Spinal cord injuries (SCI) result in significant neurological and functional impairments, with current assessments like the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) showing limited sensitivity in detecting residual motor control. Surface electromyography (sEMG) is a sensitive, non-invasive, and quantitative technique to measure muscle activity, even when there are no visible muscle contractions. In order to advance the use of sEMG in the management of SCI, this study aimed to identify distinct muscle electrophysiological profiles from sEMG features in individuals with cervical SCI, hypothesizing the existence of identifiable clusters within sEMG feature space. Using clustering algorithms such as k-means, k-medoids, DBSCAN, and hierarchical clustering, distinct and reproducible clusters were identified from curated sEMG feature sets with commonly used sEMG features in time and frequency domains. The clusters were then correlated with clinical indicators of neurological and muscle function. While significant associations were demonstrated with clinical variables including myotome level, neurological level of injury, and manual muscle testing scores, the structure in the sEMG feature space revealed by the clustering analysis was not fully accounted for by the clinical variables. These findings suggest that sEMG provides complementary information and could enhance assessments and guide clinical treatments for individuals with spinal cord impairments. Innovative tools to stratify patient groups based on detailed electrophysiological profiles may offer significant implications for clinical trial design.

The field of artificial intelligence (AI) and machine learning (ML) has seen remarkable advances with the advent of neural networks. This paper presents a comprehensive comparison between Convolutional Neural Networks (CNNs) and traditional Backpropagation Neural Networks (BPNNs). By exploring the underlying mechanisms, performance metrics, and applications of both types of neural networks, we aim to highlight their respective strengths and limitations. The study provides a detailed examination of the architecture, training processes, and suitability for various tasks, supported by recent empirical research and theoretical analysis.

The geolocation of signals from high altitudes is increasingly vital to combat terrorism and support war fighting globally. In a recent paper [1], the author described an approach to improve geolocation performance by focusing on understanding the impact of biases, particularly navigation errors, on sensor platforms. That analysis model used a hypothetical scenario with drones collecting time and frequency data from an emitter at an altitude of 10 to 15 km for several minutes, with a strong, time-varying observation geometry. This note applies that model to a case of more practical interest: shorter observations (a minute or less) from geosynchronous altitudes and a fixed observation geometry over a very large coverage area. The considered covariance approach, with the use of the full 6x6 navigation error in the covariance computation, is shown to be a valuable tool in the system engineering specification of the collection system, as well as in addressing biases and improving system geolocation performance.

Climate change and operation approaches have significant impacts on the total energy consumption of buildings. This paper proposes a two-stage operational-aware planning for energy-efficient buildings to realize the climate-resilient lowcarbon transition within buildings considering the uncertainties of local weather conditions induced by climate change. Both normal weather conditions and extreme weather conditions, i.e., heat waves, under climate change, are projected using morphing techniques, where the general circulation models and heat wave detection method are incorporated. A temperature-dependent heat pump model with finite output levels is proposed in the second-stage operational problem to capture its flexibilities. A novel sequential decomposition scheme is proposed to solve the two-stage problem with a large scale of supports, which can converge rapidly in finite iterations. The simulation is conducted on a real Stockholm educational building. Numerical results verify the effectiveness of the proposed scheme regarding climate change awareness and computational performance. Specifically, the results demonstrate a significant 8.33% reduction in heating capacity and a substantial 62.14% decrease in solution time.

