Zhiyong Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhiyong Zhang has a total of 163 co-authors. Here are the top co-authors by number of shared publications:
- Kim-Kwang Raymond Choo
Department of Information Systems and Cybersecurity, The University of Texas at San Antonio, San Antonio, USA
9 shared publications Lili Zhang
Henan University of Science and Technology, China
9 shared publicationsBrij B. Gupta
7 shared publicationsChangwei Zhao
Information Engineering College, Henan University of Science and Technology, Luoyang, China
7 shared publicationsJunchang Jing
7 shared publications- Zhongya Zhang
Information Engineering College, Henan University of Science and Technology, Luoyang, China
7 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1974-1984
Multi-stage Interpretable Temporal Mixture of Experts for Information Diffusion Prediction
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1974-1984
Open AccessMulti-stage Interpretable Temporal Mixture of Experts for Information Diffusion Prediction
- Open Access
- 20 September 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 220
A novel data-level prompt injection attack against graph prompt learning
Graph Neural Networks (GNNs) are increasingly deployed in security-critical applications, including fraud detection, recommendation integrity analysis, and scientific knowledge mining. Recently, Graph Prompt Learning (GPL) has emerged as a parameter-efficient paradigm for adapting pretrained GNNs to downstream tasks. By optimizing only a small set of prompt parameters while keeping the pretrained encoder frozen, GPL enables efficient task adaptation under limited supervision. Despite its empirical success, the security implications of this paradigm remain largely unexplored. In this work, we investigate the vulnerability of GPL to data-level manipulations and introduce a novel threat, termed the data-level prompt injection attack. Unlike conventional attacks that rely on poisoning pretraining data or modifying model parameters, the proposed attack operates entirely at the downstream prompt learning stage. Specifically, an attacker injects a small number of carefully crafted malicious samples into the training dataset, without altering the pretrained encoder or the prompt learning algorithm. To realize this threat, we propose Graph Prompt Injection Attack (GPIA), which embeds carefully designed subgraph into training graphs to influence prompt optimization. Extensive experiments on multiple benchmarks show that GPIA achieves high attack success rates under low injection ratios while maintaining clean accuracy.
- 20 September 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 220
Open AccessA novel data-level prompt injection attack against graph prompt learning
Graph Neural Networks (GNNs) are increasingly deployed in security-critical applications, including fraud detection, recommendation integrity analysis, and scientific knowledge mining. Recently, Graph Prompt Learning (GPL) has emerged as a parameter-efficient paradigm for adapting pretrained GNNs to downstream tasks. By optimizing only a small set of prompt parameters while keeping the pretrained encoder frozen, GPL enables efficient task adaptation under limited supervision. Despite its empirical success, the security implications of this paradigm remain largely unexplored. In this work, we investigate the vulnerability of GPL to data-level manipulations and introduce a novel threat, termed the data-level prompt injection attack. Unlike conventional attacks that rely on poisoning pretraining data or modifying model parameters, the proposed attack operates entirely at the downstream prompt learning stage. Specifically, an attacker injects a small number of carefully crafted malicious samples into the training dataset, without altering the pretrained encoder or the prompt learning algorithm. To realize this threat, we propose Graph Prompt Injection Attack (GPIA), which embeds carefully designed subgraph into training graphs to influence prompt optimization. Extensive experiments on multiple benchmarks show that GPIA achieves high attack success rates under low injection ratios while maintaining clean accuracy.
- Open Access
A Human-Centered Situation Security Framework Based on Multi-Sensor Data Fusion for Abnormal Behavior Recognition in the Industrial Internet
To guarantee the robust security of the industrial internet, we propose a human-centered situational security framework that comprehensively considers multi-sensor industrial elements. Constructing an industrial spatial morphism mechanism allows for in-depth analysis of abnormal behaviors. Through the acquisition and effective analysis of multi-sensor data in industrial scenarios, a human-centered situation security is established, the computational complexity of the abnormal behavior identification model is reduced, while the accuracy and efficiency of identification are improved. This method has been verified for its generality in the processing data of metallic materials in 5G sensor networks. The experimental results demonstrate that the industrial situational triplets, validated through ISA-95 standard testing, achieve an average coverage rate of 99.6% in industrial scenarios. Extensive experiments on the Gas dataset and a real-world 5G-enabled production line demonstrated that the proposed method achieved mean precisions of 97.35% and 97.41%, respectively. Compared with the non-vectorized data, the processing time for 5G data and EdgeIIoT data decreased by 79.4 s and 52.9 s, respectively. The ablation study confirms that incorporating the intention component (T) yields a 13.92-percentage-point F1-score improvement over the E + A configuration. Through multi-dimensional performance validation, the security analysis framework for multi-sensor data in industrial scenarios enhances the identification performance of abnormal behaviors in the industrial internet and has excellent generalization capabilities.
