Zhiqiang Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Zhiqiang Zhang has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Le Wang
Guangzhou University
5 shared publications - Zhaoquan Gu
Harbin Institute of Technology
5 shared publications - Zhaohua Li
Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China
3 shared publications - Liyi Zeng
Pengcheng Laboratory, Shenzhen, China
2 shared publications - Rongxin Hu
Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou 510700, China
2 shared publications - Chenhui Zhang
Department of New Networks, Peng Cheng Laboratory, Xili Street, Shenzhen 518000, China
1 shared publication
Latest Publications
- Open Access
- 30 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (13) , 6538
HARMONI: A Two-Stage Hybrid Learning Framework with Dynamic Metric Learning for Interpretable NIDS
The increasing sophistication of cyber attacks has created a growing demand for effective Network Intrusion Detection Systems (NIDSs). Although deep learning has improved NIDS performance, existing models often lack adaptive mechanisms for spatiotemporal feature fusion and struggle with complex traffic distributions characterized by severe intra-class heterogeneity and inter-class overlap. Meanwhile, current interpretability methods mainly rely on feature importance analysis and provide limited insight into the model’s decision process. To address these challenges, we propose HARMONI, a two-stage hybrid learning framework that enhances both detection accuracy and model interpretability. In the first stage, a dual-branch Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) architecture extracts spatiotemporal features, which are dynamically fused through a lightweight adaptive gating network. The representation learning process is jointly optimized using a Dynamic Class-Center Loss to enforce intra-class compactness and inter-class separability in the latent space. In the second stage, the learned deep representations are concatenated with raw traffic features and fed into an ensemble classifier. This residual-style design mitigates information loss during deep encoding while leveraging the non-linear modeling capability of ensemble learning. We further develop a multi-level interpretability framework based on SHapley Additive exPlanations (SHAP) that analyzes global feature importance, individual feature contributions, and feature interactions to provide quantitative insights into the model’s decision mechanisms. Experiments on four benchmark datasets show that HARMONI consistently outperforms state-of-the-art baselines, achieving 80.19% and 78.24% accuracy on NSL-KDD and UNSW-NB15 respectively, surpassing representative deep learning and ensemble methods.
- 30 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (13) , 6538
Open AccessHARMONI: A Two-Stage Hybrid Learning Framework with Dynamic Metric Learning for Interpretable NIDS
The increasing sophistication of cyber attacks has created a growing demand for effective Network Intrusion Detection Systems (NIDSs). Although deep learning has improved NIDS performance, existing models often lack adaptive mechanisms for spatiotemporal feature fusion and struggle with complex traffic distributions characterized by severe intra-class heterogeneity and inter-class overlap. Meanwhile, current interpretability methods mainly rely on feature importance analysis and provide limited insight into the model’s decision process. To address these challenges, we propose HARMONI, a two-stage hybrid learning framework that enhances both detection accuracy and model interpretability. In the first stage, a dual-branch Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) architecture extracts spatiotemporal features, which are dynamically fused through a lightweight adaptive gating network. The representation learning process is jointly optimized using a Dynamic Class-Center Loss to enforce intra-class compactness and inter-class separability in the latent space. In the second stage, the learned deep representations are concatenated with raw traffic features and fed into an ensemble classifier. This residual-style design mitigates information loss during deep encoding while leveraging the non-linear modeling capability of ensemble learning. We further develop a multi-level interpretability framework based on SHapley Additive exPlanations (SHAP) that analyzes global feature importance, individual feature contributions, and feature interactions to provide quantitative insights into the model’s decision mechanisms. Experiments on four benchmark datasets show that HARMONI consistently outperforms state-of-the-art baselines, achieving 80.19% and 78.24% accuracy on NSL-KDD and UNSW-NB15 respectively, surpassing representative deep learning and ensemble methods.
- Open Access
- 17 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 12203
Phish-Master: Leveraging Large Language Models for Advanced Phishing Email Generation and Detection
Phishing emails present a significant and persistent cybersecurity threat to individuals and organizations globally due to the difficulty in detecting these malicious messages. Large Language Models (LLMs) have inadvertently intensified this challenge by facilitating the automated creation of high-quality, covert phishing emails that can evade traditional rule-based detection systems. In this study, we examine the offensive capabilities of LLMs in generating phishing emails and introduce Phish-Master, a novel algorithm that integrates Chain-of-Thought (COT) reasoning, MetaPrompt techniques, and domain-specific insights to produce phishing emails designed to bypass enterprise-level filters. Our experiment, involving 100 malicious emails, validates Phish-Master’s real-world effectiveness, achieving a 99% evasion rate within authentic campus networks, successfully bypassing filters and targeting recipients, a testament to its capability in navigating complex network environments. To counteract the threat posed by Phish-Master and similar LLM-generated phishing emails, we have developed a multi-machine learning model integration framework trained on Kaggle’s phishing email dataset. This framework achieved an impressive detection rate of 99.87% on a rigorous test set of LLM-generated phishing emails, highlighting the critical role of our specialized dataset in enabling the detection tool to effectively recognize sophisticated patterns in LLM-crafted phishing emails. This study highlights the evolving threat of LLM-generated phishing emails and introduces an effective detection algorithm to mitigate this risk, emphasizing the importance of continued research in this domain.