As research into candidate waveforms for sixth generation (6G) communication systems accelerates, chirp-based multicarrier, notably the generalized version Affine frequency division multiplexing (AFDM) have garnered increased attention. This paper introduces a novel, low-complexity channel estimation (CE) method that exploits the properties of chirp carriers in the post-coded time domain (PCTD). Initially, we derive the explicit input-output relationship under practical pulse shaping, highlighting the impact of pulse mismatch in the Affine domain. Subsequently, we analyze the inherent properties of AFDM in the Affine domain and identify the limitations of current CE techniques. As such, we present a detailed representation of single and multiple chirp carriers in the PCTD both preand post-channel transmission, with a focus on interpreting the effects of delay and Doppler shifts. Building on that, we propose a two-stage CE method leveraging the zero-padded PCTD received signal, effectively decoupling delay and Doppler components while compensating for phase selectivity and pulse shaping-induced inter-carrier interference (ICI). Our approach considers data-pilot interference levels within adaptive scenariobased frame structures, leading to a significant reduction in complexity and enhanced spectral efficiency (SE) due to minimized pilot overhead. Simulation results reflect the superiority of our proposed method over existing iterative schemes, highlighting its efficacy in improving performance in practical AFDM systems.

As neural networks are trained to perform tasks of increasing complexity, their size increases, which presents several challenges in their deployment on devices with limited resources. To cope with this, a recently proposed approach hinges on substituting the classical Multiply-and-ACcumulate (MAC) neurons in the hidden layers with other neurons called Multiply-And-Max/min (MAM) whose selective behavior helps identify important interconnections, thus allowing aggressive pruning of the others. Hybrid MAM&MAC structures promise a 10× or even 100× reduction in their memory footprint compared to what can be obtained by pruning MAC-only structures. However, a cornerstone of maintaining this promise is the assumption that MAC&MAM architectures have the same expressive power as MAC-only ones. To concretize such a cornerstone, we take here a step in the theoretical characterization of the capabilities of mixed MAM&MAC networks. We prove, with two theorems, that two hidden MAM layers followed by a MAC neuron with possibly a normalization stage is a universal approximator.

Urbanization is an inevitable process linked to regional development. As cities and towns expand, they put more pressure on the environment and resources. Understanding the evolution of urban areas and its effects on landscape dynamics are crucial for sustainable urban development. Spatial urban expansion models, such as the Cellular Automata based Patch-generating Land Use Simulation (CA-PLUS) model, have proven to be effective in tracking urban growth and predicting future expansion. These models incorporate various spatial parameters to assess their influence on regional urbanization. This study forecasts urbanization trends for the years 2025 and 2030 in Bangalore City, using the CA-PLUS model. This model integrates a land expansion analysis strategy (LEAS) with Cellular Automata-based random patch seeds (CARS), along with multiple driving factors. A 16-year time series dataset (from 2006 to 2022) was used to analyze land use patterns. The results show significant expansion in Bangalore, initially driven by growth at its core and later by peri-urban development, which has now extended beyond the city's boundaries. Various driving factors influencing land use changes were also identified and the model projected land use demand for 2022, which was then evaluated against the actual LULC map for 2022. Findings indicate that urban population, proximity to metro stations and road networks have a greater impact on urban expansion than other variables. The calibrated CA-PLUS model achieved 83% accuracy and a kappa coefficient exceeding 74%, demonstrating its utility in understanding urban land use dynamics for sustainable planning.

In this paper, we propose a novel Canon Q-Transformation method for signal quantization that differs from traditional vector quantization (Q-Transforms) approaches. Our method focuses on generalizations on canonic expected signals, providing a more accurate representation of expectable signals. Unlike traditional methods, which often overlook the classification of expectable signals via fundamental measurement limitations compared to the affirmation of expected signal as detected signal via practical filter quantization or determining the expected signal by theoretical idealization, our approach integrates this key element to Q-Transform module in order to enhance signal fidelity.

We demonstrate the use of playtraces and playtrace clustering to identify strategies and card synergies in deck building card games, using Dominion as an example. We analyse two different types of playtraces generated from both online human play and a variety of AI agents. Firstly we examine playtraces defined by tracking card counts by round in a player's deck, and secondly utilising N-Grams generated from player actions. We investigate playtrace clustering using both the L knorm and Jensen-Shannon distance measures, in-conjunction with K-Means, K-Medoids and DBSCAN algorithms. We show that distinct playtrace clusters can arise from different strategies, and that playtraces and cluster centroids provide a means to identify both longer term strategies and also short-term tactics involving particular card combinations. Additionally, we use a restricted play framework to increase the variation in strategies and tactics explored by the AI agent. We conclude by highlighting how the game agnostic N-Gram based approach might be used to explore strategy spaces in tabletop games more broadly.

