Authors - Marco Torres-Umana, Ernesto Rivera-Alvarado Abstract - This paper presents a reinforcement learning (RL) approach for validating graphical user interfaces (GUIs) on networking devices. Traditional methods, including manual and automated approaches, face challenges in scalability, efficiency, and adaptability. The proposed RL solution generates test cases dynamically, exploring diverse GUI states and behaviors without relying on pre-established models or extensive data. By leveraging internal and external device observations and encoding techniques, the RL agent effectively navigates GUIs. Results demonstrate high solution similarity and shorter convergence times across various configurations, enhancing test coverage while minimizing manual effort. Future work will refine reward definitions, tackle larger state spaces, and extend the system to support additional devices and vendor interfaces.
Authors - Layan Sawalha, Jiamei Deng, Temitope Omotayo Abstract - The accurate analysis of hybrid medical datasets consisting of textual reports and diagnostic images plays an important role in the early detection and better outcomes of breast cancer patients. A novel deep learning framework is proposed in this paper that combines Bidirectional Encoder Representations from Transformers (BERT), Generative Pre-trained Transformer-2 (GPT-2), and Convolutional Neural Networks (CNN) to overcome the challenges of multimodal data in breast cancer. This framework combines the strengths of BERT and GPT-2 for extracting rich contextual features from text with CNNs for capturing complex patterns in diagnostic images. By integrating textual and visual features into unified latent representations, this fusion enables accurate classification of breast cancer, distinguishing malignant from benign cases using both text and imaging data. The proposed framework lessens the bottleneck in multimodal to achieve outstanding results with an accuracy of 1.00, hence remarkably improving the precision of breast cancer diagnostic.
Authors - Humberto Merritt Abstract - Mobile applications have become the most important digital technology for social interaction, culture formation, and knowledge sharing in recent years. Electronic platforms such as search engines, social network sites (SNS), on-demand Internet streaming media, and smartphone apps swiftly shape our daily activities politically, culturally, and technologically. Although SNS impact how people interact with one another, their usage varies depending on the type of application. This research explores what social network sites are preferred in Mexico according to the application and platform used. In particular, it seeks to answer the following questions: 1) What SNS are preferred by Mexicans, and 2) What are the sociodemographic factors that determine their preferences? The methodology follows a quantitative approach that draws on the National Survey on the Availability and Use of Information Technologies in Households (ENDUTIH) carried out in 2023 by the National Institute of Statistics and Geography of Mexico (INEGI). Empirical evidence suggests that Mexicans favor SNS such as Facebook, WhatsApp, Instagram and YouTube. Average users are in their mid-thirties, and women are more intensive users of SNS.
Authors - Guang-Jhe Lin, Chih-Heng Ke, Cheng-Feng Hung Abstract - The adequacy of the bandwidth is essential for delivering stable and high-quality transmission services. However, existing routing methods often rely on single transmission paths, which limit bandwidth utilization and place higher demands on routing decisions for bandwidth allocation. Small traffic flows may occupy high-capacity paths, leaving larger traffic flows unmet. To address these challenges, this paper proposes a novel routing algorithm that allocates paths based on bandwidth demands. By decomposing the routing problem through the overlay of multiple single paths, the algorithm reduces complexity. In addition, it integrates traffic splitting techniques and reinforcement learning to dynamically optimize path selection and improve performance in software-defined networks. The simulation results show that the proposed method achieves an average improvement of approximately 50.79% in system throughput compared to shortest path routing and 34.68% compared to maximum bandwidth path routing, demonstrating its effectiveness in optimizing network resource utilization.
