Kexin Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kexin Zhang has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Jinhu Lü
Beihang University
4 shared publications - Qing Gao
School of Automation Science and Electrical Engineering, Beihang University
4 shared publications - Maciej Ogorzałek
Jagiellonian University
2 shared publications - Yue Deng
Beihang University, Beijing, China
2 shared publications - Bing Zhao
School of Automation Science and Electrical Engineering, the Embodied Intelligence Robotics Institute, Beihang University, Beijing, China
1 shared publication - Fang Zhou
Hangzhou Innovation Institute, Beihang University, Hangzhou, P. R. China
1 shared publication
Latest Publications
- 5 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (3) , 1327-1340
A Local 2-Order Mutual Information Metric Approach for Service Community Detection in Complex Service Networks
This paper presents a novel local 2-order mutual information metric (L2oMI) approach for service community detection, addressing the limitations of traditional methods in dealing with large scale and complex service networks. Firstly, a local 2-order mutual information metric module is proposed to measure the inner-distance between each pair of service stations from different-sized receptive fields in complex service networks, based on which useful semantic node embeddings are obtained for service community detection. Next, a multi-order subgraph information bottleneck for topology module is introduced for the L2oMI approach as a topology constraint to improve the effectiveness of node embedding. Furthermore, a hierarchical interaction learning module is proposed to learn the complicated interactions between pairs of service station and to encode their node features in complex service networks. Numerical experiments using several service network datasets demonstrate the efficacy of the proposed approach.
- 5 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (3) , 1327-1340
A Local 2-Order Mutual Information Metric Approach for Service Community Detection in Complex Service Networks
This paper presents a novel local 2-order mutual information metric (L2oMI) approach for service community detection, addressing the limitations of traditional methods in dealing with large scale and complex service networks. Firstly, a local 2-order mutual information metric module is proposed to measure the inner-distance between each pair of service stations from different-sized receptive fields in complex service networks, based on which useful semantic node embeddings are obtained for service community detection. Next, a multi-order subgraph information bottleneck for topology module is introduced for the L2oMI approach as a topology constraint to improve the effectiveness of node embedding. Furthermore, a hierarchical interaction learning module is proposed to learn the complicated interactions between pairs of service station and to encode their node features in complex service networks. Numerical experiments using several service network datasets demonstrate the efficacy of the proposed approach.
- 27 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Emerging and Selected Topics in Industrial Electronics
- Vol. 6 (4) , 1828-1839
Bayesian Ridge Regression-Based Graph Injection Attack on IIoT
The systems within the Industrial Internet of Things (IIoT) have complex structures and non-Euclidean data, which are challenging to manage. Due to the advantages of Graph Neural Networks (GNNs) in processing non-Euclidean data and complex topologies, they are capable of handling problems in the context of the IIoT. In this work, the IIoT system is structured into multiple layers to facilitate the management of the system and the use of GNNs. GNNs are taken as node classifiers to analyze the state of each edge server in the IIoT system. However, in reality, adversarial attacks often arise in the IIoT, severely impacting system performance. Therefore, a black-box graph injection attack, Bayesian Ridge Regression Injection Attack(BRRIA), is proposed to study the impact of the internal relations on a system and to investigate the vulnerabilities of GNNs. Extensive experiments on two public datasets demonstrate the effectiveness of our attack method. In both experiments targeting specific victim nodes and those attacking a certain category of nodes by targeting critical nodes, BRRIA demonstrates a higher attack accuracy compared to an advanced method. Besides, a synthetic dataset designed to simulate industrial production processes was used to demonstrate the effectiveness of the BRRIA method.
- 27 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Emerging and Selected Topics in Industrial Electronics
- Vol. 6 (4) , 1828-1839
Bayesian Ridge Regression-Based Graph Injection Attack on IIoT
The systems within the Industrial Internet of Things (IIoT) have complex structures and non-Euclidean data, which are challenging to manage. Due to the advantages of Graph Neural Networks (GNNs) in processing non-Euclidean data and complex topologies, they are capable of handling problems in the context of the IIoT. In this work, the IIoT system is structured into multiple layers to facilitate the management of the system and the use of GNNs. GNNs are taken as node classifiers to analyze the state of each edge server in the IIoT system. However, in reality, adversarial attacks often arise in the IIoT, severely impacting system performance. Therefore, a black-box graph injection attack, Bayesian Ridge Regression Injection Attack(BRRIA), is proposed to study the impact of the internal relations on a system and to investigate the vulnerabilities of GNNs. Extensive experiments on two public datasets demonstrate the effectiveness of our attack method. In both experiments targeting specific victim nodes and those attacking a certain category of nodes by targeting critical nodes, BRRIA demonstrates a higher attack accuracy compared to an advanced method. Besides, a synthetic dataset designed to simulate industrial production processes was used to demonstrate the effectiveness of the BRRIA method.
- 9 March 2025
- Published by Wiley in International Journal of Robust and Nonlinear Control
Data‐Driven Information Perception for Complex Industrial Networks of Service Based on Hierarchical Interactive Graph Variational Inference
This article proposes a hierarchical interactive graph variational inference (HI‐GVI) approach to solve the data‐driven information perception problem in complex industrial networks of service. In the HI‐GVI approach, a multi‐layer latent graph structure is organized to describe various influencing factors within the industrial network, and then a hierarchical interaction attention learning method is proposed to learn the latent variables and model the hierarchical interaction process among numerous industrial entities. Furthermore, the HI‐GVI approach employs a generative self‐supervised framework to obtain low‐dimensional variables for downstream industrial tasks, which overcomes the challenge of limited industrial labels. The advantage and performance of the HI‐GVI approach are demonstrated by addressing three different downstream tasks and are assessed in four real‐world datasets.
