This paper presents a transformative cross-modal framework that unifies audio and text modalities through the synergistic use of Convolutional Neural Networks (CNNs) for rich audio feature extraction and BERT (Bidirectional Encoder Representations from Transformers (BERT) for sophisticated, context-aware text embeddings. The proposed framework seamlessly projects both modalities into a shared embedding space, facilitating precise audio-text alignment and retrieval. Unlike traditional models that treat audio and text in isolation, this framework introduces a dynamic CNN architecture tailored to capture intricate audio patterns, while BERT delivers deep semantic understanding of text. Employing cosine similarity loss, the framework ensures optimal cross-modal alignment, achieving high retrieval accuracy and robust generalization across diverse datasets. Tested on the Free Music Archive (FMA) and Speech Commands datasets, the model demonstrated a remarkable ability to cluster and align related audio and text samples with mean similarity scores of 0.82, and achieved cross-modal retrieval accuracy of 0.67, even when applied to domains outside its training scope. This innovative framework offers broad potential for applications such as music information retrieval, speech recognition, and multimedia content analysis, pushing the boundaries of multimodal learning and retrieval systems.
Software vulnerabilities provide openings for cyberattacks and therefore pose a great risk to the security of an IT system. The availability of public exploit scripts strongly influences the scale of the threat as it greatly reduces the effort that adversaries must expend to exploit these vulnerabilities. These exploit scripts also differ relative to maturity, reliability, and skill required to be executed successfully. This factor plays an important role in qualitative and quantitative vulnerability risk assessment, cyber adversary analysis, and cybersecurity training. However, not all public exploit databases offer such a rank, which impairs the accuracy of these evaluations. To address this shortcoming, this study evaluates different machine learning approaches to predict the complexity required to execute an exploit based on the exploit script. For this evaluation, different data preparation approaches and machine learning models, all of which have been proven to perform well on similar tasks, were selected. Each machine learning model was then trained using each data preparation approach to evaluate performance and determine the best approach to reliably predict the required skill to use an exploit. These results can be used to generate complexity predictions for exploits, which can then be included in skill-based assessments and ultimately improve the accuracy of cybersecurity evaluations.
This paper proposes a cutting-edge vulnerability detection method for smart contracts, combining Transformer-based embeddings and Graph Neural Networks (GNNs). Critical opcodes are identified and dynamically weighted using attention mechanisms, enhancing feature representation. The GNN captures both relational patterns and critical opcode characteristics, enabling robust detection of vulnerabilities. Experimental results show significant improvements in F1-scores for both binary and multi-class detection tasks, outperforming traditional models like LightGBM. This approach leverages modern AI advancements to address challenges in accuracy and generalization, providing a scalable and effective solution for smart contract analysis.
Precise segmentation of brain tumors is essential in medical imaging, supporting early diagnosis, personalized treatment planning, and effective monitoring of tumor development and response to therapies. While centralized deep learning models like U-Net are highly effective for this task, they raise substantial privacy concerns when healthcare or-ganizations need to exchange confidential patient information. To address these concerns, the BrainDiffU-Net framework is introduced, a diffusion-based U-Net model designed for decen-tralized learning environments. In this framework, each medical institution independently trains a model on its local dataset and exchanges model updates with neighboring institutions, ensuring data privacy without requiring a central server. The diffusion process allows model parameters to be shared securely across the network, facilitating collaborative learning across multiple institutions while preserving privacy. Simulations conducted on the Low-Grade Glioma (LGG) MRI Segmentation Dataset demonstrate that BrainDiffU-Net achieves segmentation performance comparable to centralized models and significantly outperforms non-cooperative strategies in terms of average loss and Dice coefficient. These results show that BrainDiffU-Net provides an optimal balance between segmentation accuracy and privacy preservation, making it a viable solution for decen-tralized medical applications. The source code is available at https://github.com/Parth-nXp/BrainDiffU-Net.
User satisfaction in Wireless Networks (WLANs) heavily rely on the efficient provision of Quality of Service (QoS). Nevertheless, misbehaving acts by malicious users have the potential to disrupt the QoS provisioning and impair its overall efficiency. One such abusive behavior is called Class Hijacking (CH) and has the ability to deteriorate the QoS-levels received by the well-behaving users in the network. CH is a hard to detect, yet easy to perform attack, targeting the QoS mechanism in differentiated services networks such as WLANs. This article attempts to illustrate the negative impact of CH actions on the network performance of an infrstructure-based WLAN where stealing network resources is an attractive choice for nefarious users. This is achieved through extensive simulations using the ns-3 network simulator. Applying a gradual increase in the number of deceitful users, we observe how these attacks noticeably deteriorate network performance.
