Cybersecurity attacks can arise from internal and external sources. The attacks perpetrated by internal sources are also referred to as insider threats. These are a cause of serious concern to organizations because of the significant damage that can be inflicted by malicious insiders. In this paper, we propose an approach for insider threat classification which is motivated by the effectiveness of pre-trained deep convolutional neural networks (DCNNs) for image classification. In the proposed approach, we extract features from usage patterns of insiders and represent these features as images. Hence, images are used to represent the resource access patterns of the employees within an organization. After construction of images, we use pre-trained DCNNs for anomaly detection, with the aim to identify malicious insiders. Random under sampling is used for reducing the class imbalance issue. The proposed approach is evaluated using the MobileNetV2, VGG19, and ResNet50 pre-trained models, and a benchmark dataset. Experimental results show that the proposed method is effective and outperforms other state-of-the-art methods.
19 February 2024 in Expert Systems with Applications
Most of the real-world networks exhibit community structure, a property that reveals the existence of natural vertex clusters whose inter-edge density is lower than intra-edge density between various groups. Despite providing a better understanding of network structure and characteristics, community detection has many practical applications in diverse domains. Communities obtained from the telephone network provides many useful information that can be used for churn prediction, budget control in organizations etc. Detecting communities is a fundamental need in the area of networks, yet challenging. In this paper, we propose an extension to the Girvan-Newman algorithm for finding the betweenness using the transitive closure property and the greedy technique in Dijkstra's single source shortest path method
22 March 2018 in Journal of Intelligent & Fuzzy Systems: Applications in Engineering and Technology
Computing the all pair shortest paths in a graph is a widely used solution, but a time-consuming process too. The popularly used conventional algorithms rely solely on the computing capability of the CPU, but fail to meet the demand of real-time processing and mostly do not scale well for larger data. In this paper, we propose the ex-FTCD (extending Full Transitive Closure with Dijkstra’s) algorithm for finding the all pair shortest path by merging the features of the greedy technique in Dijkstra’s single source shortest path method and the transitive closure property. Experiments show that the process improves computing speed and is more scalable. We re-designed the algorithm for the parallel execution and implemented it in mapreduce on Hadoop that supports the conventional map/reduce jobs. This work also includes the implementation on Spark that supports the in-memory computational capability which uses Random Access Memory for computations. The experiments show that the numbers of iterations are relatively small for even large networks.
In this paper, we propose a scalable method for finding the evolving communities in complex networks. The various network properties that are prominent in real-world networks are studied. The proposed algorithm computes the edge betweenness based on the transitive closure property combined with the greedy approach applied in Dijkstra's single source shortest path method. The major contribution is an improvement to GN algorithm in linear time for weighted undirected networks. The proposed algorithm is applied on Mapreduce to prove its scalability and enhance the performance in managing and analyzing the large scale networks. The experimentation of the distributed algorithm is tested on the private cluster and the results are as expected in the theoretical analysis.
The study investigates how adversarial training techniques can be used to introduce backdoors into deep learning models by an insider with privileged access to training data. The research demonstrates an insider-driven poison-label backdoor approach in which triggers are introduced into the training dataset. These triggers misclassify poisoned inputs while maintaining standard classification on clean data. An adversary can improve the stealth and effectiveness of such attacks by utilizing XAI techniques, which makes the detection of such attacks more difficult. The study uses publicly available datasets to evaluate the robustness of the deep learning models in this situation. Our experiments show that adversarial training considerably reduces backdoor attacks. These results are verified using various performance metrics, revealing model vulnerabilities and possible countermeasures. The findings demonstrate the importance of robust training techniques and effective adversarial defenses to improve the security of deep learning models against insider-driven backdoor attacks.