This study investigates market differentiation strategies within the luxury perfume industry and their economic impacts on brand value through an Exploratory Data Analysis (EDA) approach. In a highly competitive market, brands employ strategies beyond mere fragrance, emphasizing brand experience, packaging, and marketing to enhance perceived value. Understanding the economic impact of these strategies is vital for gaining a competitive edge and commanding premium pricing. The research utilizes Optical Character Recognition (OCR) to extract ingredient lists from perfume bottle labels, converting physical label information into digital data. This digitized data is then matched with a predefined cost table to determine the cost of each ingredient, providing insights into the cost structure of perfumes.By calculating the cumulative cost of ingredients, a benchmark for comparing production costs with retail prices is established, highlighting the added value from branding and marketing efforts. Comparing these costs with retail prices offers a direct measure of brand value, indicating the premium consumers pay for perceived value. Significant price differentials point to strong brand value, reflecting effective marketing, premium packaging, and robust brand reputation. Graphical representations, such as bar graphs and pie charts, facilitate understanding of the economic dynamics, revealing patterns and trends in brand value.The study identifies key drivers of brand value, including marketing strategies, premium packaging, and strong brand reputation, which justify higher retail prices and emphasize the successful implementation of market differentiation strategies. The findings underscore the importance of strategic market positioning, enabling luxury perfumes to target niche markets and sustain competitive differentiation. This research provides valuable insights for companies to make informed strategic decisions, leveraging brand value for higher profit margins and sustained market differentiation.

The task of knowledge graph completion poses significant challenges due to the vast scale and complexity of relationships that need to be predicted within large datasets. Introducing a novel concept of partial dormancy within the GPT-Neo architecture offers a significant advancement by enabling selective neuron activation, thereby reducing computational overhead while maintaining high accuracy. Through dynamic gating mechanisms, the model efficiently handles diverse input complexities, resulting in optimized performance across various knowledge graph sizes. The model's ability to balance computational efficiency with high accuracy highlights its potential for deployment in resource-constrained environments. Experimental results demonstrated that the partially dormant GPT-Neo outperforms fully active models in both efficiency and scalability, suggesting a transformative approach for future applications in knowledge graph-related tasks. The contributions made extend beyond enhancing model efficiency, providing a foundational framework that may influence the development of more adaptable and energy-efficient language models.

Image reconstruction is the manipulation of digitized information obtained during body imaging into interpretable pictures that represent anatomical details and diseases. It is a vital step in producing visual results from a Computed Tomography (CT), Magnetic Image Resonance (MRI) etc. scanning devices. In all modalities (CT, MRI etc.) the data is acquired in form of slices or projections. In simple terms, the process of putting the projections together in a two-dimensional (2D) or three dimensional (3D) scans can be called image reconstruction. Back projection and forward projections are two commonly used operations in the image reconstruction pipeline. And since the image reconstruction algorithms can be compute intensive, there is desire to fasten the processing time for it, to get the final scan faster. The scans are what a medical professional such as a radiologist use to provide diagnosis to a patient, hence faster access to scan can result in more timely diagnosis.  With the goal to reduce the processing time for image reconstruction pipeline, the authors worked to analyze and optimize representative back projection and forward projection algorithms. This paper provides details on the various optimization and code migration work that was done to achieve targeted performance metrics using a hardware vendor neutral programming language.