Wireless Body Area Networks (WBANs) have emerged as a promising technology in the field of healthcare monitoring. The aim target of the paper is to design an Ultra-Wide Band (UWB) (3.1 GHz -10.6 GHz) antenna for WBAN. A Micro-Strip Patch Antenna (MSPA) with MS line feed method is used. The antenna is tested on different substrates; FR4, Rogers, and Jeans. The simulation results confirm that jeans fabric substrate (epsilon(r) = 1.4)shows the best performance in terms of flexibility, comfort, and efficiency. The antenna is then tested on flat and bended human tissue models. The CST MW Studio software is used for the simulations. The Specific Absorption Rate (SAR) values on the simulated human tissue (flat-bended) is within the standard safe limit. The textile antenna is designed and examined on a human body. A fair agreement is achieved between measured and simulation results. The designed antenna is promising for telemedicine systems.
The retail landscape has significantly transformed due to the impacts of war and the Covid-19 pandemic, which have altered consumer preferences and shopping behaviors. During war, consumer boycotts and changing demand patterns emerge, while the pandemic has shifted traffic away from large hyperstores to smaller, compact stores due to health concerns. This study develops an automated model to address space limitations in compact stores, using machine learning (ML) and deep learning (DL) algorithms. The model combines clustering and classification techniques, utilizing methods such as K-means, k-Nearest Neighbor (KNN), Recurrent Neural Networks (RNNs), DBSCAN, and Gaussian, Agglomerative, and Birch clustering to generate assortments tailored to store characteristics like size and segment. The study finds that K-means clustering is highly accurate, with an SSE score of 0.93, followed by KNN for classification. Additionally, regression models, including Multiple Linear Regression (MLR), Lasso, Random Forest Regression (RFR), Support Vector Regression (SVR), and Decision Tree Regression (DTR), alongside a Deep Neural Network (DNN), analyze sales factors to enhance prediction accuracy. The DNN effectively captures complex relationships between sales data inputs, with R- squared values of 0.81 for MLR and 0.87 for DNN. The model was applied to the grocery department, resulting in a 3% sales increase in compact stores. An ongoing survey with 12 merchandise experts is evaluating the model’s impact on category sales behavior. Overall, the model has improved customer experience by making products more accessible, driving increased foot traffic, visit frequency, and profitability.
Signing documents using electronic signature services can pose security risks due to problems pertaining to signature tracing, forging, and identity verification. These services also rely on centralized networks, which require that the signing parties trust the provider’s ability to keep their documents and sensitive information safe. However, the shift to digital contracting remains essential, as the traditional contracting process is often complex and inefficient. This work proposes a distributed network model that allows any two parties to sign a legally binding agreement using verifiable digital signatures. This eliminates the need for third-party mediation, which results in a more cost-effective and time-efficient contracting process. Integrating a hybrid encryption scheme and full client-side processing in this model ensures the confidentiality of the contract and the privacy of the parties involved. This, in addition to the model’s emphasis on minimizing traffic and power consumption, results in a highly scalable and robust ledger, with the potential for integration with blockchain or other Distributed Ledger Technologies (DLTs). In conclusion, this work is a step forward towards the adoption of digital contracts and their integration with current legal frameworks. Eventually, it could aid in clearing the confusion that surrounds smart contract technology.
This paper explores the impact of dynamic entropy tuning in Reinforcement Learning (RL) algorithms that train a stochastic policy. Its performance is compared against algorithms that train a deterministic one. Stochastic policies optimize a probability distribution over actions to maximize rewards, while deterministic policies select a single deterministic action per state. The effect of training a stochastic policy with both static entropy and dynamic entropy and then executing deterministic actions to control the quadcopter is explored. It is then compared against training a deterministic policy and executing deterministic actions. For the purpose of this research, the Soft Actor-Critic (SAC) algorithm was chosen for the stochastic algorithm while the Twin Delayed Deep Deterministic Policy Gradient (TD3) was chosen for the deterministic algorithm. The training and simulation results show the positive effect the dynamic entropy tuning has on controlling the quadcopter by preventing catastrophic forgetting and improving exploration efficiency.
