially in certain localized brain areas. This indicates the possibility of developing easily accessible risk screening for AD in the future, using standard EEG scans (e.g. 10-20 channels-as opposed to the more expensive 120/240 channel set ups) on resting patients (i.e. without any need for cognitive tests). I want to build on my findings and explore similar correlations between AD and non AD patients and also conduct research on other medical markers as well (i.e. in addition to the presence / absence of G allele). The result, approach and resources used in this paper will also be useful for new researchers wishing to explore EEG analysis in general and low cost AD screening, in particular.

This paper presents a comprehensive wideband indoor propagation characterization via real-world experiments at 105.8 GHz with a 4 GHz bandwidth in a lecture hall environment. Utilizing a double-directional channel sounding system with rotating horn antennas, we measure power angular delay profiles (PADPs) at multiple receiver (RX) positions. Our study derives a path loss model, delay characteristics, and angle characteristics, including path loss exponent, root mean square (RMS) delay spread (DS), K-factor, RMS azimuth angle spread of arrival (ASA), and RMS azimuth angle spread of departure (ASD). We compared these results with the 3GPP indoor hotspot (InH) office model (up to 100 GHz), highlighting both similarities and differences in propagation characteristics. Notably, the path loss exponent was found to be 1.38, which is lower than the 1.73 observed in the 3GPP InH-Office LoS model. Moreover, there were positive correlations between the above spread values and transmitter (TX)-RX distance, and negative correlations between K-factor and TX-RX distance, which were not considered in the 3GPP InH-Office LoS model. Furthermore, our analysis revealed a strong negative correlation in RMS ASA vs K-factor and RMS ASD vs K-factor, which are not accounted for in the current 3GPP InH office model. These findings emphasize the need to consider such correlations in accurate propagation models in lecture hall scenarios at the 105 GHz sub-THz band.

The modular approach to ledger technology (MLT) offers a fresh viewpoint and method for putting distributed ledger systems into practice. Aiming to address the rigidity and limitations of traditional blockchains, MLT has a modular structure, in contrast to conventional blockchains, which are frequently thought of as monolithic. This makes it easier to modify the ledger to meet certain needs without having to completely overhaul the system. The intrinsic ability of MLT to evolve is one of its most notable advantages. Drawn from extensive literature reviews and various case studies, it's evident that as financial systems have grown increasingly complex, scalable and adaptable solutions are now essential more than ever. MLTs allow for the independent addition, removal, and modification of components. The usage of MLT to tackle new issues or adapt to shifts in the financial landscape is accentuated by their versatility. The capacity for interoperation is another significant benefit. The modern global financial system consists of numerous organizations and platforms that necessitate seamless communication. Because of its modular construction, MLTs may be able to incorporate components adhering to diverse standards or protocols, facilitating more efficient interactions between platforms. In conclusion, MLT not only offers a breakthrough approach to distributed ledgers but also signifies a transformative tool in addressing the modern challenges of the financial ecosystem.

Design-oriented modelling of switched reluctance machines is gaining more and more attention as it provides an attractive solution to the considerable computational burden associated to preliminary design stages. This work proposes a novel analytical model to determine the unsaturated inductance values, particularly in non-overlap conditions, based on a twofold improvement of the permeance method: the use of elliptic flux tubes, and the ability to self-tailor the flux tube shapes to any machine geometry. Finally, the accuracy of the proposed model is proven against finite element analysis applied to four designs, along with experimental results of a physical prototype. Selftailoring is also assessed and its implementation in an automated design routine is analysed.

