Grid operators in areas with lots of solar generation need solar power forecasts to support decision making. Most forecasts that are available are deterministic, and most of the ones that are probabilistic are for single plants, meaning don't have combined probabilities for aggregations of all the plants in a region. Additionally, few methods for producing probabilistic forecasts are easy to replicate or deploy for grid operators. This work presents a method for producing regional probabilistic forecasts from an existing deterministic forecast using open source tools and freely available data. The method is demonstrated using a deterministic forecast that is produced with similar tools and data. Reference code to replicate this work, which is freely available, is also introduced.

Previous research on jailbreak attacks has mainly focused on optimizing the adversarial snippet content injected into input prompts to expose LLM security vulnerabilities. A significant portion of this research focuses on developing more complex, less readable adversarial snippets that can achieve higher attack success rates. In contrast to this trend, our research investigates the impact of the adversarial snippet's position on the effectiveness of jailbreak attacks. We find that placing a simple and readable adversarial snippet at the beginning of the output effectively exposes LLM safety vulnerabilities, leading to much higher attack success rates than the input suffix attack or promptbased output jailbreaks. Precisely speaking, we discover that directly enforcing the user's target embedded output prefix is an effective method to expose LLMs' safety vulnerabilities.

Answer assessment is a critical aspect of various fields including education, artificial intelligence, and natural language processing. This paper aims to provide a comprehensive overview of answer assessment, exploring its methodologies, applications, challenges, and future directions. By examining the different approaches to evaluating responses, both human and automated, this paper highlights the significance of accurate and reliable answer assessment in enhancing learning outcomes, improving automated systems, and advancing research in related areas.

The goal of this study is to reduce rotation angles for faster scans in a gradient-coil-free portable MRI system, employing rotational spatial encoding magnetic fields (rSEM) with non-linear gradients. This is achieved by utilizing a proposed evaluation metric for encoding capability, thereby making optimization feasible. The local k-space-based evaluation metric for each sub-region of image is calculated by summing the area covered by the spokes in the local k-space when the SEM rotates; each spoke corresponds to one angle. This evaluation metric is expected to correlate closely with the image quality of the reconstructed images and is applied for angle selection to accelerate scans in rSEM systems. Greedy algorithm is used for the selection. Both simulations and experiments were conducted to validate the effectiveness of the evaluation method and the selected angle set. The results demonstrate that the proposed evaluation metric for the encoding capability of rSEM has correlation coefficient greater than 0.90 with image quality metrics of the corresponding reconstructed images. Using the angle set selected by greedy algorithm with proposed method, higher image quality with reduced angles is achieved. This further validates the accuracy of the proposed evaluation metric and underscores the necessity of rapid assessment for feasible optimization. A significant reduction of rotation angles is achieved by rotation angles selection using the proposed evaluation method. This facilitates fast scans in gradient-coil-free portable MRI systems with rotating SEM while maintaining high image quality. This advancement is crucial for the practical application of this type of portable MRI.

The detection of Pulmonary Hypertension (PH) from the computer analysis of digitized heart sounds is a low-cost and non-invasive solution for early PH detection and screening. We present an extensive cross-domain evaluation methodology with varying animals (humans and porcine animals) and varying auscultation technologies (phonocardiography and seisomocardiography) evaluated across four methods. We introduce PH-ELM, a resource efficient PH detection model based on the extreme learning machine that is smaller (300× fewer parameters), energy efficient (532× less watts of power), faster (36× faster to train, 44× faster at inference), and more accurate on outof-distribution testing (improves median accuracy by 0.07 area under the ROC curve (auROC)) in comparison to a previously best performing deep network. We make four observations from our analysis: (a) digital auscultation is a promising technology for the detection of pulmonary hypertension; (b) seismocardiography (SCG) signals and phonocardiography (PCG) signals are interchangeable to train PH detectors; (c) porcine heart sounds in the training data can be used to evaluate PH from human heart sounds (the PH-ELM model preserves 88 to 95% of the best indistribution baseline performance); (d) predictive performance of PH detection can be mostly preserved with as few as 10 heartbeats and capturing up to approximately 200 heartbeats per subject can improve performance. The results of this work may guide future methods for data collection, predictive modeling, and model evaluation towards the practical implementation of PH detectors.