Open AccessA Human-Centered Situation Security Framework Based on Multi-Sensor Data Fusion for Abnormal Behavior Recognition in the Industrial Internet
To guarantee the robust security of the industrial internet, we propose a human-centered situational security framework that comprehensively considers multi-sensor industrial elements. Constructing an industrial spatial morphism mechanism allows for in-depth analysis of abnormal behaviors. Through the acquisition and effective analysis of multi-sensor data in industrial scenarios, a human-centered situation security is established, the computational complexity of the abnormal behavior identification model is reduced, while the accuracy and efficiency of identification are improved. This method has been verified for its generality in the processing data of metallic materials in 5G sensor networks. The experimental results demonstrate that the industrial situational triplets, validated through ISA-95 standard testing, achieve an average coverage rate of 99.6% in industrial scenarios. Extensive experiments on the Gas dataset and a real-world 5G-enabled production line demonstrated that the proposed method achieved mean precisions of 97.35% and 97.41%, respectively. Compared with the non-vectorized data, the processing time for 5G data and EdgeIIoT data decreased by 79.4 s and 52.9 s, respectively. The ablation study confirms that incorporating the intention component (T) yields a 13.92-percentage-point F1-score improvement over the E + A configuration. Through multi-dimensional performance validation, the security analysis framework for multi-sensor data in industrial scenarios enhances the identification performance of abnormal behaviors in the industrial internet and has excellent generalization capabilities.
- 30 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 350
SRD: A model-agnostic semantic framework for jailbreak mitigation in large language models
- 30 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 350
SRD: A model-agnostic semantic framework for jailbreak mitigation in large language models
- Open Access
- 13 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Cyber-Physical Systems
- Vol. 10 (4) , 1-28
A Situation-Based Adaptive Defense Framework For IIoT Using Hierarchical Multi-Agent Reinforcement Learning
The deep convergence of Information Technology (IT) and Operational Technology (OT) exposes Industrial Internet of Things (IIoT) systems to complex cross-layer attacks. However, traditional defense methods are mostly designed for a single domain and rely on static or rule-based mechanisms, making them ineffective in capturing cross-layer attack evolution and coordinating adaptive responses. Therefore, this article proposes a Situation-Based Hierarchical Multi-Agent Reinforcement Learning (Situ-HMARL) adaptive defense framework, which formulates defense strategies based on real-time global industrial situations and proactively enhances the resilience of IIoT systems. Firstly, a three-stage industrial situational awareness architecture is proposed to continuously fuse heterogeneous data from the IT and OT layers into a structured Global Industrial Situation Vector (GISV), which serves as a unified global observation space for defense decision-making. Secondly, a lightweight Moving Target Defense (MTD) mechanism is designed to adaptively trigger IP-hopping based on the global industrial situation, thereby reducing overhead while preserving system availability. On this basis, a two-level Hierarchical Multi-Agent Reinforcement Learning (HMARL) framework is developed to decouple perception, decision-making, and enforcement. Low-level agents perform real-time local situational perception and execute defense actions, while a high-level agent reasons over the global industrial situation to generate coordinated defense strategies that minimize system losses. Extensive experiments conducted on the Cyber Operations Research Gym (CybORG) platform validate that the proposed framework effectively mitigates cross-layer attacks and significantly improves the availability of IIoT systems.
