We employ the Hopfield model as a simplified framework to explore both the memory deficits and the biochemical processes characteristic of Alzheimer's disease. By simulating neuronal death and synaptic degradation through increasing the number of stored patterns and introducing noise into the synaptic weights, we demonstrate hallmark symptoms of dementia, including memory loss, confusion, and delayed retrieval times. As the network's capacity is exceeded, retrieval errors increase, mirroring the cognitive confusion observed in Alzheimer's patients. Additionally, we simulate the impact of synaptic degradation by varying the sparsity of the weight matrix, showing impaired memory recall and reduced retrieval success as noise levels increase. Furthermore, we extend our model to connect memory loss with biochemical processes linked to Alzheimer's. By simulating the role of reduced insulin sensitivity over time, we show how it can trigger increased calcium influx into mitochondria, leading to misfolded proteins and the formation of amyloid plaques. These findings, modeled over time, suggest that both neuronal degradation and metabolic factors contribute to the progressive decline seen in Alzheimer's disease. Our work offers a computational framework for understanding the dual impact of synaptic and metabolic dysfunction in neurodegenerative diseases.

Understanding the limitations of Multi-Agent Systems (MAS) is crucial for advancing their development and application. This paper explores the key technical, functional, ethical, and practical constraints of MAS, highlighting their impact on various applications and suggesting potential areas for future research.

The rapid evolution of ransomware poses an increasing threat to information security, with malicious actors deploying advanced techniques to evade traditional detection systems. Introducing a method for ransomware detection that leverages Gaussian algorithms for classifying low-level file system I/O Request Packet (IRP) patterns offers a novel and statistically grounded approach to identify deviations in system behavior. The methodology captures abnormal IRP sequences that signify ransomware activity, outperforming conventional signaturebased and heuristic detection techniques, especially when dealing with zero-day ransomware variants. A controlled experimental environment demonstrated the effectiveness of this approach, achieving high detection rates and minimizing false positives. Additionally, the model's adaptability to dynamic ransomware behavior positions it as a robust solution for environments where early intervention is critical. Results indicate that the Gaussian algorithm-based model provides a scalable and efficient method for real-time detection of ransomware, significantly enhancing the capabilities of modern cybersecurity frameworks.

CubeSat systems have the potential to make significant contributions to multiple areas of astrophysics research, including detection of gravitational wave counterparts, Gamma Ray Bursts (GRBs) and GRB afterglow, X-ray and ultraviolet astrophysics, exoplanet studies, solar, ionospheric and space physics, and variable star astrophysics including pulsar detection, Active Galactic Nuclei (AGN) physics, transient, time domain, and multi-messenger astrophysics. Several of these areas of astrophysical research utilize the radio frequency (RF) spectrum between 10 MHz and 10 GHz. This spectrum accounts for significant opportunities for research with space-based CubeSat systems and networks. While other frequency ranges such as millimeter-wave bands are important for research, the 10 MHz to 10 GHz frequency bands are optimum for CubeSat applications. Given this wide frequency range, the question becomes how to incorporate this capability in small form factor CubeSats. Swarms or constellations of CubeSats can be configured to perform multi-aperture Very Long Baseline Interferometry (VLBI), which is a fundamental technique for radio astronomy. The applications of CubeSats for detection of radio frequency counterparts of high energy phenomena with VLBI will be the focus of this study. This study is divided into two sections. Section 1 is a high level overview of the basic concepts, architectures, and technologies of a CubeSat system utilized for radio frequency counterparts of high energy astrophysical phenomena. This section was independently published in the Journal of the Society of Amateur Radio Astronomers (JSARA) and can be used as a standalone document for introduction to the subject discussed herein. Section 2 is the detailed analysis and design of the specific topics introduced in Section 1.

Modular-Multilevel-Cascade Converters are the preferable choice for high-voltage and high-power applications. The price of simple scalability, modularity, and flexibility is paid by a large number of components, especially semiconductors and capacitors. Such components are prone to failure, so fault detection and diagnosis, fault-tolerant control concepts, and reliability analysis are research fields within this topology. This paper presents several fault-tolerant approaches for the class of Modular-Multilevel Converters under the severe failure of a whole cluster and compares them in detail. Furthermore, the different fault modes are emphasized, and operation recommendations are given. All approaches are extensively derived and verified by simulation and experimental results in a steady-state. Additionally, experimental results of transients, i.e., the transition from normal-to fault-operation and power steps, are provided for the most suitable proposed approach. The results show the proper functionality of the control strategies. The simple implementation enables easy integration into (existing) systems and thus provides better fault-tolerance.

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.

Magnetic flux tubes and magnetic monopoles are interconnected concepts in theoretical physics, particularly in the concept of gauge theories and Superconductivity.This paper discusses this concept from the view of superconductivity.

