Elif Degirmenci

The adoption of deep learning has exposed significant vulnerabilities, especially to adversarial attacks that cause misclassifications through subtle small perturbations. Such attacks challenge security-critical applications. This study addresses these vulnerabilities by proposing a novel adversarial attack detection method leveraging data reconstruction errors. We evaluate this approach against three well-known adversarial attacks—Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), and Basic Iterative Method (BIM)—on Intrusion Detection Systems. Our method combines reconstruction error alongside aleatoric, epistemic, and entropy metrics to distinguish between original and adversarial samples. Experimental results show that our approach achieves a detection success rate of 92% to 100%, outperforming existing methods, particularly at low perturbation levels. This research enhances the robustness and reliability of machine learning models against adversarial threats by using effective error metrics in adversarial detection.

The emergence of robotic technologies has made a significant contribution in industry. Robot Operating System (ROS) is becoming a standard framework for industrial systems uses as a middleware system with many versions. However, the initial design of ROS does not include cyber-security concepts. The intense interest in robot systems, the security concerns and vulnerabilities of these systems have started to attract the attention of attackers. One of these attacks is DoS attack that targeting system availability by slowing down or crashing a service rather than obtaining the information or system. In this study, the impact of DoS attack has been analyzed in various scenarios for both in application and transport layer of the ROS middleware.In the experiments four different volume of DoS attacks are performed in five different experiment scenarios on ROS. To understand the impact of DoS attack, network traffics are monitored using Tshark. The resulting effects measured with some Quality of Service parameters that are delay and packet loss.

Robotic systems are widely used in industry, agriculture, the inspection of infrastructure, and even in our daily lives. The safety and security of robotic systems have become a primary concern as their interaction with humans increases. In this context, attacks on robotic systems have increased for diversified field applications. It is necessary to accurately detect these abnormal events in these systems as soon as possible. However, these systems also need a runtime verification approach on whether they conform to the established specifications. In this study, runtime verification for anomaly detection methods is proposed for the security of the robot operating system (ROS). Firstly, an anomaly detection method is proposed to detect unexpected situations, such as the number of the received packages being decreased under DoS attacks. Then, a holistic runtime verification architecture is proposed for the anomaly detection method. This architecture consists of three major entities: a verification device, an attacker device, and a robotic platform without losing generality. In the verification device, ROSMonitoring and Oracle are used to implement runtime verification. The proposed architecture is verified through an experimental setup. It is shown that the architecture can be used for runtime verification of different anomaly detection algorithms. A discussion on the security of robotic systems is also presented.

The increasing connectivity of smart systems opens up new security vulnerabilities. Since smart systems are used in various sectors such as healthcare, smart cities, and the intelligent industries, security becomes a fundamental concern. For this reason, security measurements are often added to security systems, such as authorization, authentication, encryption, Intrusion Prevention System (IPS) and Intrusion Detection Systems (IDS). However, these systems are still vulnerable to attacks such as DoS, DDoS. Recently, Deep Learning methods have been used to detect these attacks as early as possible. This method, however, has its own vulnerabilities to adversarial attacks. This paper presents three types of adversarial effects on the deep learning models analysis. In this study we use the open dataset CICIDS2017. Initially, DDoS attacks are detected using deep learning methods. Then, adversarial examples are created using adversarial attack methods. The untargeted adversarial attacks are used in this study, these being; FGSM, PGD and BIM. The results show that all adversarial attacks are effective on Deep Learning models. However, the results show that attacks from PGD and BIM are more effective. In addition, the deep learning models training is evaluated with K-fold cross-validation. The results show that, deep learning models that occur during each fold could be get different accuracy results against adversarial attacks. Adversarial attacks may be used in a K-fold cross validation process as a parameter for best model selection.

Adversarial attacks aim to deceive the target system. Recently, deep learning methods have become a target of adversarial attacks. Even small perturbations could lead to classification errors in deep learning models. In an intrusion detection system using deep learning method, adversarial attack can cause classification error and malicious traffic can be classified as benign. In this study, the effects of adversarial attacks on the accuracy of deep learning-based intrusion detection systems were examined. CICIDS2017 dataset was used to test the detection systems . At first, DDoS attacks weredetected using Autoencoder, MLP, AEMLP, DNN, AEDNN, CNN and AECNN methods. Then, the Fast Gradient Sign Method (FGSM) is used to perform adversarial attacks. Finally, the sensitivity of the methods against the adversarial attacks were examined. Our results showed that the classification performance of deep learning based detection methods decreased up to %17 after the adversarial attacks. The results obtained in this study form the basis for the validation and validation studies of learning-based intrusion detection systems.

With the novel developments in Wireless Sensor Network (WSN) technologies, environmental data collection and processing services are applied in diverse industrial and scientific areas. However, energy limitations and vulnerabilities of WSN nodes are still the main drawbacks of technological developments in the area. Understanding the energy utilization patterns of nodes helps to detect abnormal node behaviors and prevent malicious nodes. In this study, we observe the energy utilization behaviors of nodes and found that nodes have distinctive activity patterns based on their types. We also found that source, sink, and relay nodes on the data propagation path have higher energy consumption patterns compared to other nodes

Due to the failure-prone and vulnerable structure of WSN nodes, understanding the typical activity patterns of nodes helps to identify the faulty and malicious node activities, and to differentiate normal node behaviors from abnormal node behaviors. In this study, for understanding the typical node activities, we observe the effects of the network load on energy consumption, packet throughput, and latency parameters. We run simulations of networks with a variable number of sinks. We found that an increase in the number of sinks affects the overall network load and causes communication delays between the source and sink node.