Authors - Shahinul Hoque, Farhin Farhad Riya, Jinyuan Sun, Hairong Qi, Kevin Tomsovic Abstract - Machine learning (ML) models hosted on cloud platforms are increasingly susceptible to security vulnerabilities, particularly due to their exposure to external queries in untrusted environments. In this paper, we explore this specific vulnerability by leveraging fuzzing techniques to systematically generate diverse input samples (X) to query cloud-hosted ML models. By capturing the corresponding outputs (y), we attempted to train a shadow model that mimics the behavior of the target model. This methodology allows us to systematically assess the security risks associated with such models, including information leakage, extraction of decision boundaries, and model inversion. The core of our study is to determine the feasibility of mimicking cloud-hosted ML models using shadow models trained via various fuzzing attacks. We focus on computationally efficient fuzzing methods to evaluate the practicality of these attacks. Our findings demonstrate that fuzzing effectively creates a comprehensive dataset for training the shadow model, thereby minimizing the number of queries needed to mount successful attacks. Moreover, we discuss the broader implications of these security breaches on the confidentiality, integrity, and availability of the models, identifying significant security deficiencies in current deployment practices of cloud-hosted ML models. We conclude with proposed countermeasures designed to defend the security of these systems, underscoring the importance of implementing robust defensive strategies in cloud-based ML frameworks.
Authors - Julkar Nine, Akif Ahmed, Shibbir Ahmed, Wolfram Hardt Abstract - The push toward full autonomy in the automotive industry has been bolstered by advancements in Artificial Intelligence, particularly Deep Learning and Machine Learning techniques, which facilitate efficient processing of sensor data for environmental object detection and recognition. Detection of objects with precision and robustness becomes particularly critical on roads, where increased vehicle speeds require fast and reliable decision-making. German highways, known for their high average vehicle speeds, underscore the need for specialized object detection algorithms tailored for such environments. However, these technologies typically depend on large datasets for optimal performance. This research presents a novel approach to detecting and recognizing German highway traffic signs using YOLOv8 while addressing the challenge of limited data availability. By curating high-quality custom datasets for training, the model achieved approximately 90% accuracy on test data and 80% on real-world data, even with reduced dataset size, and demonstrated semi-real- time to real-time performance, highlighting its potential for practical deployment in autonomous driving systems.
Authors - Wolf A. Kafka, Gunther Bernatzky, Pravir Malik Abstract - Chronic fatigue, sleep disturbances, and mental exhaustion are increasingly prevalent in modern societies, necessitating innovative, non-invasive therapeutic solutions. This study evaluates the impact of the Centropix KLOUD Pulsed Electromagnetic Activation (PEMA) system on various dimensions of human well-being. Conducted by an independent research group (SFU Vienna), the study involved 428 participants (aged 19–84) across Germany, Austria, Switzerland, and the USA. Participants used the KLOUD device—a full-body electromagnetic field mat delivering broadband pulsed stimulation—for 15 minutes daily over 42 days. Validated assessment tools, including the Veterans RAND- 12, Fatigue Assessment Scale (FAS), Jenkins Sleep Scale, and NRS-11 for discomfort, were used to measure outcomes at baseline (T-0) and after the intervention (T-1). Results revealed statistically significant reductions in exhaustion, and improvements in mental performance, energy levels, and sleep quality. In a key subgroup (n = 103) identified with severe exhaustion symptoms at baseline, only 14.9 % continued to report such symptoms after the intervention—an 85.1 % reduction. These findings suggest that the KLOUD-PEMA system offers clinically relevant benefits for enhancing vitality and resilience without side effects, supporting its role in primary and secondary preventive health strategies. The technology’s simplicity, accessibility, and safety profile make it a compelling candidate for broader integrative health applications.
Authors - Tatavarthi Lakshmi Chandrasena, Arashdeep Kaur Abstract - The progress in Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) has significantly enhanced the technologies used for predicting solar flares. These advancements enable better analysis and interpretation of complex solar data, leading to improved accuracy in forecasting potentially disruptive solar events. Since solar flares can impact global communication networks, power systems, and pose health risks to astronauts due to elevated radiation levels, there is a pressing need for precise and timely predictive models. This paper explores various modern approaches to solar flare prediction, highlighting the contribution of AI-based tools in enhancing their accuracy and lead time. The integration of ML and DL not only refines these models but also presents challenges related to their complexity and data requirements. By assessing the strengths and weaknesses of these techniques and proposing potential improvements, this paper aims to provide the extensive review of various existing techniques and methods that have been deployed in the literature for solar flare prediction. This paper also gives a comparative study of various solar flare prediction models.