The real-time knowledge of the junction temperature of power devices is an essential asset to improve the power density of modern power converters. In fact, temperature is the main stressor in term of device reliability, and its accurate measurement enables dynamic derating and overload without any lifetime penalties. On-state resistance is a renown temperaturesensitive electric parameters for MOSFETs, which can provide reliable information, provided that an accurate characterization procedure is undertaken, for each particular device used. In this paper, we propose four possible ways to improve the accuracy and speed of this characterization procedure, especially when done in an in-circuit setup (as opposed to a laboratory one). These improvements pertain to: consideration of self heating also for short-time pulses, compensation of the nonidealities of the conditioning circuits, changes in the cooling system to reduce the duration of the procedure while improving the accuracy of the reference temperature sensor, adoption of simplified models for the fitting of the experimental data. The proposed solutions are evaluated using a custom active gate driver; final design guidelines for the improvement of the procedure are eventually presented, demonstrating the possibility to simplify the characterization procedure with respect to the state of the art, still retaining a high accuracy.

Language models have transformed automated text generation by enabling systems to generate coherent, contextually relevant content across diverse domains. However, when faced with domain-specific tasks that require the precise use of specialized terminologies, these models frequently encounter challenges in tokenization and knowledge retrieval. The novel combination of token factorization and retrieval-augmented generation (RAG) presents a significant advancement, offering a more accurate and contextually informed generative process. Token factorization improves the representation of complex terms through the decomposition of tokens into smaller, semantically meaningful units, which significantly enhances the model's ability to generate domain-specific text with precision. Simultaneously, the integration of RAG allows the system to access and incorporate external domain-specific knowledge dynamically, ensuring that the generated content remains factually accurate and relevant. Through extensive experimentation, the research demonstrates substantial reductions in error rates, improvements in retrieval accuracy, and increases in fluency when these techniques are applied to specialized domains such as legal, medical, and technical fields. Although the enhancements introduce additional computational complexity, the overall improvements in performance justify the trade-offs. The findings contribute to the broader application of language models in domains requiring high accuracy and contextual depth, paving the way for more effective text generation systems in specialized industries.

The rapid advancement of artificial intelligence (AI) in medical imaging has transformed the landscape of diagnostic practices, particularly in the classification of chest X-ray (CXR) images. This paper explores the application of deep learning techniques, specifically convolutional neural networks (CNNs) and transfer learning, in enhancing the accuracy and efficiency of CXR analysis. We address the challenges posed by limited labeled data, a common issue in medical imaging, and demonstrate how transfer learning can leverage pre-trained models to improve classification outcomes. Our study reviews various methodologies, focusing on Residual Networks (ResNet) and innovative data augmentation strategies that enhance model robustness. The results indicate significant improvements in the detection of critical pathologies such as pneumonia, underscoring the potential of AI-assisted diagnostics in clinical settings. This work highlights the need for continued research to optimize these models and integrate them effectively into healthcare practices, ultimately contributing to better patient outcomes and more efficient use of radiological resources.

This paper presents WideConvNet, a novel convolutional neural network (ConvNet or CNN) that outperforms three established end-to-end models-EEGNet, Shallow Convolutional Neural Network (ShallowConvNet), and Deep Convolutional Neural Network (DeepConvNet)-in classifying event related potentials (ERPs) from electroencephalogram (EEG) signals. At the core of WideConvNet is an adapted inception module for 1D data, utilizing varying-size 1D filters inspired by the wavelet scattering transform, which are learned during the training phase to capture temporal patterns across different scales within the EEG signals. We evaluated the performance of WideConvNet on multichannel EEG data from 20 male subjects with alcoholic use disorder (AUD) and 20 healthy male controls, each providing 16 1.0-s ERPs. Using 10-fold leave-subjects-out cross-validation, WideConvNet demonstrated a significantly higher classification accuracy (𝒑 = 𝟎. 𝟎𝟎𝟒) at detecting AUD, and an area under the curve of the receiver operating characteristic (AUC-ROC, 𝒑 = 𝟎. 𝟎𝟎𝟒), compared to EEGNet and ShallowConvNet. WideConvNet also had a comparable performance to DeepConvNet but with 28% fewer parameters. These findings suggest that WideConvNet, specifically on short duration EEG signals, offers a balanced solution, combining robust classification performance with improved computational efficiency.

In recent years, Unmanned Aerial Vehicles (UAVs) have attracted a lot of attention due to their flexibility and high mobility. Nowadays, UAVs have been widely used in various fields, such as military fields and social fields. However, due to the increasingly complex environments faced by UAVs and the rising demands on UAV systems, traditional UAV control methods are no longer able to achieve efficient control of UAVs under multi-constraint situations. Reinforcement Learning (RL), as an emerging robot control technology, is well suited to the needs of UAV systems in terms of its ability to interact with and learn from the environment. Therefore RL-based UAV systems are gradually becoming a new trend in research. Nonetheless, as a new research field, it faces many challenges, such as high dimensional spaces and dynamic environments. It is necessary to provide a comprehensive overview and analysis of existing specific RL methods applied to UAV systems, and to understand the challenges as well as possible solutions for RL-based UAV systems. In this paper, we first provide a comprehensive overview and summary of the application of RL in different UAV application scenarios based on the classification of RL methods. After that, we systematically analyze the challenges and solutions faced by RL applied to UAV systems given the existing relevant literature. Finally, we discuss the potential research directions for RL-based UAV systems.