This paper presents a cost-effective and scalable hybrid methodology for evaluating retrieval-augmented generation (RAG) systems using specialized pretrained models and advanced metrics, designed for critical domains like healthcare and finance. LLM judge-based evaluation approaches are hindered by significant score inconsistencies across identical input runs (up to 35% variations) and high computational costs while traditional NLP approaches, overly reliant on entity or phrase matching, lack a multi-faceted perspective and fail to capture deeper semantic understanding. Our methodology addresses these challenges with a novel approach that combines an ensemble of fine-tuned pretrained models and advanced NLP techniques, ensuring consistent, reproducible evaluations across multiple dimensions, while being tailored for offline scenarios to eliminate reliance on internet-connected systems or proprietary LLMs, offering a cost-effective solution with high accuracy in assessing semantic relevance, factual correctness, and context adherence. To enhance reliability, the approach employs a robust weighted scoring system using harmonic means combined with PCA, adaptive, and entropy-weighting techniques for trustworthy and consistent evaluation enabling seamless integration with existing systems, continuous metric adaptation and domain-specific customization, ideal for high-stakes applications that demand rigorous quality assessments without relying on external APIs.

This paper explores the need to implement the use of AI in Educational Institutions through a personal journey as a student in the School of Advanced Technology at Algonquin College during the appearance of ChatGPT in November 2022. The work done in the following months as President of the Student's Association to remove some of the fears and unsupported prejudices that developed into restrictive practices for the use of AI within College. This paper argues for the necessity of policy-driven approaches over mere guidelines to harness AI's potential while mitigating risks. Emphasizing the importance of ethical considerations, the paper advocates for a balanced view of AI as a complementary tool rather than a threat, urging educational institutions to shift from traditional control methods to more adaptive and inclusive strategies. Looking for a human-centered approach to AI, this paper intends to describe the latest technological trend from a social sciences perspective.

In this paper, we extend our previously proposed Adaptive Continuous Adversarial Training (ACAT) method beyond the problem-space of SPAM filters to encompass the feature-space of Machine Learning (ML)-based Network Intrusion Detection Systems (NIDS). ACAT continuously improves model robustness by incorporating adversarial samples detected from the ML-NIDS input network traffic into the training process. Problem-space evasion attacks rely on inverse-feature mapping, where modifications to real-world objects result in targeted perturbations within the feature vector after feature extraction. Therefore, protecting ML systems against feature-space attacks inherently provides defense against both feature-space and problem-space evasion attacks. Our results demonstrate that ACAT significantly reduces adversarial sample detection time compared to traditional techniques. Moreover, the performance of the ML-based NIDS under attack improved dramatically, with accuracy increasing from 50.92% to over 99% after only three retraining sessions.

SAINT (Space-Time Analysis with Implicit Neutron Transport) is a newly developed general computational framework for high-fidelity, multigroup space-time kinetics analyses. It is based on using an implicit time-discretization scheme to reduce the time-dependent neutron transport equation to a series of time-independent, fixed-source transport problems that can be solved directly by any static neutron transport solver without the need for code modifications. SAINT can thus be used with any angle and spatial discretization approach. Results presented here use the collision-probability method as implemented by the transport code DRAGON. To demonstrate its functionality and to verify its results, SAINT is applied to the two-dimensional cases of Phase I, the kinetics phase, of the OECD/NEA C5G7-TD benchmark. The test cases consist of control-rod insertions and moderator voiding in four fuel assemblies of a miniature light-water reactor with each assembly consisting of a 37×37 arrangement of fuel pins. The transient total core fission rates for the various test cases are found to be within +2% of the values available from the MPACT transport code. During maximum rod insertion and moderator voiding, the distribution in the fission rates of the individual fuel pins (normalized to steady state) are found to vary by as much as 18% from their steady-state values for the rod-insertion cases and by 28% for the moderator-voiding cases.

Ultra-wideband (UWB)-based localization systems offer a cost-effective solution with relatively accurate performance. However, in a distributed ROS 2 multi-robot system, sudden failures of localization-related programs such as the UWB ranging error mitigation program, can degrade overall system accuracy. While Kubernetes provides automated deployment capabilities, its performance with ROS 2 remains unevaluated within a specific robotic application. This letter addresses this gap by integrating Kubernetes with ROS 2 in a UWB-based multi-robot localization system. An edge cluster composed of five Jetson Nano and one laptop is orchestrated with Kubernetes. The node of Jetson Nano is set on each robot. We then deploy LSTM-based UWB ranging error mitigation programs on edge nodes. By inducing failures in various combinations of LSTM programs, we comprehensively assess the system's resilience and robustness through position error evaluation under different failure scenarios.