- 9 March 2025
- Published by Wiley in International Journal of Robust and Nonlinear Control
Data‐Driven Information Perception for Complex Industrial Networks of Service Based on Hierarchical Interactive Graph Variational Inference
This article proposes a hierarchical interactive graph variational inference (HI‐GVI) approach to solve the data‐driven information perception problem in complex industrial networks of service. In the HI‐GVI approach, a multi‐layer latent graph structure is organized to describe various influencing factors within the industrial network, and then a hierarchical interaction attention learning method is proposed to learn the latent variables and model the hierarchical interaction process among numerous industrial entities. Furthermore, the HI‐GVI approach employs a generative self‐supervised framework to obtain low‐dimensional variables for downstream industrial tasks, which overcomes the challenge of limited industrial labels. The advantage and performance of the HI‐GVI approach are demonstrated by addressing three different downstream tasks and are assessed in four real‐world datasets.
- 1 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. 54 (5) , 3124-3133
Cooperative Security Analysis of Industry Cloud Control Systems Under False Data Injection Attacks
This article analyzes a security problem for industry cloud control systems (CCSs), where a malicious false data injection (FDI) attacker may deteriorate the system performance by tampering with the data being transmitted. To improve the robustness of CCSs against FDI attacks, a redundancy-based sensor configuration scheme is provided through analysis of the observability under attacks of the nodes in CCSs. Then, a defending resource allocation scheme is developed based on a two-stage Stackelberg game in order to optimize the overall defense capability of the CCSs with limited defending resource. In this two-stage Stackelberg game, the defender of the CCSs acts first and allocates the defending resource to secure the measurements of wireless sensors. With the knowledge of the defender’s strategy, the attacker then decides which target nodes to launch attacks on. The optimal resource allocation strategy is then designed in the sense of the Stackelberg equilibrium. Furthermore, the defense problem under different attacking resource constraints is investigated. Numerical examples illustrate the effectiveness of the proposed approach.
- 1 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. 54 (5) , 3124-3133
Cooperative Security Analysis of Industry Cloud Control Systems Under False Data Injection Attacks
This article analyzes a security problem for industry cloud control systems (CCSs), where a malicious false data injection (FDI) attacker may deteriorate the system performance by tampering with the data being transmitted. To improve the robustness of CCSs against FDI attacks, a redundancy-based sensor configuration scheme is provided through analysis of the observability under attacks of the nodes in CCSs. Then, a defending resource allocation scheme is developed based on a two-stage Stackelberg game in order to optimize the overall defense capability of the CCSs with limited defending resource. In this two-stage Stackelberg game, the defender of the CCSs acts first and allocates the defending resource to secure the measurements of wireless sensors. With the knowledge of the defender’s strategy, the attacker then decides which target nodes to launch attacks on. The optimal resource allocation strategy is then designed in the sense of the Stackelberg equilibrium. Furthermore, the defense problem under different attacking resource constraints is investigated. Numerical examples illustrate the effectiveness of the proposed approach.
- 16 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (5) , 2146-2156
A Subgraph-Based Hierarchical Q-Learning Approach to Optimal Resource Scheduling for Complex Industrial Networks
This paper proposes a subgraph-based hierarchical Q-learning network (SgHQN) approach to solve the optimal resource scheduling problem for complex industrial networks. In the industrial network, each connection between two individual stations has limited communication bandwidth, while each station has limited computing and storage capability and is only accessible to its local information. The resource packages flowing within the industrial network are treated as agents that have different sizes and different levels of decision-making priority. This makes the industrial resource scheduling problem on the industrial network a multi-level decision-making problem with information asymmetry. Specifically, the resource packages with lower decision-making priority have knowledge of the decisions made by those with higher priority, but not vice versa. To solve this resource scheduling problem with information asymmetry, an SgHQN model is developed by exploiting partial observations. It is found that the proposed SgHQN can be used to solve resource scheduling problems for general industrial networks. Numerical experiments simulating industrial scheduling scenarios demonstrate the effectiveness and advantages of our method.
- 16 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (5) , 2146-2156
A Subgraph-Based Hierarchical Q-Learning Approach to Optimal Resource Scheduling for Complex Industrial Networks
This paper proposes a subgraph-based hierarchical Q-learning network (SgHQN) approach to solve the optimal resource scheduling problem for complex industrial networks. In the industrial network, each connection between two individual stations has limited communication bandwidth, while each station has limited computing and storage capability and is only accessible to its local information. The resource packages flowing within the industrial network are treated as agents that have different sizes and different levels of decision-making priority. This makes the industrial resource scheduling problem on the industrial network a multi-level decision-making problem with information asymmetry. Specifically, the resource packages with lower decision-making priority have knowledge of the decisions made by those with higher priority, but not vice versa. To solve this resource scheduling problem with information asymmetry, an SgHQN model is developed by exploiting partial observations. It is found that the proposed SgHQN can be used to solve resource scheduling problems for general industrial networks. Numerical experiments simulating industrial scheduling scenarios demonstrate the effectiveness and advantages of our method.