- 13 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Cyber-Physical Systems
- Vol. 10 (4) , 1-28
Open AccessA Situation-Based Adaptive Defense Framework For IIoT Using Hierarchical Multi-Agent Reinforcement Learning
The deep convergence of Information Technology (IT) and Operational Technology (OT) exposes Industrial Internet of Things (IIoT) systems to complex cross-layer attacks. However, traditional defense methods are mostly designed for a single domain and rely on static or rule-based mechanisms, making them ineffective in capturing cross-layer attack evolution and coordinating adaptive responses. Therefore, this article proposes a Situation-Based Hierarchical Multi-Agent Reinforcement Learning (Situ-HMARL) adaptive defense framework, which formulates defense strategies based on real-time global industrial situations and proactively enhances the resilience of IIoT systems. Firstly, a three-stage industrial situational awareness architecture is proposed to continuously fuse heterogeneous data from the IT and OT layers into a structured Global Industrial Situation Vector (GISV), which serves as a unified global observation space for defense decision-making. Secondly, a lightweight Moving Target Defense (MTD) mechanism is designed to adaptively trigger IP-hopping based on the global industrial situation, thereby reducing overhead while preserving system availability. On this basis, a two-level Hierarchical Multi-Agent Reinforcement Learning (HMARL) framework is developed to decouple perception, decision-making, and enforcement. Low-level agents perform real-time local situational perception and execute defense actions, while a high-level agent reasons over the global industrial situation to generate coordinated defense strategies that minimize system losses. Extensive experiments conducted on the Cyber Operations Research Gym (CybORG) platform validate that the proposed framework effectively mitigates cross-layer attacks and significantly improves the availability of IIoT systems.
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. PP (15513203) , 1-12
Differential Random Fault Attack of Ultra-Lightweight Cipher ANU for Industrial Internet
Lightweight ciphers are widely used in the Industrial Internet of Things (IIoT), covering data encryption, communication encryption, and authenticated encryption. Due to their efficiency and low resource consumption, these cryptographic algorithms have become core solutions for network security, yet they still face security challenges. The ANU algorithm, characterized by its small hardware implementation area and low power consumption, is suitable for securing the Industrial Internet of Things (IIoT). In this article, we propose two differential fault attack methods for ANU based on the random nibble and byte models, respectively. Theoretically, the two attacks require 82.4 and 41.2 faults respectively to recover the master key. Simulation results show that the two attacks require 85.66 and 42.86 faults, respectively. The proposed attacks demonstrate significant improvements in terms of analysis difficulty, model applicability, and the number of faults required, offering a new perspective for algorithm security research and evaluation in IIoT environments.
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. PP (15513203) , 1-12
Differential Random Fault Attack of Ultra-Lightweight Cipher ANU for Industrial Internet
Lightweight ciphers are widely used in the Industrial Internet of Things (IIoT), covering data encryption, communication encryption, and authenticated encryption. Due to their efficiency and low resource consumption, these cryptographic algorithms have become core solutions for network security, yet they still face security challenges. The ANU algorithm, characterized by its small hardware implementation area and low power consumption, is suitable for securing the Industrial Internet of Things (IIoT). In this article, we propose two differential fault attack methods for ANU based on the random nibble and byte models, respectively. Theoretically, the two attacks require 82.4 and 41.2 faults respectively to recover the master key. Simulation results show that the two attacks require 85.66 and 42.86 faults, respectively. The proposed attacks demonstrate significant improvements in terms of analysis difficulty, model applicability, and the number of faults required, offering a new perspective for algorithm security research and evaluation in IIoT environments.
- 8 June 2026
- Published by Wiley in Security and Privacy
- Vol. 9 (4)
A Situation Prediction Driven Stochastic‐Stackelberg Game Approach for Defense Decision‐Making in Industrial Internet
In the highly interconnected industrial internet, security threats often exhibit multi‐stage and cross‐domain cascading characteristics. Traditional defense strategies that depend on static configurations are insufficient to support real‐time and anticipatory protection. To address this limitation, this paper proposes an industrial situation‐prediction‐driven role‐switching stochastic Stackelberg game method. First, we construct an industrial network situation prediction model constrained by attack graphs. By incorporating a reachability‐mask matrix and vulnerability exposure metrics, the proposed model enables interpretable and forward‐looking situation prediction while strictly preventing illegal state transitions. Then, we propose a role‐switching stochastic Stackelberg game framework that integrates leader‐follower role transitions induced by real‐time and predicted situations into a unified decision‐making model. Finally, we introduce a situation‐prediction‐driven multi‐agent reinforcement learning algorithm to approximate game equilibria, enabling efficient computation of optimal pre‐deployment defense strategies. Experimental results on a simulation testbed demonstrate that the proposed method improves prediction accuracy by approximately 11% over baseline approaches and completely eliminates unreachable state transitions. The prediction‐enhanced adaptive defense strategy reduces attack success rates by 57%, decreases critical asset loss by about 48%, and significantly extends the attack chain length, yielding a defense strategy that is both cost‐efficient and loss‐minimizing.
