Adversarial attacks, widely recognized as significant threats to machine learning (ML) models in computer vision and natural language processing, can have more severe consequences when targeting ML-based Network Intrusion Detection Systems (NIDS). These attacks, characterized by data manipulation, necessitate a focused investigation grounded in the unique attributes of the data and practical constraints inherent to the target scenario, as opposed to indiscriminately applying methodologies borrowed from other domains. Since network traffic is complex unstructured data, ML models are commonly used in existing studies to explore how perturbations can defeat ML-based IDS. However, two challenges persist in the realm of traffic-space adversarial attack generation. First, raw traffic data cannot be directly input into ML models. Second, determining the appropriate perturbation scale and direction is challenging, particularly in the case of multi-class NIDS. In this work, we propose a projection-based adversarial attack generation framework, ProGen, to address these two challenges. ProGen is inspired by two observed characteristics of the NIDS scenario: flexible representation and clear objective. ProGen uses a basic feature sequence (BFS) space to represent network traffic in a way that aligns with realistic requirements. To achieve a clear objective, ProGen utilizes a traffic space generative adversarial network (GAN) to approximate distribution mapping between malicious traffic and benign traffic. To better apply the generative model for adversarial attacks, we further design constraints to preserve the functions of the adversarial traffic. We’ve successfully demonstrated the effectiveness of ProGen on six common ML models using the CSE-CIC-IDS2018, CIC-IDS-2017, and UNSW-NB15 datasets; however, we’re yet to validate these findings in real network environments. We visualize the generated distributions of the BFS elements to illustrate the projecting effect under the designed realistic constraints. The results of attack effectiveness tests show that attacks generated from ProGen can significantly reduce the detection performance across different ML models.

Developing early and transferable Network Intrusion Detection Systems (NIDSs) is essential for robust network security. Early detection prevents further damage, while transferable NIDSs enables reuse across diverse networks. For Machine Learning (ML) and Deep Learning (DL)-based NIDSs, transferability significantly reduces data collection and annotation costs for timely attack mitigation. Current DL-based early intrusion detection studies often focus on identifying attacks from the first few packets, neglecting the crucial aspect of adjustable early detection. Additionally, most DL-based NIDS methods overlook transferability during both design and evaluation phases. To address these limitations, we propose K-GetNID, a knowledge-guided graph learning-based NIDS that excels in both early and transferability. We introduce a Heterogeneous Temporal Graph (HTGraph) to represent the dynamic feature series of network flows, providing enough information for early detection. Additionally, we construct this HTGraph format based on prior knowledge about feature types and correlations to assist the neural network in learning general and transferable knowledge for intrusion detection. We develop a corresponding Heterogeneous Temporal Graph Neural Network (HTGNN) model to learn from the HTGraph format. Furthermore, an Adjustable Early Detection Decoder is designed to enhance the generalization of the proposed model to the input distribution shifts caused by early detection. Experiments on CIC-IDS-2017 and UNSW-NB15 datasets show that K-GetNID matches the performance of deep learning methods, excelling in adjustable early intrusion detection and transferability.

Children with autism spectrum disorder (ASD) often require long-term intervention. Experienced therapists with better observation on children’s feedback and interaction could adjust their intervention strategies to achieve the responsiveness and dynamic adaptation. However, there is large unmet need of experienced therapists to provide high quality therapy. Most children’s intervention was run by less experienced therapists or parents lacking observation skills. This motivates us to provide some observation assistance for improving the effectiveness of intervention of ASD. In this paper, we present an Observational Therapy-Assistance Neural Network (OTA-NN) for bridging the observation ability gap between experienced therapists and unprepared therapists (include parents of children). In OTA-NN, an attention based neural network extracts features from children’s behavior data and a multiple instance learning (MIL) network predicts score for reflecting child’s training state during a therapy session. We adopt weakly-supervised learning to train our model. The experiments on DREAM dataset show our model can successfully distinguish children’s unpredictable behaviors under different training states.

Named data networking (NDN) recently arises as a promising networking paradigm to support the Internet of Things (IoT) due to its data-centric architecture. However, NDN integrates application-layer semantics into the packet forwarding plane, which presents new attack faces. In this article, we aim to handle two attacks that exploit such vulnerabilities, namely the blackhole attack and the content poisoning attack. The two attacks are not handled efficiently by existing approaches due to the challenge in minimizing routers that need to be detoured to isolate attackers. Therefore, in this article, we propose a novel method named SmartDetour to tackle the challenge in a distributed manner. SmartDetour contains two components: 1) a proactive reputation updating algorithm and 2) a reputation-based probabilistic forwarding strategy. The former updates the reputation of forwarding candidates based on whether they must be detoured upon packet failures. The latter selects the next-hop router for interest packets probabilistically based on the reputations of forwarding candidates. The two components work together to isolate attackers with minimal detouring needed. Extensive ndnSIM-based simulation shows that SmartDetour can effectively identify and isolate attackers.

The full-buffer Shrew (FB-Shrew) denial of service (DoS) attack is a variant of the classic Shrew attack that exploits the congestion control mechanism of transmission control protocol (TCP). Here, an attacker sends a high-rate burst of attack packets only after the router buffer is filled with TCP packets, causing the router to drop legitimate packets, and forcing the retransmission of TCP packets. As such, an FB-Shrew attack can cause maximum damage with minimum resources. In this paper, we challenge an assumption of constant round trip time adopted in the original FB-Shrew model. As a result, this model fails to achieve its expected attack effect. In response, we analyze the TCP congestion window and queue behaviors to develop two low-high burst models for maximizing the potency of the FB-Shrew attack. Model 1 is designed to achieve the attack effect expected of the original model. Then, the attack potency of Model 1 is enhanced by simply adjusting the starting time of the attack burst to form Model 2. Mode 1 only exploits the retransmission timeout (RTO) mechanism. Model 2 takes advantage of both the RTO mechanism and the fast retransmission mechanism. In this way, Model 2 further slows down the growth of the congestion window and extends the attack period. A combination of theoretical analyses and simulations are adopted to first validate the proper functioning and effectiveness of the two models for a standard network configuration, and then we assess their attack performances with variations in different network parameters. Our performance assessment demonstrates that one attack unit of Model 2 damages almost twice the number of TCP units as one attack unit of Model 1, which represents an increase in attack potency of nearly 200 percent. The present study provides an expanded basis to explore FB-Shrew attack patterns that may be utilized by attackers. Moreover, the damage that could be inflicted by such attack and the extent to which defense strategies are capable of mitigating the attack's impact could be assessed more precisely by defenders.

A TCP-targeted low-rate denial of service (LDoS) attack exploits the vulnerabilities of TCP congestion control mechanism. The most widely used TCP congestion control algorithm, CUBIC, increases the resilience to LDoS. This paper explores high-potency patterns of LDoS attacks against CUBIC TCP under the RED queue management scenario, and develops two attack models, the D- and S-models to maximize attack potency (i.e., the damage-to-cost ratio). Theoretical analyses and extensive experiments are conducted to validate the proper function of the models and evaluate their performance. Test results show that the models can effectively throttle CUBIC TCP throughput. Under standard-configured network parameters, one attack unit can damage up to about 21 and 26 TCP units for the D- and S-models, respectively, which represents an increase in attack potency about 20%. The attack potencies of our proposed models are at least 250% greater than that of the traditional attack model. In addition, with variations in different network parameters, these two models are still efficient and alternatively maximize the attack potency. Finally, attack countermeasures are outlined. The present study offers a basis to explore new attack manners which may be exploited by attackers and inspires researchers to develop new measurements against such attack.