This paper considers multiuser millimeter wave (mmWave) MIMO communication systems where both the base station (BS) and the user equipments (UEs) use analog beamforming (BF). We propose a beam acquisition approach to find the best directions for the narrow beams, required to achieve BS-UE links above a minimum signal-to-noise ratio (SNR) threshold. The multiple UEs first select their beams simultaneously, based on signals transmitted from the BS on wide training beams. The BS then switches to a beam acquisition mode in order to select the best beam direction for each UE, based on training signals the UEs send on the uplink. We propose a novel parallel beam acquisition scheme for this uplink phase. We also propose a user rejection policy, in which weak UEs that do not satisfy a target SNR threshold, equivalent to a certain bit error rate, are rejected. To do so, we derive the posterior probability that a BS beam has a resultant SNR below the target SNR threshold. We also propose a generalized likelihood ratio test (GLRT) to reduce the computational burden. We derive closed-form error probability expressions for the proposed GLRT approach, and examine the trade-off between the beam acquisition time and accuracy in selecting the beams. We show that the beam acquisition time of our proposed approach is as low as 3% of existing approaches.

Currently, the 5th generation mobile networks, known as 5G, have been standardized by the 3rd generation partnership project (3GPP) and are already being used commercially. The system is characterized by its flexibility and heterogeneity in mobile cellular communication and networking. Simultaneously, the issue of 5G security has emerged swiftly with the utilization of cloud and non-3GPP networks in the 5G communications network. Extensive research has been carried out on system vulnerabilities and attacks, but there is a relatively new and understudied area known as handover vulnerabilities and their security solutions. Handover refers to the process of transferring connections from one network to another or from a base station to another within the same network. It is important to ensure that any proposed solution for handover vulnerabilities in 5G networks maintains the key characteristics of low latency and network availability. Several security solutions have been suggested, but they involve a trade-off between the additional communication overhead they introduce to the system and the system's requirement for low latency. Our study proposes a new and comprehensive model leveraging the unique physical layer characteristics of the users to enhance the security of the Xn-based handover in 5G networks. We have validated our solution using Scyther tools, which demonstrate that our approach ensures secure communication during the handover process without introducing any additional overhead to the existing system. Furthermore, our solution does not introduce any complexity that could negatively impact the low latency of the system.

This article is devoted to promising methods of detection of unmanned aerial vehicles by phased antenna array with a wide bandwidth and providing distributed power supply of this array using a controller in the power supply circuits. Range extension of this phased antenna array is achieved by differentiating the antenna web into sub-bands to which a particular antenna section corresponds. This method allows the antenna to work on aerodynamic targets of different configurations and having different effective scattering surface. In contrast to the modeling described in articles [1], [2] modeling of the proposed antenna was carried out in the frequency range of 1 GHz-2 GHz. The antenna circuit uses a single power supply and a controller that allows the antenna sections to operate algorithmically depending on the type of target, target acquisition, acquisition reset, etc. The developed antenna can be used both in military and space technology, but when used in space technology, there will be problems, the impact of rocket engine plasma on the electromagnetic radiation of the antenna as it was described in the article [3], but the analysis of the use of short radio pulse [4] shows that this problem is solvable. In this paper, electrodynamic modeling of frequency characteristics of the radiators of a four-section phased antenna array is carried out.

A new method for estimating the aspect angle of ships at sea from an ISAR is developed. The ISAR AutoTrack (IAT) algorithm uses the information from the adaptive motion compensation velocity to improve the tracker estimation of the ship aspect angle and thus to improve the estimation of the ship length. The IAT is based on classical methods of autofocus for synthetic aperture radar. The mean of this velocity yields the error in the in-range component of the ship velocity; the linear time trend of the velocity determines the cross-range component of the ship velocity. Since the IAT creates its estimates during the ISAR time window it is unaffected by ship maneuvers. The IAT has two methods for implementing the algorithm. The Search Method tests a range of errors in course and speed to find the values that best model the data. The Analytical Method uses explicit formulas for the in-range and cross-range components of the ship velocity vector. Both methods benefit from an intelligent smoothing process that removes system errors, random noise, and ocean waves. The goal of the IAT is to measure ship length to within 10 percent over all azimuth angles and ranges relative to the aircraft and for (unsigned) aspect angles from 5 to 85 degrees. To develop a method that satisfies these requirements the IAT is evaluated for three types of tracking errors-errors in the ship speed, course, and Lat/Lon position. But errors in range are assumed to be corrected by routine algorithms in the ISAR processor. Thus, the position error reduces to error in the initial bearing to the ship. The IAT is validated by the analysis of data from an airborne test radar for 22 ship cases. This analysis showed an improvement in the RMS aspect angle from 26 degrees for the tracker to 8 degrees after the IAT. When the tracker position estimate is replaced with the AIS ship position data the error is reduced to 5 degrees; and the ship length error is nearly the same as if the full AIS results for course, speed and position were used. The data analysis also revealed some interesting insights into the effects of shadowing and multipath scattering on the length estimates. Algorithm testing with simulated data shows that the IAT can readily correct for errors in ship speed and course when the initial bearing is known well. However, at higher aspect angles errors in bearing lead to large aspect and length errors. Thus, for the desired range of operating conditions the IAT may need to be supplemented by hardware or software improvements in the bearing estimate. In addition, at azimuth angles ahead or behind the radar flight direction the IAT has a partial blind spot where the cross-range velocity cannot easily be estimated. To remedy this problem the time derivative of the bearing also needs to be estimated, giving a more direct estimate of the cross-range velocity. Recommendations for further development and testing of the IAT are given in the summary.