This paper presents a novel approach for diagnosing neuromuscular disorders, specifically Myopathy and amyotrophic lateral sclerosis (ALS), from electromyography (EMG) signals using a data augmentation-enhanced convolutional neural network (CNN) with an integrated attention mechanism. Given the limited availability and variability of labeled EMG data, we employed multiple data augmentation techniques including noise addition, time warping, scaling, magnitude warping, and jittering, to expand the dataset and create a more robust model training process. Each augmentation method was evaluated through distinct multiclass classification tasks using a CNN model, enhanced with attention blocks that focus on the most relevant temporal and spatial features within the EMG signals. Our results demonstrated that the CNN model, with attention mechanisms, achieved high classification accuracy of 98.49% using jittering technique, showcasing the effectiveness of our approach in improving the early diagnosis of neuromuscular disorders.
Integrating ontologies in tourism applications represents a transformative step toward achieving a structured digital transformation to enhance the competitive edge of tourism destinations. In Morocco’s tourism sector, particularly in the culturally and naturally rich Draa-Tafilalet region, implementing an ontology is essential to organize diverse tourism-related data and provide a more personalized and accessible visitor experience. Despite Morocco’s significant efforts, the tourism industry continues to face persistent challenges in delivering customized and accessible information aligned with tourists' preferences and needs. To address this issue, this study develops comprehensive ontology using a semi-automated methodology that combines natural language processing, language models, and expert validation. The ontology encapsulates attractions, services, and visitor preferences specific to the region. The main objective is to modernize and enhance the informational structure of the tourism sector, facilitating better navigation for tourists and supporting the strategic promotion of the region’s assets.
In this paper, we investigate intrusion detection within IoT networks. We employ the Support Vector Machine (SVM) algorithm and utilize the BoTNeTIoT-L01 dataset as a training dataset. It comprises traffic data from nine IoT devices. The traffic were monitored in a local network during botnet attacks, carried out by Mirai and Gafgyt. We extracted twenty-three features from the packet capture (PCAP) files and applied statistical methods to enhance the detection performance of the SVM model. Our simulation results shows that SVM algorithm effectively identifies malicious traffic with an overall accuracy of 99.17%.
Throughout the past years, Egypt has been one of the main producers of dates around the globe, with the trees covering a huge portion of Egypt from Aswan to the north Delta. However, the red palm weevil insect has recently become a great threat to these trees. If its infestation is not detected early, it can feed on the tree’s fibers, causing harmful damage, which could lead to the tree’s death. That is why detecting its infestation early is very important. In recent years, IoT has grown to play a significant role in many areas, and the agriculture sector is no exception; smart agriculture methods can address such infestation issues. This paper focuses on building an acoustic device that can listen to sounds inside the palm trees and extract features from the sound using MFCC (Mel-Frequency Cepstral Coefficients), a feature extraction technique used in audio signal processing and analysis. The device then detects whether an infestation is present, achieving an accuracy of 99.02%. Another target was making the device able to run 24/7 without requiring human intervention but by using renewable solar energy. Moreover, a mobile application was developed to help farmers monitor their trees remotely, empowering them to detect infestations at an early stage, allowing them to prevent further spread and minimize the damage caused by the weevil.
This study examines the quality differences between AI-generated and human-generated code through an evaluation of multiple software quality metrics, including maintainability, complexity, and documentation. Using a dataset of 5,312 code samples—2,700 human-generated and 2,612 AI-generated—we applied machine learning techniques to classify and analyze the code based on these metrics. The results revealed that AI-generated code tends to excel in maintainability and documentation, demonstrating higher maintainability index scores and a higher ratio of comments. Additionally, AI-generated code often features simpler control structures, reflected in its lower cyclomatic complexity. In contrast, human-generated code showcased greater adaptability and flexibility, particularly in addressing complex problem statements. A neural network classifier achieved 88.05% accuracy in distinguishing between the two code origins, with comments ratio, maintainability index, and cyclomatic complexity being the most significant differentiators. These findings highlight the complementary roles of AI and human contributions in software development, suggesting strategic integration of both for enhanced efficiency and quality.
This paper investigates the use of Probabilistic Principal Component Analysis (PPCA) in enhancing algorithmic trading strategies, particularly for latent feature extraction in Mean Reversion trading. The motivation behind this work lies in the need to uncover hidden alphas that traditional methods fail to expose, especially in high-dimensional financial datasets. We propose a novel integration of PPCA with Mean-Variance Optimization to improve portfolio performance while addressing Survivorship Bias through Missing Value Imputation. The contribution of this work lies in validating PPCA's ability to extract latent information from financial data and mitigate biases introduced by delisted stocks, thereby enhancing the robustness of trading algorithms. Our backtest results demonstrate the potential impact of this approach on strategic investment decision-making.