- 8 June 2026
- Published by Wiley in Security and Privacy
- Vol. 9 (4)
A Situation Prediction Driven Stochastic‐Stackelberg Game Approach for Defense Decision‐Making in Industrial Internet
In the highly interconnected industrial internet, security threats often exhibit multi‐stage and cross‐domain cascading characteristics. Traditional defense strategies that depend on static configurations are insufficient to support real‐time and anticipatory protection. To address this limitation, this paper proposes an industrial situation‐prediction‐driven role‐switching stochastic Stackelberg game method. First, we construct an industrial network situation prediction model constrained by attack graphs. By incorporating a reachability‐mask matrix and vulnerability exposure metrics, the proposed model enables interpretable and forward‐looking situation prediction while strictly preventing illegal state transitions. Then, we propose a role‐switching stochastic Stackelberg game framework that integrates leader‐follower role transitions induced by real‐time and predicted situations into a unified decision‐making model. Finally, we introduce a situation‐prediction‐driven multi‐agent reinforcement learning algorithm to approximate game equilibria, enabling efficient computation of optimal pre‐deployment defense strategies. Experimental results on a simulation testbed demonstrate that the proposed method improves prediction accuracy by approximately 11% over baseline approaches and completely eliminates unreachable state transitions. The prediction‐enhanced adaptive defense strategy reduces attack success rates by 57%, decreases critical asset loss by about 48%, and significantly extends the attack chain length, yielding a defense strategy that is both cost‐efficient and loss‐minimizing.
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 4, 335-345
A Multi-Point Collaborative Defense Method Against Unknown Attacks for Industrial Cyber-Physical Systems
As industrial cyber-physical systems (CPS) become increasingly interconnected and attack methods diversify, industrial equipment remains vulnerable to unknown attacks, potentially causing severe consequences. Traditional single-point defense solutions struggle to counter rapidly emerging unknown threats and fail to meet the practical security and reliability demands of industrial environments. To overcome this limitation, this paper proposes a multi-point collaborative defense framework for unknown attacks for CPS. For single-point defense, a unified unknown attack detection model is deployed to perform independent detection and identification, completing preliminary judgment. In multi-point scenarios, a blockchain-based collaborative defense framework is constructed. When any local detection model identifies an unknown attack, it will initiate a collaborative request via the blockchain. Other collaborative nodes will independently verify the attack and generate a final verdict. A dynamic trust evaluation mechanism has also been designed to assess the trustworthiness and weighting of nodes, thereby effectively suppressing interference from malicious nodes and ensuring the robustness of the collaborative process. Meanwhile, through the model's self-evolution mechanism, continuously optimize the local detection model by using collaboratively validated unknown attack samples to achieve model updates. Experimental results demonstrate that this framework effectively defends against unknown attacks, significantly enhancing detection capabilities and overall system defense levels.
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Cyber-Physical Systems
- Vol. 4, 335-345
A Multi-Point Collaborative Defense Method Against Unknown Attacks for Industrial Cyber-Physical Systems
As industrial cyber-physical systems (CPS) become increasingly interconnected and attack methods diversify, industrial equipment remains vulnerable to unknown attacks, potentially causing severe consequences. Traditional single-point defense solutions struggle to counter rapidly emerging unknown threats and fail to meet the practical security and reliability demands of industrial environments. To overcome this limitation, this paper proposes a multi-point collaborative defense framework for unknown attacks for CPS. For single-point defense, a unified unknown attack detection model is deployed to perform independent detection and identification, completing preliminary judgment. In multi-point scenarios, a blockchain-based collaborative defense framework is constructed. When any local detection model identifies an unknown attack, it will initiate a collaborative request via the blockchain. Other collaborative nodes will independently verify the attack and generate a final verdict. A dynamic trust evaluation mechanism has also been designed to assess the trustworthiness and weighting of nodes, thereby effectively suppressing interference from malicious nodes and ensuring the robustness of the collaborative process. Meanwhile, through the model's self-evolution mechanism, continuously optimize the local detection model by using collaboratively validated unknown attack samples to achieve model updates. Experimental results demonstrate that this framework effectively defends against unknown attacks, significantly enhancing detection capabilities and overall system defense levels.
