Search

The growing complexity and volume of modern software have led to an increase in source code vulnerabilities, posing significant security risks. In response, deep learning-based automated source code vulnerability detection methods, particularly those utilizing source code similarity analysis, have recently emerged as promising solutions. However, existing similarity-based source code vulnerability detection methods frequently fail to fully utilize information from the hierarchical structure of source code and are often computationally expensive, limiting their practicality in real-world scenarios. In this paper, we introduce XTransformer, a novel deep learning-based source code vulnerability detector tailored for comparing target source code against archived vulnerable codes across various levels of the source code’s hierarchical structure by leveraging extra cross-attention imposed on the transformer architecture. Additionally, we propose a specialized training strategy based on supervised contrastive learning to improve XTransformer’s ability to effectively learn and differentiate between vulnerable and non-vulnerable source codes. Comprehensive experiments demonstrate that XTransformer outperforms current state-of-the-art methods across different datasets and code lengths while significantly reducing the inference time compared to other similarity-based methods that utilize hierarchical information from source code.

The rapid evolution of software development, propelled by competitive demands and the continuous integration of new features, frequently leads to inadvertent security oversights. Traditional security practices, often reactive in nature, primarily focus on identifying known vulnerabilities, creating a significant shortfall in detecting emergent, zero-day threats. This paper introduces CODE-SMASH, a novel deep learning-based source code vulnerability detector that utilizes a Siamese neural network with a hierarchical architecture integrating BiGRU and attention mechanisms. Our experiments using real-world datasets, specifically the Chromium and Debian datasets, demonstrate CODE-SMASH’s superiority over existing methods. It achieves significant improvements in detection performance across all key metrics, including accuracy, precision, recall, and F1-score, with average improvements of approximately 8.3%, 11.6%, 27.75%, and 17.7%, respectively, compared to the best-performing existing methods in our experiments. Moreover, CODE-SMASH shows its superior capability in handling complex and lengthy code sequences, with performance improvements for long-length code (60 to 80 lines) in F1 scores of 4.53 percentage points on the Chromium dataset and 5.62 percentage points on the Debian dataset compared to the second-best model’s performance. We believe our research makes a significant contribution to the field of automated vulnerability detection by providing a high-precision solution to the growing challenges in software security. Furthermore, based on our findings, we anticipate that future research could enhance CODE-SMASH by expanding its generalizability to various programming languages and reducing computational demands to improve efficiency.

Today’s cyber-attacks have become increasingly sophisticated and diverse, targeting systems that hold sensitive information, creating the need for continuous cyber exercise and skill development for cyber professionals. Because cyber exercises require training activities and environments that can support a variety of situations, significant technological efforts have been made to build training environments. In line with technological trends, current cyber exercise simulations are being studied to create various cyber scenarios that can help build an intelligent cyber battlefield using big data and artificial intelligence (AI). This requires a large amount and different types of data for learning, as well as a technical system that can manage and update them periodically. The objective of this study is to develop network topology generation and traffic prediction technologies based on intelligent network traffic analysis and AI models for cyber exercise technology systems. To automate training network scenarios, a path generation technology based on graph theory was developed, and the network environment was analyzed based on the amount of transmission by building a software-defined network capable of analyzing and predicting network traffic. A comparison of AI models such as long short-term memory (LSTM), bidirectional LSTM (BiLSTM), and gated recurrent units (GRU) to predict the amount of transmission showed good performance, with BiLSTM showing a better prediction error. The proposed methodology provides insights that can be used to adjust training scenarios during the network design and operation phases, which is expected to help manage the network, increase efficiency, and address security issues.

Terahertz (THz) wave imaging has potential features for medicine, such as cancer detection. Nevertheless, traditional lenses are heavy, bulky, low spatial resolution, and difficult to integrate. Recently, metasurface has emerged as a compelling approach for achieving lightweight, ultrathin, and easy integration. Although many THz metasurface lenses have been proposed for the wave-focusing applications, there are a few THz metalenses that are polarization insensitive and have a high spatial resolution for THz imaging. In this paper, an ultra-thin, planar, polarization-independent, and high numerical aperture metalens has been proposed using double split-ring resonator (DSRR) structures for high-resolution THz wave focusing. The combination method of the propagation phase (adjusting the diameter of the outer ring of the DSRR) and the geometric phase (changing the slit of the DSRR) is applied to build the unit cell library including eight DSRRs, which can cover the full phase from 0 to $2\pi $ and is independent with the polarization of the incident wave. By arranging the DSRRs into the concentric rings on a thin substrate, an ultrathin, polarization-insensitive, and high numerical aperture metalens is designed with a thickness of $0.55\lambda $ , a focal length of $660~\mu \text{m}$ ( $4\lambda$ ), radius of 4.678 mm ( $28.35~\lambda$ ), and numerical aperture of 0.99 that can work at 1.82 THz (wavelength ( $\lambda$ ) of $165~\mu \text{m}$ ). The designed metalens with a near-unity numerical aperture achieves a high spatial resolution of $89~\mu \text{m}$ ( $0.54\lambda$ ), which can produce high-quality images. It means that the proposed metalens can resolve the microscale features separated by a sub-wavelength distance ( $ < 90~\mu \text{m}$ ). Therefore, the suggested metalens can serve as an objective lens for the miniaturized microscopy, opening a new avenue for microscopic THz imaging and showing potential usage in tiny THz imaging systems for cancer detection applications.