Using hydrogen instead of natural gas in pipelines is considered as an alternative energy source. Therefore, it is proposed in this paper to transport hydrogen using one of the new methods of transporting Natural Gas, "dense phase". As it was possible to transport natural gas in the dense phase, the purpose of the study is to know the possibility of transporting hydrogen in the dense phase as well. In a dense phase, gas density remains as high as a liquid, whereas its viscosity is very low, closer to that of a gas. In the present study, the governing equations and conditions are presented in a dimensionless form to define the influential parameters describing flow behavior. The numerical solution of these equations is implemented with appropriate boundary conditions using ProMax software for two cases of Hydrogen to provide user-friendly alternatives and offer a wide range of simulation patterns based on selected sets of equations and components. It was also used for comparison and to ensure it produces the same results as previous studies. Velocity inside the pipes is kept appropriately low to prevent corrosion. Velocity, pressure drop, temperature, density, viscosity, and liquid holdup results are obtained based on the influential dimensionless parameters. The results is then validated with previous studies. Despite its density high value decreases, Hydrogen in the dense phase was found to be in liquid form due to its low critical temperature and pressure. Accordingly, the transportation of hydrogen is not recommended.
Data plays a crucial role in computer science, serving as its backbone in various applications. Understanding and interacting with data stored in databases often requires technical knowledge of query languages, posing a challenge for non-technical users. This paper aims to bridge this gap by developing a Text-to-SQL generation system, facilitating natural language queries to database operations. In SQLIFYME, we adopted a dual-path approach to implement the Text-to-SQL generation model. The Seq2Seq model, utilizing transformers, processes natural language queries to generate corresponding Structured Query Language (SQL) queries. Simultaneously, generative AI techniques are employed, applying prompt engineering to the Llama2 LLM to manage complex and nuanced queries and further refine and enhance query generation capabilities. After extensive experiments with Seq2Seq models, LLMs, and different prompt engineering techniques, we reached the optimal configuration for our proposed model. The experimental results demonstrate the effectiveness of both approaches in accurately translating natural language queries into SQL commands. The Seq2Seq model achieves high accuracy in handling structured queries, while the generative AI approach excels in handling complex and nuanced queries, highlighting the versatility of our system. Notably, the system achieved an impressive execution score of 0.8 using the T-5 base model.
In this paper, an integrated school management system is introduced to enhance educational interactions by connecting students and teachers. Moreover, it simplifies the sharing and retrieval of academic resources in pdf format. The primary objectives include creating a user-friendly web-based platform for improved student-teacher connectivity, supporting multiple languages (starting with English and Arabic), and providing introductory videos for new users. In addition, an attendance tracking system for student safety is implemented. Furthermore, key features of the system include user registration and profile management, secure login and authentication, a user-friendly dashboard, messaging tools for direct communication, and scheduling, pdf file upload with effective organization and search options, real-time notifications, and robust data security and privacy measures are introduced. The HTML5, CSS3, JavaScript, Tailwind for the front, Node.js technology is introduced for the backend, MongoDB for the database, JWT for authentication, and cloud-based storage.
Cryptocurrency-related crimes peaked in 2021, and current trends indicate the enduring presence of blockchain technology. With transaction volumes projected to rise, manual investigations into criminal activities will soon become impractical. Consequently, anomaly detection systems are essential for upholding the financial integrity of blockchain networks. Despite the security measures implemented through consensus mechanisms, blockchain platforms remain vulnerable to illicit behavior. Leveraging machine learning presents a promising strategy for mitigating these risks. This study undertakes a thorough comparison of various machine-learning algorithms aimed at detecting suspicious transactions within blockchain networks. The evaluated techniques encompass Support Vector Machine, Decision Tree, Logistic Regression, K-means, Random Forest, XGBoost, Local Outlier Factor, Light Gradient, and Isolation Forest.
This paper describes an efficient framework for querying RDF (Resource Description Framework) graphs, which contain billions of labeled entities, using simplified SPARQL queries. Due to the schema-free nature of RDF data, it is challenging for users to understand the underlying structure and create complex queries. The paper proposes a solution that extends simplified queries using knowledge semantics to retrieve approximate answers. The framework mines RDF graphs for semantically equivalent patterns, known as topic graphs, by using large language model (LLM) embeddings to generate semantic vectors. It then constructs approximate queries to retrieve top-k results based on semantic similarity. Extensive tests on the DBpedia dataset and QALD-4 benchmark demonstrate the effectiveness and efficiency of the approach.