It is tremendously costly to produce large volumes of datasets with extensive annotation, which object detection requires. In this paper, we provide a framework in which GAN and Knowledge Distillation can be effectively combined to train object detectors from lightly annotated data. Given a source domain that is fully annotated, a small (20%) annotated target domain, and a large non-annotated target domain, the proposed method uses GAN to generate images in the synthetic target domain and creates pseudo-labels with a teacher model. Training an object detector model with detection and distillation losses yielded an overall mean average precision (mAP) on the target domain of 60.2%, outperforming the baseline with only 20% annotations at 51.8% and closing the gap to the 100% annotation baseline at 62.8%. Source code is available upon request.

The rapid expansion of multimodal data, encompassing text, images, and audio, presents significant challenges in efficiently processing and interpreting incomplete or partiallydefined information. The proposed framework introduces a novel approach to token propagation, addressing the inherent difficulties of grounding partially-defined tokens across multiple modalities without human intervention. Through an open-source LLM model, Llama, the study successfully demonstrated how cross-modal interactions can resolve ambiguities in token definitions, particularly in text and image data, while also identifying the limitations associated with audio processing. The methodology showed that contextual data from various modalities can be effectively integrated to refine token representations, leading to improved accuracy in multimodal tasks. The comprehensive evaluation of token grounding performance highlights the potential for further extending LLM capabilities in real-world applications, where heterogeneous data sources are increasingly prevalent.

In this article, we introduce a novel integral equation (IE) solver for inductance extraction that incorporates both Fast Multipole Method (FMM) and Fast Fourier Transform (FFT) acceleration. Crucially, FMM allows to deal with arbitrary polyhedral grids and thus with both tetrahedral and voxelized grids, whereas FFT requires exclusively voxelized grids. The FFT acceleration in the proposed framework differs from other FFT-accelerated approaches thanks to the use of element-wise constant basis functions, which reduce the computational cost and enable a simple singularity extration. We demonstrate pros and cons of three different combinations of acceleration method and type of grid, namely FMM-Tetrahedra (FMM-Tet), FMM-Voxel and FFT-Voxel, by applying them to a set of toyproblems as well as a real-world structure of a printed circuit board involving millions of unknowns. The results point out two inherent limitations of FFT methods, due to the necessity of using voxels of the same size along fixed Cartesian directions: the poor geometric approximation accuracy of curved geometries, and the high matrix fill-in resulting from the modeling of thin structures immersed in air. We conclude that FFT acceleration is better suited as a specialized tool in a more general framework, to be used only when its strict favorable conditions are met. This conclusion puts in perspective the performance gains of alternative methods proposed in literature that employ FFTacceleration on voxelized grids for the case of thin and curved geometries, thus suggesting new future directions of research.

Simulating the fine geometry of Metasurfaces (MTS) structures in a conventional way is a difficult task requiring huge computational resources. On the other hand, the metallization can usually be modeled as a (cascaded) impedance sheet(s) with modulation scale of the order of the operating wavelength. Such modeling, while homogenizing the currents and fields on the surface, is accurate in predicting the far-field as well as the near-field down to a fraction of wavelength from the surface. However, analyzing an impedance sheet lying on a grounded slab with the Method of Moments (MoM) often leads to ill-conditioning; it is certainly the case when the range of impedance spans both the capacitive and inductive domains. Such impedance range is in practice required for Reflective Intelligent Surfaces (RIS), and for some MTS antennas. This paper proposes a preconditioner aiming to solve ill-conditioning issues caused by the wide range of the surface impedance. The preconditioner involves a multiplication by the conjugate of the MoM matrix followed by a block diagonal preconditioner. The preconditioning is implemented in an accelerated scheme relying on FFTs. Given the resulting well-conditioned system of equations matrix, the unknown current distribution is efficiently computed iteratively using the Generalized Minimal RESidual (GMRES) method. A fast convergence is observed independently of the impedance range.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  