With the proliferation of network systems, the boundaries between cyber and physical environments are blurring, leading to an increased risk of sophisticated cyber-attacks equipped with advanced technologies. In particular, as advancements in artificial intelligence through learning models have led to automated attacks and attack scenarios, countries are implementing cyber training and constructing training systems to respond to cyber security threats. This cyber training is based on existing cyber-attacks and conducted in virtual spaces similar to reality, generating network traffic through simulators and focusing on training for attack response and cyber resilience. However, the exponential increase in the number of network-based devices and the amount of network traffic they generate is leading to a gradual increase in threats to cyber security. In this study, first investigated the existing port number-based network traffic classification technologies and payload-based network traffic classification technologies to identify their shortcomings in the current network environment. We then categorized existing studies into supervised, unsupervised, and reinforcement learning to analyze the technology of classifying network traffic based on learning models as well as classification methods, procedures, performance standards, evaluation methods, quality of service/quality of experience, etc. Based on the analysis, presented limitations for application to training networks according to the learning method and suggested recommendations for establishing future research directions. Therefore, refining learning model-based network traffic classification technology will contribute to the construction of automated cyber training grounds such as cyber-attack–defense scenarios, network traffic anomaly detection, and maximizing cumulative rewards.

Monitoring construction site safety through physical observations is inherently flawed due to the complex and dynamic nature of construction sites. To overcome these challenges and enhance worker safety management, decentralized model training-assisted edge intelligence emerges as a promising solution. However, despite the potential benefits, our investigation reveals that no research for worker safety prediction has been grounded in the Federated Learning (FL) approach. In this context, we present a novel approach to worker safety prediction, leveraging FL in outdoor construction environments while preserving the privacy and security of sensitive data. Our methodology involves deploying sensor-based IoT devices at construction sites to collect highly granular spatial and temporal weather, building, and worker data. This data is then collaboratively utilized for training Deep Neural Network (DNN) models on the edge nodes in a cross-silos manner. To implement our approach, we establish a test-bed utilizing the EdgeX framework and constrained devices such as Raspberry Pi 4Bs, acting as edge nodes. Following the collaborative training, the resultant global model is deployed on participating nodes for edge inference, ensuring optimal network resource utilization and data privacy. The experimental results demonstrate the efficacy of the proposed approach in improving the utilization of construction safety management systems and reducing the risk of accidents and fatalities in the future. The outcome is a system that exhibits enhanced speed and responsiveness, a crucial aspect for time-sensitive applications such as the prediction of worker safety.

Water drilling machines are used to drill boreholes in the ground to extract groundwater. The resources required for water drilling vary from region to region due to underground water table depth and ground soil layer. Water drilling on a hard underground soil layer requires different resources than a soft underground. The proposed study facilitates the drilling industry by selecting the region with a soft land layer and increasing the penetration rate. Furthermore, the number of days and water table depth prediction allows the drilling industry to estimate the depth of the water table and time resources to reach the water table at different locations. The classification techniques classify the region based on the soil land layer. Regression techniques are used for predicting water table depth and number of days. The experiments are performed on a borehole log dataset provided by a research organization. This study used Support Vector Machine, TabNet, and Deep Tabular models to predict the land soil layer and compare the results with our proposed Ensemble Weighted Voting Soil Layer Classifier (EWV-SLC). The performance of the classification model is evaluated using accuracy, Precision, Recall, and F1 Score. The experimental finding shows that the EWV-SLC model performs better in accuracy and F1 score than other machine learning techniques. The performance of the regression model is evaluated using Mean Square Error (MSE), Mean Absolute Error (MAE),Root Mean Square Error (RMSE), and Mean Absolute Percentage Error(MAPE). In a days and water table depth prediction phase Support, Vector Regressor, Deep Neural Network, and TabNet Regressor are used, and compare the results with our proposed Ensemble Number of Days (E-NOD) and Ensemble Water Table depth (E-WTD) Regressor model. E-NOD and E-WTD models achieved less MAE, RMSE, and MSE than other machine learning methods.