This paper presents the design, modeling, and feasibility study of a magnetic resonance (MR)-conditional, steerable neurosurgical robot for the intracranial delivery of therapeutic agents. Immunotherapy is an emerging technique for brain tumor treatment but faces challenges due to low cell trafficking with systemic infusions, particularly in the case of large tumors. To address this limitation, we have developed a novel robotic system capable of delivering therapeutic agents throughout a brain tumor. The robot comprises a straight, rigid outer tube and a flexible inner tube that can navigate along curved trajectories in multiple directions. Integrated into the system is a customdesigned injection mechanism, which consists of a motorized syringe, a long hydraulic transmission tube, and local syringes. This integration with the robot's nonmagnetic actuation system enables precise navigation to various locations within the tumor and targeted delivery of therapeutic agents at each site, thus maximizing their trafficking in the tumor. The robotic system is designed to be MR-conditional, enabling the use of intraoperative MR imaging to guide the delivery process and ensure accurate and sufficient delivery of therapeutic agents. Along with the design and manufacturing, we also present the modeling and characterization of the robot's motion capability, evaluation of open-loop motion control, and assessment of MR-compatibility. Additionally, we demonstrate therapeutic delivery in phantom models, collectively validating the feasibility of robot-assisted, MRI-guided therapeutic delivery.

This paper proposes PixelPrune, an approach to address two primary challenges in artificial intelligence of things (AIoT) vision systems: (1) the energy-intensive analog-to-digital converters (ADCs) required in the sensing unit for converting analog pixel arrays to digital tensors, and (2) the high data transfers between the sensing unit and computing unit. Our proposed solution involves the implementation of an in-sensor binary segmentation model on analog memristive crossbars to identify the important pixels and prune out the background information. Additionally, we propose a data transfer scheme that adaptively selects between dense and sparse data transfer formats based on the sparsity ratio measured from the segmentation mask obtained by the segmentation model. Our results demonstrate that the proposed object detection system achieves significant energy savings along with a considerable up to 95% reduction in data transfer, all while maintaining high accuracy.

This research introduces a reconfigurable intelligent surface (RIS)-assisted UAV service delivery system designed to enhance the network quality-of-experience (QoE) of remote users (RUs) by improving signal propagation and mitigating channel uncertainty. The RIS-enabled UAVs are deployed to ensure robust connectivity between RUs and the network infrastructure service provider (NISP). The objective is to maximize RU communication rates while satisfying quality of service (QoS) requirements. This optimization problem is formulated as a mixed-integer nonlinear programming (MILP) challenge, decomposed into channel allocation and RIS phase shift optimization sub-problems. These sub-problems are solved iteratively using matching game and whale optimization algorithms. Simulation results demonstrate that the proposed system significantly outperforms benchmark schemes in improving communication rates.

Automatic and accurate classification of lesions in magnetic resonance imaging (MRI) are essential in early diagnosis and postoperative assessment of human brain tumor diseases. Clinically, neurologists and oncologists primarily diagnose brain tumor lesions within MRI images by scrutinizing their locations, signal characteristics, edge conditions, blood supply, and interactions with surrounding tissues. In this paper, we propose a novel joint-attention convolutional neural network (JACNN) tailored for brain tumor classification, in which tumor lesions within MRI images are extracted and incorporated within an efficient neural network to achieve superior classification accuracy. Specifically, we first introduce joint attention module (JAM) to generate fused feature map from the MRI image. This module is then seamlessly integrated into a concise streamlined classification network called BaseNet to form the proposed JACNN. Guided by the JAM, the proposed approach can leverage the information combined from global channel and local spatial region to further accelerate the training process and improve the brain tumor classification. Qualitative and quantitative experimental results compared with several state-of-the-art methods demonstrate the effectiveness and efficiency of proposed JACNN method in MRI brain tumor classification. The outcomes of this study will further facilitate the practical application of AI-driven MRI brain tumor classification.