Authors - Ha L Thu Abstract - The rapid advancement of Big Data has transformed business network analysis, enabling organizations to extract meaningful insights for decision-making and optimization. This paper explores key methodologies in business network analysis, focusing on how Big Data techniques enhance network visualization, relationship mapping, and performance optimization. We discuss applications in supply chain management, financial networks, and corporate strategy. Additionally, we address challenges such as data privacy, computational scalability, and real-time processing. Finally, we propose future research directions for integrating artificial intelligence and advanced analytics into business network optimization.
Authors - Armida Gonzalez-Lorence, Jose Alejandro Ascencio-Laguna, Cornelio Morales-Morales, Jose Gabriel Ayala-Landeros, Juan Emigdio Soto-Osornio Abstract - This research develops an integrated system that combines the Internet of Things with machine learning for the purpose of optimizing environmental conditions in Mexican poultry farms. A four-module architecture is implemented: IoT Module for real-time environmental data collection through various sensors (DHT22, MQ-7, MQ-137, MG-811), processing and storage module, multivariable machine learning module, and visualization module. Experimental validation was conducted over 62 days in a commercial poultry farm, continuously monitoring critical variables of temperature, humidity, CO₂, and NH₃. The data were processed using classification and regression algorithms, including Random Forest, neural networks, and Gradient Boosting, to generate real-time recommendations. Random Forest algorithms showed the best classification performance (68% accuracy), while Gradient Boosting achieved the lowest mean square error in regression (RMSE=1.32). Through variable importance analysis, it was identified that indoor temperature (37.5%), CO₂ levels (18.3%), and bird age (15.7%) are the most significant variables. Therefore, an Agglomerative Hierarchical Clustering analysis (k=5) was executed, which allowed categorize 5 specific microenvironments. The system implementation makes predictions about the trend of temperature, humidity, NH3, and CO2. The developed system establishes a significant evidence-based advancement for poultry farming in Mexico.
Authors - Leonardo Juan Ramirez Lopez, Juan Sebastian Orozco Duran Abstract - The Interoperable Electronic Records Platform (PIRE) is an interoperable solution designed to unify existing telemedicine systems and integrate Medical Internet of Things (IoMT) platforms into a single, secure, and scalable ecosystem. This approach enables real-time consolidation of clinical information, facilitating more agile and effective decision-making. From a project management perspective, PIRE represents an innovative strategy by centralizing data from multiple sources, thereby optimizing the planning, execution, and monitoring of digital health initiatives. The project combines exploratory, descriptive, and analytical methods within the Scrum framework, structured across four iterative cycles: (1) architecture definition and IoMT device integration, (2) security planning and clinical data protection, (3) graphical interface and relational model design, and (4) testing and validation in simulated and real-world environments. PIRE adopts international standards such as HL7 FHIR, DICOM, and SNOMED CT to ensure interoperability among heterogeneous healthcare systems. Preliminary results demonstrate a significant reduction in clinical data access times, enhanced information security, and increased operational efficiency through redundancy elimination. Overall, PIRE not only strengthens care continuity and patient safety but also enables a replicable model for digital health project management, grounded in interoperable platforms that unify processes, reduce costs, and improve care quality.
Authors - Falk Pospischil, Yves Burkhardt, Oliver Klar, Johann Lichtblau Abstract - This paper presents the optimization problem of distributing traction forces among parallel group drives in railway vehicles to reduce energy consumption while considering constraints. The model requirements, including temperature-dependent losses, are deduced to precisely determine energy losses. The approach for online implementation of optimal traction force distribution using a Look-Up Table (LUT) is analyzed through simulation on a realistic drive cycle, resulting in an overall energy demand reduction of approximately 0.5 %. The effects of alternating driving strategies on wheel life and lifespan consumption of switching power electronics are examined, revealing that lifespans can be negatively affected by energy-efficient traction force distribution.
Authors - Nguyen T. Lan Huong Abstract - This paper explores the role of Data Science in modern economic anal- ysis. We discuss key methodologies, including machine learning, econometrics, and big data analytics, and their applications in economic forecasting, market analysis, and policy evaluation. Additionally, we highlight challenges such as data quality, ethical concerns, and computational complexity. The study concludes with future research directions and potential advancements in integrating Data Science with economic modeling.
