Marvy Badr Monir Mansour

The deployment of deep-learning vulnerability detectors in production remains difficult. Models are large, false-positive rates are high, output is opaque, and a persistent gap separates benchmark performance from real-world utility. The objective of this work is to close part of that gap by combining a learned detector with interpretable rules so that accuracy, efficiency, and actionability are achieved together. A hybrid framework is therefore presented in which a lightweight edge-conditioned GNN of 71,810 parameters, named FastVulnGNN, trained in 96.2 s on a single CPU core, is paired with rule-based heuristic detection for six C/C++ vulnerability classes, namely buffer overflows, format-string defects, null-pointer dereferences, double-free errors, integer overflows, and race conditions. On the MegaVul dataset, an accuracy of 71.1%, an F1 score of 0.70, and an AUC-ROC of 0.77 are obtained by the GNN component. On a production codebase of 499 files and 312,758 lines of code, the full hybrid scan completes in 5.5 s, which corresponds to about 57,000 lines per second, without any GPU hardware. Per-file risk tiers and pattern-level explanations are produced, and these are suitable for continuous-integration use. The significance of this work lies in demonstrating that a deployable, explainable detector can be assembled from compact components, and an edge-type ablation study, a cross-dataset evaluation, and a per-vulnerability analysis are reported to characterize the approach.

Software vulnerabilities are a primary cause of security breaches, and their automated detection at scale has therefore become a pressing concern for both industry and academia. Most Graph Neural Network (GNN) approaches to vulnerability detection treat code graphs as homogeneous structures, and the semantic distinctions between Abstract Syntax Tree (AST) edges, Control-Flow Graph (CFG) edges, and data-flow dependency edges are consequently discarded. The main objective of this study is to determine whether explicitly conditioning message passing on edge type yields accurate yet lightweight detection. To this end, an edge-conditioned GNN named FastVulnGNN is proposed, in which the message-passing computation is conditioned on edge-type information drawn from Code Property Graphs (CPGs). FastVulnGNN operates on Joern-produced CPGs that contain 33 node types and 21 edge types, so that the full heterogeneous graph structure is preserved. A multi-scale readout mechanism that combines mean, maximum, and learned attention pooling is employed for graph-level classification, and the training configuration, which combines focal loss, label smoothing, and cosine annealing warm restarts, is individually validated by an ablation of the training objective. On the MegaVul dataset of 1904 balanced C/C++ samples, an accuracy of 71.1%, an F1 score of 0.70, and an AUC-ROC of 0.77 are achieved with only 71,810 parameters. Training completes in under two minutes on a single CPU core, and no GPU resources are required. The significance of this work lies in its demonstration that a compact, edge-aware architecture can match independently reproduced results of far larger models while remaining deployable in resource-constrained settings, such as continuous-integration pipelines and developer workstations. This study is deliberately framed as a controlled and reproducible engineering and evaluation contribution rather than as an architectural advance. An edge-type ablation study, a cross-dataset evaluation, and a per-vulnerability analysis are additionally reported to characterize the behavior and limitations of the model.

Software defect prediction using machine learning models is widely adopted to identify defects before release and to guide testing efforts. However, the black-box nature of these models limits their trustworthiness and usability among developers. To address this issue, recent research has focused on providing explanations for model predictions, helping developers understand why a prediction was made and increasing their confidence in the results. Despite this progress, current explanation techniques for software defects are often computationally intensive and produce complex, overwhelming outputs that are difficult for developers to interpret and act upon. To overcome these challenges, this paper proposes a lightweight, developer-friendly, rule-based explanation strategy that combines the strengths of two popular local explanation methods: LIME and Anchor. This is done by extracting the intersection of conditions common between LIME and Anchor. The proposed approach generates concise, high-confidence explanations that preserve fidelity while enhancing interpretability. We evaluated our approach on 1,000 defective instances from the NASA PROMISE Ant dataset. Compared to standalone LIME and Anchor outputs, the reduced explanations provided more actionable insights with fewer and more consistent conditions. Overall, the proposed method offers a practical and scalable path toward more interpretable and developer-friendly defect prediction.