In recent years, the rapid growth of research on Large Language Models (LLMs) has led to a surge in survey papers, making it difficult for beginners to take in all the latest advancements. This report aims to overcome this challenge by analyzing a collection of LLM survey papers to look at key trends and provide useful insights into the field. We aggregate a dataset containing titles, summaries, categories, and release dates of LLM survey papers, then preprocess the data and apply a Random Forest Model to predict the categories of surveys based on their textual content. Our evaluation demonstrates the use of Random Forest Classifiers, achieving a precision of 0.70 but an accuracy of 0.38, highlighting room for improvement. Overall, this work contributes to making LLM research more accessible by offering a structured exploration of trends in survey publications and laying the foundation for automatic categorization of future research surveys.

This paper presents a methodology to allocate the savings from expected reduction in grid investment costs in a voltage-constrained distribution grid due to Energy Communities’ (ECs) flexibility. Incentives to provide flexibility are distributed both spatially to the contributing ECs in the grid, and temporally in the form of hourly incentives for providing flexibility. The savings from investment cost reduction are distributed using Shapley values, based on each EC’s marginal contributions to voltage improvement. Given the computational intensity of calculating Shapley values, especially with numerous ECs, we further introduce a stratified sampling method to improve the computational efficiency without compromising the fairness of incentive allocation to the ECs. Finally, using Bayesian regression, an approximation method is proposed to map the EC’s location in the grid to its incentive allocation. Bayesian regression offers a probabilistic approach to estimate the Shapley value, accommodating uncertainties inherent in the calculation by using probability distributions for the parameters of regression. The performance of the proposed methodology and approximation methods are demonstrated through extensive case studies on a representative Norwegian medium voltage distribution grid.

The deployment of fifth-generation (5G) networks across various industry verticals is poised to transform communication and data exchange, promising unparalleled speed and capacity. However, the security concerns related to the widespread adoption of 5G, particularly in mission-critical sectors, present significant challenges. This article investigates the potential of a Zero Trust (ZT) security philosophy as a viable countermeasure to these concerns. It delves into the practicalities of implementing ZT principles within 5G networks, with a specific focus on harnessing AI/ML technologies for proactive security measures, dynamic policy adaptations, and advanced risk assessments. Further, the article underscores the importance of developing a tailored ZT maturity model for 5G networks. Furthermore, the paper outlines key future research directions aimed at improving the ZT maturity of 5G deployments, contributing to the safe and secure integration of 5G technology in various sectors.

Ransomware is one of the most critical cybersecurity threats, targeting systems globally and causing substantial financial and data losses. Developing an effective and efficient detection mechanism within operating systems is essential for mitigating these risks, which is where the novel integration of LSTM-based machine learning within the Linux kernel offers a significant advancement. The proposed approach involves the deployment of an LSTM model capable of analyzing system call sequences in realtime, identifying anomalous behavior indicative of ransomware activities with high accuracy. The research outlines the model's design, including data collection, preprocessing, and the intricate integration process into the Linux kernel to facilitate seamless communication and classification. Extensive experimental evaluation demonstrated that the LSTM model achieved superior detection performance when compared to traditional methods, maintaining a high level of accuracy and low false positive rates. The findings indicate that this method not only provides a proactive defense mechanism against emerging ransomware threats but also enhances the overall security framework of Linux systems through its adaptive learning capabilities and efficient resource utilization.

Federated Learning (FL) is a powerful Machine Learning (ML) technique that allows multiple clients to collaborate on training models while keeping their data private. Unlike traditional centralized methods, FL ensures that data are kept separate, which helps to protect privacy. However, an important area that needs more research while using the FL system is detecting harmful models within the Internet of Things (IoT) context. For example, poisoning attacks, where compromised clients introduce harmful data, can degrade the model's overall performance or lead to incorrect predictions. This paper comprehensively reviews of recent attacks in FL within IoT networks, along with defense mechanisms and common FL frameworks. It begins by highlighting the significance of FL in IoT networks, exploring its applications, benefits, and inherent security challenges. It then explores specific attacks targeting FL in IoT networks. The defensive strategies are evaluated, including their performance metrics, datasets used, and related work, providing a comparative analysis of these techniques. Common FL frameworks and their criteria are reviewed. Our goal is to offer a detailed understanding and solutions to enhance the strength and resilience of FL systems in IoT networks.