Authors - Gopal S. Gawande, Virendra Shete, Maheshkumar Kolekar, Sachin Takale, Sanjaykumar Nipanikar Abstract - As a potent DSP technology, multirate DSP allows for inexpensive digital filter implementations, and it is extensively used for satisfying the sampling rates of various systems. An assortment of state-of-the-art digital signal processing (DSP) methods are at your disposal for optimizing multirate digital filters, including retiming, parallel processing, pipelining, folding, unfolding, and polyphase decomposition. This work presents new multirate filter designs that provide high computation rates, throughputs, and speeds by using a variety of optimization strategies. In order to reach the spartan-6 xc6slx150T-4fgg676 Field Programmable Gate Array device, decimation filters use multirate filter topologies and the Xilinx System Generator. It is observed that the speed is increased by 135 MHz for the transpose pipelined decimation filter as compared to simple decimation filter in average. Due to the parallelism property of polyphase decomposed structures, the throughput and computation rate are found maximum in case of efficient polyphase decimation filter structure. It has been observed that the polyphase decomposition technique enhances the average throughput and computation rate of the simple decimation filter by almost 14MSPS and 345 MMACPS respectively.
Authors - Youna Jung, Songyon Shin, Second Sujeong Park Abstract - Since Generative AI (GenAI) has introduced in 2022, it has transformed teaching and learning in higher education in diverse aspects. Students are increasingly using diverse GenAI tools to understand complex theories, generate ideas, draft work, create presentations with scripts, and obtain instant solutions to assignments. This dramatic shift in learning styles raises concerns on how to integrate this emerging technology while ensuring students are trained as problem- solvers who can effectively utilize AI without over-relying on it. As students’ needs vary across disciplines, it is critical to address this potential disparity in the ways in which different student groups learn, understand, and apply GenAI may differ, including in interdisciplinary courses. Towards this goal, this paper reviews the literature on GenAI in higher education, highlighting the lack of research addressing the differences among student groups. To bridge this gap, we propose the GAITE project, which aims to identify and address the unique needs of these groups through surveys and tailored solutions.
Authors - S. M. Mizanoor Rahman Abstract - We developed a human-robot collaborative system in the form of a hybrid cell to perform a collaborative assembly task in a flexible manufacturing setup. We conducted an experiment where 20 human subjects separately performed the assembly task in collaboration with the robot. We observed the collaborative assembly task, reviewed related state-of-the-art research results, interviewed the subjects, relevant researchers and industry experts and thus derived a comprehensive set of assessment and benchmark metrics and methodologies for the collaborative manipulation (co-manipulation) task. The metrics included the core requirements of the task expressed as the key performance indicators (KPI) and the effectiveness of physical interactions between a human collaborator and the robot for the task. The proposed benchmark metrics and methods can help assess and benchmark the overall performance and interactional effectiveness of human-robot collaborative manipulation tasks and systems for various applications such as manufacturing and assembly, logistics and transport, military operations, construction, disaster operations, etc.
Authors - Kanyanut Homsapaya, Bonnyakorn Leelakarnsakul, Waraporn Aumarm, Orawan Watchanupaporn Abstract - Congestive heart failure (CHF) in cats is a serious condition characterized by the heart’s inability to pump blood effectively, resulting in fluid accumulation in the lungs (pulmonary edema), chest cavity (pleural effusion), or abdomen (ascites). This life-threatening disorder manifests with clinical signs such as respiratory distress, profound lethargy with progression often culminating in organ failure if left untreated. Early detection is critical, and the integration of artificial intelligence (AI) offers significant potential for improving outcomes by analyzing diagnostic imaging, physiological data, and medical records to predict CHF onset and facilitate timely, targeted interventions. In this experiment, the data consists of 181 cats diagnosed with pulmonary edema, collected from Kasetsart University Veterinary Medicine Hospital. It includes relevant features such as clinical signs, diagnostic test results, medical history, and physiological data, providing a comprehensive basis for building a predictive model for congestive heart failure. The research utilizes multiple algorithms, including Support Vector Machine (SVM), KNN etc. to develop predictive models for congestive heart failure. The performance of each model is evaluated, and the algorithm with the highest accuracy and reliability is selected as the optimal approach. This work demonstrates the superior performance of Support Vector Machine (SVM) in predicting congestive heart failure (CHF), underscoring their significant advantages in both accuracy and efficiency compared to other models.