Vision-based Transformers have emerged as a revolutionary architecture in the field of robotics, offering significant advancements in perception, decision-making, and control. This survey aims to provide a comprehensive review of the application of Vision-based Transformers in robotic systems. We explore the fundamental principles of Transformers, highlighting their ability to handle longrange dependencies and contextual understanding, which are critical for complex robotic tasks. The paper delves into various implementations of Vision-based Transformers across different robotic domains, including object detection, autonomous navigation, manipulation, and human-robot interaction. We also discuss the challenges and limitations associated with integrating Transformers into robotic systems, such as computational demands and real-time processing constraints. Furthermore, we present a comparative analysis of Vision-based Transformers with traditional convolutional neural networks and other state-of-the-art approaches, underscoring their unique advantages. Finally, the survey identifies promising research directions and potential future applications, aiming to guide and inspire further innovations in this rapidly evolving field.

The advent of autonomous driving technology necessitates accurate prediction of human trajectories to ensure safety and efficiency in urban environments. This paper explores advanced methodologies for human trajectory prediction, focusing on the integration of machine-learning techniques and sensor fusion. We propose a novel framework that combines recurrent neural networks (RNNs) with attention mechanisms to enhance the prediction accuracy of pedestrian and cyclist movements. By leveraging data from LiDAR, cameras, and GPS, our approach dynamically adapts to the diverse and often unpredictable behavior of humans in traffic scenarios. We evaluate our model on several benchmark datasets, demonstrating significant improvements over existing state-of-the-art methods. The results indicate that our method can reliably forecast shortterm and long-term trajectories, contributing to the development of safer and more reliable autonomous driving systems. Future work will address scalability and real-time implementation to further bridge the gap between research and practical deployment.

This paper introduces a groundbreaking framework that establishes a cohesive link between first principles-based simulations and circuit-level analyses using a machine learning-based compact modeling platform. Beginning with meticulous atomistic simulations, the framework explores the microscopic intricacies of material behavior, serving as the foundational layer for subsequent stages. The generated datasets, enriched with molecular insights, undergo a transformative process guided by machine learning algorithms, enabling the discernment of complex patterns and relationships. As these machine-learning models mature, they evolve into powerful tools capable of predicting intricate behaviors beyond the reach of conventional modeling and simulation approaches. Applied to circuit simulation, the framework excels in providing a holistic understanding of electrical interactions, significantly enhancing accuracy, and expediting design automation. As a proof of concept, we perform first principles-based simulations of the Graphene Nanoribbon Field Effect Transistor (GNRFET), an exploratory device, and create a symbolic-regression-based machine learning model that can readily be integrated into advanced circuit simulation. This framework presents a versatile template offering a unified approach that synergizes the strengths of first principles-based simulations and circuit-level design tools.

Verification tools for cyber-physical systems (CPS) often have their own input specification language based on formal models of CPS. This requires tool users to have acquaintance to formal models and thus poses a barrier for new users to get introduced to the verification technology. To address this concern, Verse is a recent Python library that aims to make modeling and verification of CPS more accessible. This library does not require the users to have prior knowledge of formal models of CPS. With the growing popularity of Python, the library has the potential to broaden the user base for CPS verification. In this paper, we study the Verse library and its capabilities and limitations through two case studies: one in vehicle control and the other in swarm robotics, demonstrating the differences from conventional CPS modeling.

The cybersecurity for IoT devices is often an afterthought, as they often have little need for a "human interface". If IoT designers think about security at all, they may follow the "hard and crunchy on the outside, soft and chewy inside" model, which emphasizes perimeter defense, on the theory that if hackers can't get in, they can't do damage. The catch is that even the smallest IoT devices may have a software stack that's too complex for the designers to comprehend as a whole, usually leading to obscure vulnerabilities in dark corners of the code, where normal designers would never look.

Since the discovery of carbon nanotubes, they have been studied widely and used in many applications. In this work, we have designed, fabricated, and tested several carbon nanotube-based patch antennas. Carbon nanotubes (CNTs) are grown in-house on ceramic substrates using a chemical vapor deposition technique. Iron and carbon are sputtered as the catalyst material before CNTs are grown. Patch antennas are designed for 2.4 GHz and 5 GHz frequencies. The measured S11 parameter and radiation pattern of each antenna are presented. A comparison with copper patch antenna is made.