This study presents a physics-informed online learning method for modeling the flux linkages of synchronous machines (SMs). The approach trains neural networks (NNs) to learn the physical laws governing the flux linkages while adhering to the physical constraints inherent in the model. The learning rules are designed to satisfy the first-order optimality conditions with quadratic convergence. The proposed method can be used for both online state estimation and online model learning, ensuring fast convergence for local behavior and gradual, comprehensive convergence for global behavior, respectively.

This paper provides a timely snapshot of works that focus on using Unmanned Aerial Vehicle(s) (UAVs) for Integrated Sensing, Communication (ISAC) or/and Computation (ISACC). Specifically, it covers works where UAVs use their radio as a radar when communicating with users to collect information such as the range and velocity of targets. Example targets include obstacles, aerial/ground eavesdroppers or jammers, users or an area. Further, sensed data collected by UAVs may require computation, where data is either computed locally by a UAV or offloaded to another UAV or a ground base station. We classify prior works based on whether they have a (i) single, or (ii) multiple UAVs. Within category (i) and (ii), we further divide works according to whether they have a single or multiple users or/and targets. Further, we outline their research aim, performance metrics, optimization problem and key constraints. Lastly, we present a number of possible research directions.

In 2021, approximately 537 million people were diagnosed with diabetes mellitus. With rates expected to rise, health expenditures are projected to reach one trillion USD by 2030. Thus, measuring glucose levels is essential for rationalizing the costs of public health systems. In this context, this paper presents two major contributions. First, it demonstrates that using deionized water (DI-water) as a reference for glucose sensing is not a reliable approach for representing human blood plasma (BP), as it lacks ions and suppresses essential effects such as losses. As an alternative, we investigate the use of an artificial blood plasma solution (ABPS) that closely resembles real human BP. Characterized over a range from 500 MHz to 10 GHz, ABPS shows marginal differences in real permittivity but significant differences in imaginary permittivity compared to DIwater. The second contribution is the design of a highly sensitive microwave resonator sensor based on concentric double circular split ring resonator (DCCSRR) on a ROGERS 5880 TM substrate. This sensor can differentiate glucose concentrations from 0 to 400 mg/dL, exceeding the relevant range for diabetic individuals (50-300 mg/dL). The DCCSRR operates at 2.48 GHz and can detect minimal concentration variations of 25 mg/dL in low concentrations, representing a significant advancement in the field. Differently from most sensitive approaches available to date, this structure operates in a non-licensed band and in a fully passive form, offering flexibility for implementation and low cost. These characteristics position it as a state-of-the-art solution in microwave glucose sensors. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.

The wettability of a surface is a critical parameter for heat transfer and has significant implications for the development of heat-exchanging surfaces, self-cleaning surfaces, electrical cooling components, etc. A commercial surface wettability measurement device determines static and dynamic contact angles but is highly expensive. The purpose of this study is to build and validate a low-cost, computer-controlled device that is in-house designed and capable of measuring contact angles with high accuracy. The correct analysis of the captured images frame by frame from the video is ensured by an open-source image processing program called ImageJ and a MATLAB image code for calculating contact angle. Static contact angle measurements were found to be 93° ± 2.7°, 85° ± 6.35°, 25.0° ± 6.7°, 88° ± 8.35°, and 87° ± 5° for aluminum, copper, glass, brass, and steel, respectively. The data from the open literature has been compared with the measured static liquid contact angle. The range of the deviance was from 1.93º to 4.97º. Copper and Brass have hysteresis contact angles that are larger than those of the other metals and nearly identical (~ 34º). On untreated glass, the minimum hysteresis contact angle was found to be approximately 21º.

A document by Rami Amiri. Click on the document to view its contents.

Cardiovascular diseases are the primary cause of death globally. With the prevalence of electrocardiogram (ECG) machines within and outside the clinical environment, it is now possible to passively monitor a patient's heartbeat for cardiovascular diseases. The goal of this work is to emphasize the importance of selfsupervised learning for arrhythmia detection, leveraging the large amounts of unlabelled data recently made publicly available and demonstrating significant performance improvements as it reduces overfitting to class imbalance and noise. We propose Masked Patch Modelling (MPM) and leverage 8.2 million unlabelled ECGs to perform large-scale self-supervised pre-training and create a foundational 1dimensional Transformer model, PatchECG, that can be fine-tuned for any downstream tasks involving ECG data. We obtain state-of-the-art results on standard benchmark datasets, including PTB-XL multi-label classification, while setting new benchmarks on the largest and highest quality multi-label classification dataset to date. We find that PatchECG outperforms the current state-of-the-art with regard to computational efficiency, requiring only 1/5 of the computational resources while increasing model capacity by a factor of 14. We also compare the 1-dimensional PatchECG model to a state-of-the-art 2-dimensional vision Transformer and observe significantly higher performance. Finally, we perform ablation studies to investigate other methods for addressing the critical issues incurred with automated arrhythmia detection, resulting in a performance improvement of more than 2% under conditions of class imbalance, label noise, and over-parameterization.