Authors - Sharukesh, Sai Sundarakrishna, Vignesh, Pradeep Abstract - Artificial Intelligence (AI) Technologies such as Generative AI prompting, Retrieval Augmented Generation (RAG) and Agentic AI have started to emerge in cybersecurity operations and use-cases independently. We propose an intelligent orchestration mechanism that sifts the requirements of contextual injection, smart autonomy, integrated use case and intent capture. It is capable of reducing hallucinations and capable of enhancing semantic reasoning using prompting, RAG and Agentic AI, smartly, simultaneously and on demand. We propose such an integrated novel suite of framework tools unifying System Integration and Events Management (SIEM) and security Orchestration Automation and Response (SOAR). We develop, discuss and demonstrate the system as a plug and play with popular cybersecurity platforms. We provide 2 real world case studies to demonstrate the efficiency over the current state of the art performing SIEM and SOAR tools. The plug-in is scalable for the Model context protocol (MCP) and Agent to Agent (A2A) ecosystems
Authors - Sunish Vengathattil, Shamnad Mohamed Shaffi Abstract - Citation resolution is essential for maintaining the accuracy and integrity of academic research. However, inconsistencies in citation formats, author name variations, and bibliographic errors make this process challenging. Traditional manual and rule-based methods are time-consuming and prone to errors, highlighting the need for advanced solutions. This study evaluates various machine learning models, including BERT, LSTM, and Random Forest, to improve citation resolution. Using AWS services for data storage, processing, and model training, the models were assessed based on accuracy and efficiency. Results show that deep learning models, particularly BERT, perform best in handling citation inconsistencies, though computational costs remain a concern. The findings emphasize the potential of machine learning in improving citation management for digital libraries and reference tools while suggesting future research for scalability and multilingual support.
Authors - Kazunori Minetaki Abstract - This study analyzed care workers' fatigue and stress levels through facial expression analysis, specifically using Percentage of Eye Closure (PERCLOS) and action units. A weighted average of AU04 and AU07 for the action unit analysis was used as an index. The results showed that several participants exhibited relatively high values in both PERCLOS, measuring the percentage of time eyes are closed, indicating drowsiness or fatigue based on eyelid closure duration. and the weighted average of AU04 and AU07, indicating higher fatigue or stress levels. Follow-up interviews with these individuals revealed that they were experiencing significant mental and physical strain due to responsibilities such as overseeing the overall operation of the care facility or dealing with interpersonal issues. Facial expression analysis, being non-invasive and time-efficient, shows potential as a practical tool for monitoring the condition of care workers in real-world settings. The findings are expected to inform interventions that support caregiver well-being and foster healthier work environments.
Authors - Christopher Agbonkhese, Omobola Gambo, Teslim Akande, Ishaya Gambo, Adebowale Adewuyi Abstract - This paper presents the design and development of a mobile application that serves as an AI-powered study assistant to support personalized learning and academic management. The app helps students create study plans, practice questions, track their progress, and receive feedback based on their unique learning styles and academic goals. The system uses a client-server architecture, making it scalable and efficient. It also includes useful features such as a scheduler, reminders, progress tracking, and integration with Google Calendar. Gamification elements are used to make studying more interactive and enjoyable for students. For the intelligent part of the system, we used TensorFlow and Scikit-learn libraries. These tools help the app understand each student’s learning patterns and recommend personalized study content. This way, each user gets a learning experience that fits their specific needs. In order to evaluate the system, we con-ducted user testing and analyzed student performance before and after using the app. We looked at metrics like academic improvement, task completion rates, and user satisfaction. The results showed that students became more focused, better at managing time, and more confident in their studies. Compared to existing apps like QANDA and SmartPal, our solution offers a more complete and personalized approach to learning. It combines AI, good design, and useful features to create a powerful tool for modern students
Authors - Serge Chaumette Abstract - Swarms of drones have been used for a few years in the military and civil domains, and their size has increased significantly. However, this has made them really challenging to manage. Among the issues, shape formation is one of the difficult operations to address. This increase in size is particularly significant for shows where large numbers (several hundreds to thousands) of drones are operated every day. These drones must obey precise flight plans to achieve the expected visual effects. Still, for these applications, there is no real need for dynamic shapes organization, since they can be computed offline prior to the mission. Shape formation is thus manageable. Even though military use cases are less visible to the general public, swarms of moderate yet significant sizes are used on theaters of operation, for instance, to achieve a saturation effect. There is little time in operation to prepare a mission, and the shape of the swarm is subject to changes that cannot be anticipated and that depend on the scenario and on its dynamic evolution, which is by nature unknown in advance. A process is thus required that can be initiated without major ground intervention (to ensure stealthiness and resilience), and that can be achieved in a collaborative manner within the swarms. In this paper, we address this issue of seamlessly organizing a swarm of drones in a given shape. We propose an innovative approach based on a DNA inspired code that circulates among the drones. This initial work raises a number of issues that relate to the nature of the target shape. These issues will be addressed in our future work.