Today, Vehicular Networks (VNs) can significantly improve the safety and sustainable growth of Intelligent Transportation Systems (ITSs) for smart cities. VNs offer efficient solutions to persistent urban problems such as congested roads and roadway safety. VNs adopt a wide range of technologies such as Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), and Vehicle-to-Everything (V2X) communications. These technologies allow cars to communicate with each other and with road infrastructure such as Roadside Units (RSUs) in real time, as well as enable the sharing of data. This can effectively mitigate accidents, enhance traffic flow, and minimize air pollution. Our paper extensively explores the state-of-the-art VN technologies used to establish safe and sustainable transportation systems. Additionally, our study highlights the importance of connectivity among automobiles for smarter urban environments to enhance transportation and support environmentally friendly growth. This is done by utilizing the Veins simulator in our work for conducting different mobility and traffic simulations. We formulate various scenarios, where we use the vehicles’ speed for our performance evaluation. Furthermore, we use OMNeT++ and SUMO for the accurate simulation of traffic motion in the city of Cairo, where results demonstrate that the speed of vehicles employing Vehicular Ad-Hoc Networks (VANETs) communications is not affected by traffic congestion.

Autonomous driving systems (ADSs) hold promise for enhancing safety and efficiency on the roads; yet, concerns persist due to rising fatalities involving vehicles equipped with ADSs. This research comprehensively examines the technical components of ADSs, including current challenges, system designs, evolving techniques, and critical features like sensor technologies such as Light Detection and Ranging (LiDAR) and cameras. These sensors enable vehicles to perceive their environment accurately, facilitating tasks such as navigation and obstacle avoidance. Advanced edge detection strategies for lane detection and the usage of Lane Keeping Assist (LKA) structures are crucial technologies for ADS. Hence, in this paper, we implement a modified Sobel edge detection algorithm to improve its performance for lane detection and integrate a CNN-based approach into our system. By trying various Gaussian filter parameters, we develop an optimized edge detection system that performs well in different lighting and weather conditions, such as low light or rainy weather. In our work, we implement a Convolutional Neural Network (CNN) for edge detection and train it using a comprehensive dataset of road images and traffic scenes. The dataset includes a diverse range of conditions, such as different lighting (day and night), weather (clear, rainy, foggy), and road types (highways, urban streets, rural roads). This extensive dataset allows the CNN to learn features robustly and generalize well across various driving scenarios. Simulation and results show that our CNN-based approach has high performance, as it exhibits high accuracy and low processing time needed for ADSs.

Marvy Badr Monir Mansour, Amr Ayman, and Marwan Yehia. International Journal of Fuzzy Logic and Intelligent Systems 2024;24:295-305. https://doi.org/10.5391/IJFIS.2024.24.3.295

In this paper, a novel image encryption scheme is presented that leverages a combination of hyperchaotic systems, Singular Value Decomposition (SVD), RC5 operations, permutation techniques, and a custom S-box generated via the XORshift algorithm. Multiple satellite images are merged into a single augmented image prior to encryption by this algorithm, effectively protecting against traffic analysis attacks that could compromise individual encrypted images. Once the augmented image is created, it is split into RGB color channels, and each channel undergoes a four-stage encryption process. The encryption process is initiated by the generation of a chaotic sequence from a six-dimensional (6D) hyperchaotic system. This initial key sequence is then manipulated and transformed into a new key using SVD. The transformed key sequence is utilized to XOR the original image, providing an initial layer of encryption. Subsequently, the image is divided into blocks, and RC5 operations are applied to each block to enhance the security of the encryption process. Following this, a permutation operation is performed on the image using the chaotic sequence generated earlier. Finally, an S-box, created from a random sequence generated by the XORshift algorithm, is applied to the image to further obscure the pixel values. The robustness of the algorithm is validated by numerical tests, demonstrating its effectiveness against both statistical, differential attacks and brute-force attacks, making it extremely difficult for third parties, such as cloud service providers, to identify any relational patterns between the encrypted satellite images. Specific numerical values that demonstrated an enhanced security level include a key space of 27016, entropy exceeding 7.999, and a cross-correlation of ≈ 0. This encryption method not only strengthens the protection of satellite imagery against unauthorized access but also ensures the integrity and confidentiality of the images, which are crucial for applications ranging from national security to environmental monitoring in an increasingly data-driven world.