- Open Access
- 17 May 2026
- Published by MDPI AG in Electronics
- Vol. 15 (10) , 2156
SD-Fuzz: A State-Aware Industrial Control Protocol Fuzzing Framework Based on Diffusion Models
Current fuzzing techniques for industrial control protocols (ICPs) encounter notable challenges, including model training instability, limited sample diversity, and the inability to manage complex state dependencies in protocol interactions. To address these issues, this paper presents SD-Fuzz, a state-aware fuzzing framework that integrates a discrete denoising diffusion probabilistic model (DDPM) with an online Hidden Markov Model (HMM). The discrete DDPM is designed to generate syntactically valid and diverse protocol messages using cosine noise scheduling and Denoising Diffusion Implicit Model (DDIM) sampling, while the HMM performs unsupervised learning of state transitions from real traffic to guide the creation of logically consistent multi-step interaction sequences. The framework is evaluated on three representative Modbus/TCP slave implementations. Evaluations based on 5 h benchmark campaigns across multiple independent runs indicate that SD-Fuzz achieves a mean test case recognition rate (TCRR) of 91.3% and an HMM-inferred state transition coverage of 50.1%, exhibiting statistically significant improvements over the evaluated baselines. Furthermore, an extended 8 h vulnerability mining campaign demonstrates its capability to trigger deep-seated exceptions, including buffer overflows and protocol state violations, which are typically challenging to access using traditional stateless approaches. This work illustrates the feasibility of combining diffusion-based generation with lightweight state inference for automated vulnerability discovery in industrial control systems. Directions for future work include validation on physical programmable logic controller (PLC) hardware to acquire internal code coverage feedback.
- 17 May 2026
- Published by MDPI AG in Electronics
- Vol. 15 (10) , 2156
Open AccessSD-Fuzz: A State-Aware Industrial Control Protocol Fuzzing Framework Based on Diffusion Models
Current fuzzing techniques for industrial control protocols (ICPs) encounter notable challenges, including model training instability, limited sample diversity, and the inability to manage complex state dependencies in protocol interactions. To address these issues, this paper presents SD-Fuzz, a state-aware fuzzing framework that integrates a discrete denoising diffusion probabilistic model (DDPM) with an online Hidden Markov Model (HMM). The discrete DDPM is designed to generate syntactically valid and diverse protocol messages using cosine noise scheduling and Denoising Diffusion Implicit Model (DDIM) sampling, while the HMM performs unsupervised learning of state transitions from real traffic to guide the creation of logically consistent multi-step interaction sequences. The framework is evaluated on three representative Modbus/TCP slave implementations. Evaluations based on 5 h benchmark campaigns across multiple independent runs indicate that SD-Fuzz achieves a mean test case recognition rate (TCRR) of 91.3% and an HMM-inferred state transition coverage of 50.1%, exhibiting statistically significant improvements over the evaluated baselines. Furthermore, an extended 8 h vulnerability mining campaign demonstrates its capability to trigger deep-seated exceptions, including buffer overflows and protocol state violations, which are typically challenging to access using traditional stateless approaches. This work illustrates the feasibility of combining diffusion-based generation with lightweight state inference for automated vulnerability discovery in industrial control systems. Directions for future work include validation on physical programmable logic controller (PLC) hardware to acquire internal code coverage feedback.
- 2 April 2026
- Published by Elsevier in Pattern Recognition
- Vol. 179
CM-PIUG: Cross-modal prompt injection unified modeling and game-theoretic defense strategies
- 2 April 2026
- Published by Elsevier in Pattern Recognition
- Vol. 179
CM-PIUG: Cross-modal prompt injection unified modeling and game-theoretic defense strategies