Authors - Sudharsan Vasudevan, Venkata Durga Kavya Bhatta, Krishna Mohan Bathula Abstract - New York City is experiencing continued population growth, driven by both documented and undocumented immigration. With over 8.8 million residents and a population density exceeding 29,000 individuals per square mile, this surge has intensified pressure on the city’s housing and shelter systems. The resulting overcrowding has contributed to a decline in living conditions and a rise in homelessness, especially among vulnerable and underserved populations. One particularly affected group is runaway and homeless youth (RHY), who face heightened risks such as trauma, exploitation, and limited access to essential services like education and healthcare. This study explores trends within the New York City Department of Homeless Services (DHS) shelter system, analysing key factors that influence shelter occupancy, exit patterns, and disparities based on demographics, education, and race or ethnicity. The research applies data science and machine learning techniques to forecast occupancy trends and identify variables associated with successful exits from shelters. It examines the operational efficiency, accessibility, and equity of the shelter system to uncover systemic challenges and areas for improvement. This study aims to provide actionable insights by utilizing a data-driven approach that can support informed decision-making, enhance service delivery, and guide long-term policy development. The findings are intended to help optimize resource allocation and promote more effective housing strategies. Overall, this work seeks to contribute to sustainable solutions for reducing homelessness in New York City and improving outcomes for those experiencing housing instability.
Authors - Jude Hardee, Ivana Strumberger, Eva Tuba Abstract - In this paper, the effectiveness of computer vision on the classification of visually similar minerals was evaluated. A dataset was created with special consideration to visually similar mineral groups, and was partitioned into three subsets, each increasing in number of similarity groups and number of mineral classes. A series of transfer learning models were tested on this dataset and on these subsets, and the results of each were analyzed. Each model was evaluated by the value of it’s training and validation accuracy, and how the number of mineral classes and similarity groups impacted that accuracy. Confusion matrices for each model were then analyzed to evaluate the per-class accuracy for the full dataset. Based on these accuracies, the effects of visual similarity on computer vision accuracy was determined. In every case, the transfer learning model’s accuracy decreased linearly as the number of total mineral classes increased, indicating that computer vision may be ineffective in classifying the 200 rock forming minerals [5]. Full dataset average accuracy for these models varied between 68.85% and 35.58%. The two models which proved most effective were EfficientNetB0, which had the highest accuracy (68.85%) and little overfitting, and VGG16, which had an accuracy of 43.60% with effectively no overfitting. Lastly, across every model tested, accuracy varied significantly per class, but this variance showed no connection to visual similarity, indicating both that computer vision is not effective in the consistent classification of minerals, and that visual similarity has no negative impact on the efficacy of computer vision. This information holds great significance to the geological community, the computer science community, and industries such as mineral exploration, oil and gas, green energy solutions, and construction.
Authors: Shreya Joshi, Vijay Ukani, Priyank Thakkar, Mrudangi Thakker, Dhruvang Thakker Abstract: This study introduces a hybrid malware detection approach that combines machine learning (ML) and deep learning (DL) techniques to enhance detection accuracy. By applying models like K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Gradient Boosting, we achieved 99.47% accuracy during training and 99.21% accuracy during testing, focusing on analyzing Portable Executable (PE) header data. Additionally, incorporating Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks improved performance, achieving 99% accuracy, 97% precision, and 98% recall after 30 epochs. The proposed hybrid method reduces false positives and negatives while demonstrating scalability across various datasets, offering a reliable and efficient solution for contemporary malware detection.