During the last decade, substantial resources have been invested to exploit massive amounts of boreholes data collected through groundwater extraction. Furthermore, boreholes depth can be considered one of the crucial factors in digging borehole efficiency. Therefore, a new solution is needed to process and analyze boreholes data to monitor digging operations and identify the boreholes shortcomings. This research study presents a boreholes data analysis architecture based on data and predictive analysis models to improve borehole efficiency, underground safety verification, and risk evaluation. The proposed architecture aims to process and analyze borehole data based on different hydrogeological characteristics using data and predictive analytics to enhance underground safety verification and planning of borehole resources. The proposed architecture is developed based on two modules; descriptive data analysis and predictive analysis modules. The descriptive analysis aims to utilize data and clustering analysis techniques to process and extract hidden hydrogeological characteristics from borehole history data. The predictive analysis aims to develop a bi-directional long short-term memory (BD-LSTM) to predict the boreholes depth to minimize the cost and time of the digging operations. Furthermore, different performance measures are utilized to evaluate the performance of the proposed clustering and regression models. Moreover, our proposed BD-LSTM model is evaluated and compared with conventional machine learning (ML) regression models. The $R^{2}$ score of the proposed BD-LSTM is 0.989, which indicates that the proposed model accurately and precisely predicts boreholes depth compared to the conventional regression models. The experimental and comparative analysis results reveal the significance and effectiveness of the proposed borehole data analysis architecture. The experimental results will improve underground safety management and the efficiency of boreholes for future wells.

Drilling data for groundwater extraction incur changes over time due to variations in hydrogeological and weather conditions. At any time, if there is a need to deploy a change in drilling operations, drilling companies keep monitoring the time-series drilling data to make sure it is not introducing any changes or new errors. Therefore, a solution is needed to predict groundwater levels (GWL) and detect a change in boreholes data to improve drilling efficiency. The proposed study presents an ensemble GWL prediction (E-GWLP) model using boosting and bagging models based on stacking techniques to predict GWL for enhancing hydraulic resource management and planning. The proposed research study consists of two modules; descriptive analysis of boreholes data and GWL prediction model using ensemble model based on stacking. First, descriptive analysis techniques, such as correlation analysis and difference mechanisms, are applied to investigate boreholes log data for extracting underlying characteristics, which is critical for enhancing hydraulic resource management. Second, an ensemble prediction model is developed based on multiple hydrological patterns using robust machine learning (ML) techniques to predict GWL for enhancing drilling efficiency and water resource management. The architecture of the proposed ensemble model involves three boosting algorithms as base models (level-0) and a bagging algorithm as a meta-model that combines the base models predictions (level-1). The base models consist of the following boosting algorithms; eXtreme Gradient Boosting (XGBoost), AdaBoost, Gradient Boosting (GB). The meta-model includes Random Forest (RF) as a bagging algorithm referred to as a level-1 model. Furthermore, different evaluation metrics are used, including mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE), mean absolute percentage error (MAPE), and R2 score. The performance of the proposed E-GWLP model is compared with existing ensemble and baseline models. The experimental results reveal that the proposed model performed accurately in respect of MAE, MSE, and RMSE of 0.340, 0.564, and 0.751, respectively. The MAPE and R2 score of our proposed approach is 12.658 and 0.976, respectively, which signifies the importance of our work. Moreover, experimental results suggest that E-GWLP model is suitable for sustainable water resource management and improves reservoir engineering.

This paper proposes a method that can reliably monitor the adoption of existing technology by term frequency-inverse document frequency (TF-IDF) and K-means clustering using cited patents. TF-IDF and K-means clustering can extract patent information when the number of patents is sufficiently large. When the number of patents is too small for TF-IDF and K-means clustering to be reliable, the method considers patents that were cited by the originally set of patents. The mixed set of citing patents and cited patents is the new subject of analysis. As a case study, we have focused in agricultural tractor in which new technologies were adopted to achieve automated driving. TF-IDF and K-means clustering alone failed to monitor the adoption of new technology but the proposed method successfully monitored it. We anticipate that our method can ensure the reliability of patent monitoring even when the number of patents is small.

For the purposes of technology planning and research and development strategy development, we present a semi-automated method that extracts text information from patent data, uses natural language processing to extract the key technical information of the patent, and then visualizes this information in a matrix form. We tried to support qualitative analysis of patent contents by extracting functions, components, and contexts, which are the most important information about inventions. We validated the method by applying it to patent data related to nanosensors. The matrix can emphasize technical information that have not been exploited in patents, and thereby identify development opportunities.