Over the last decade, digital twin technology has gained increasing interest in various fields, such as industrial engineering and manufacturing, healthcare, transportation, energy, logistics, or city operations. While digital twins of single assets dominated in the beginning, we are seeing more and more digital twins for systems of assets, where digital twins need to collaborate with other digital twins to accomplish their objectives. As in service-oriented systems, digital twins provide services that can be used by other twins. While in service-oriented systems a service is typically identified by type information, this is not sufficient in digital twin systems, where there may be a large number of service instances of the same type that are not interchangeable. In this paper, we present a proximity model that allows digital twins to specify the required services by so-called neighborhoods, whose definition is based on spatial relationships between client and service instances. This concept not only simplifies service discovery in digital twin systems, but also allows for efficient implementations. To support this, we illustrate how service discovery based on the proposed proximity model can be implemented on top of a spatial DBMS.

Graph neural networks (GNN) have become a powerful framework for analyzing structured data in the form of graphs, with applications spanning diverse fields such as social networks, biology, recommender systems, etc. This survey explores methodology, development, and advances in GNN architectures. We methodically survey the major classes of GNNs, including Convolutional GNNs (ConvGNNs), Spatial-Temporal Graph Neural Networks (STGNNs), Recurrent-based GNNs (RecGNNs), and Graph Autoencoders (GAEs). Every model is discussed in terms of underlying mathematical formulations, design principles, and practical applications. This survey goals to supply a comprehensive conception of GNNs for practitioners and researchers alike, highlighting their versatility and potential for future innovations in graph neural networks. We will furthermore discuss applications of graph neural networks across different fields and epitomize open-source codes, benchmark datasets, and model valuation for graph neural networks. In the end, this survey specifies existing challenges in interpretability, generalization, and scalability and proposes possible future research trends to further promote the performance of GNNs across various graph-based learning missions.

A document by Suman Kundu. Click on the document to view its contents.

Orbital Angular Momentum (OAM)-based Free-Space Optical (FSO) communication systems offer immense potential for high-capacity, secure links, but their performance is highly sensitive to pointing errors. This paper provides a comprehensive analysis of OAM systems under pointing inaccuracies for both short-range terrestrial links and long-range intersatellite communications. For short links, we demonstrate the trade-offs between increasing modulation order and the number of OAM modes, highlighting the potential of low-complexity two-stage detectors to mitigate computational overhead while maintaining robust performance. For inter-satellite links, where severe pointing errors dominate, we propose an innovative asymmetric mode design and optimal beam waist adjustment to enhance robustness against angular misalignments. Simulation results reveal critical insights into the interplay between pointing error intensity, mode selection, and system performance, offering practical guidelines for OAM system design in diverse scenarios. This work bridges the gap between theoretical modeling and real-world implementation, paving the way for reliable OAMbased communication in next-generation terrestrial and space networks.

Progressions in technologies for renewable energy have rendered feasible fresh avenues for energy production that is sustainable. Nonetheless, melding renewable energy sources into pre-existing power grids encounters hurdles, owing to their sporadic nature. Machine learning methodologies present an auspicious remedy to enhance the assimilation of renewable energy sources and augment the efficiency and dependability of power infrastructures. Through the application of data-centric models, machine learning algorithms possess the capability to scrutinize extensive datasets to anticipate renewable energy production, streamline power generation timetables, and forecast prospective system malfunctions. This inquiry endeavors to probe the potential of utilizing machine learning to amplify the sustainability of renewable energy mechanisms. By exploiting artificial intelligence's capabilities, we can craft pioneering resolutions to tackle the intricacies linked with incorporating renewable energy sources into the grid, ultimately setting the stage for a future where energy is more sustainable.