- 17 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 12203
Open AccessPhish-Master: Leveraging Large Language Models for Advanced Phishing Email Generation and Detection
Phishing emails present a significant and persistent cybersecurity threat to individuals and organizations globally due to the difficulty in detecting these malicious messages. Large Language Models (LLMs) have inadvertently intensified this challenge by facilitating the automated creation of high-quality, covert phishing emails that can evade traditional rule-based detection systems. In this study, we examine the offensive capabilities of LLMs in generating phishing emails and introduce Phish-Master, a novel algorithm that integrates Chain-of-Thought (COT) reasoning, MetaPrompt techniques, and domain-specific insights to produce phishing emails designed to bypass enterprise-level filters. Our experiment, involving 100 malicious emails, validates Phish-Master’s real-world effectiveness, achieving a 99% evasion rate within authentic campus networks, successfully bypassing filters and targeting recipients, a testament to its capability in navigating complex network environments. To counteract the threat posed by Phish-Master and similar LLM-generated phishing emails, we have developed a multi-machine learning model integration framework trained on Kaggle’s phishing email dataset. This framework achieved an impressive detection rate of 99.87% on a rigorous test set of LLM-generated phishing emails, highlighting the critical role of our specialized dataset in enabling the detection tool to effectively recognize sophisticated patterns in LLM-crafted phishing emails. This study highlights the evolving threat of LLM-generated phishing emails and introduces an effective detection algorithm to mitigate this risk, emphasizing the importance of continued research in this domain.
- 16 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 30379-30397
GBFKAN: An Adaptive Multilayer Interpretable Architecture for Intrusion Detection in Various Internet of Things Scenarios
As Internet of Things (IoT) technologies continue to evolve and gain widespread integration across various sectors, IoT devices have become increasingly interconnected and ubiquitous. However, this surge has also led to more sophisticated, diverse, and covert attacks, posing severe security challenges to IoT ecosystems and their defense mechanisms. Intrusion detection systems (IDSs) are instrumental in securing the IoT by uncovering and mitigating malicious behaviors in real time. Nevertheless, the growing complexity of network traffic presents significant challenges for IDS, often resulting in a higher false alarm rate and a lower detection rate, due to the intricate characteristics of space and notable temporal dependencies present in network traffic. We conduct an in-depth dimensionality reduction analysis using the t-distributed stochastic neighbor embedding methodology to examine the relationships among attack examples. Then, to enhance the generalization and robustness of IDS, we propose gated BiLSTM-fused Kolmogorov-Arnold network (GBFKAN), a novel interpretable three-layer network architecture. The structured design of GBFKAN can comprehensively extract and perceive key information from multiple abstract levels in the raw traffic, enhancing the architecture’s generalization ability and robustness. Furthermore, we utilize the Shapley additive explanations method to provide detailed explanations of the detection outcomes, augmenting its credibility. Building on this, we successfully identify the erroneous decision-making paradigms within GBFKAN by incorporating empirical knowledge. Extensive simulation experiments conducted on the four widely recognized intrusion detection datasets have conclusively demonstrated that our proposed GBFKAN model surpasses existing methods in terms of intrusion detection performance.
- 16 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 30379-30397
GBFKAN: An Adaptive Multilayer Interpretable Architecture for Intrusion Detection in Various Internet of Things Scenarios
As Internet of Things (IoT) technologies continue to evolve and gain widespread integration across various sectors, IoT devices have become increasingly interconnected and ubiquitous. However, this surge has also led to more sophisticated, diverse, and covert attacks, posing severe security challenges to IoT ecosystems and their defense mechanisms. Intrusion detection systems (IDSs) are instrumental in securing the IoT by uncovering and mitigating malicious behaviors in real time. Nevertheless, the growing complexity of network traffic presents significant challenges for IDS, often resulting in a higher false alarm rate and a lower detection rate, due to the intricate characteristics of space and notable temporal dependencies present in network traffic. We conduct an in-depth dimensionality reduction analysis using the t-distributed stochastic neighbor embedding methodology to examine the relationships among attack examples. Then, to enhance the generalization and robustness of IDS, we propose gated BiLSTM-fused Kolmogorov-Arnold network (GBFKAN), a novel interpretable three-layer network architecture. The structured design of GBFKAN can comprehensively extract and perceive key information from multiple abstract levels in the raw traffic, enhancing the architecture’s generalization ability and robustness. Furthermore, we utilize the Shapley additive explanations method to provide detailed explanations of the detection outcomes, augmenting its credibility. Building on this, we successfully identify the erroneous decision-making paradigms within GBFKAN by incorporating empirical knowledge. Extensive simulation experiments conducted on the four widely recognized intrusion detection datasets have conclusively demonstrated that our proposed GBFKAN model surpasses existing methods in terms of intrusion detection performance.
- Open Access
- 20 November 2024
- Published by MDPI AG in Preprints.org
Pre-Trained Language Model for Temporal Knowledge Graph Completion
Large Language Models (LLMs) have been proven remarkable in natural language processing, which prompts numerous works on knowledge extraction, knowledge fusion, knowledge representation, and knowledge completion. Existing works mainly focus on static multi-relational knowledge graph (KG). Unlike static knowledge graph, temporal knowledge graph (TKG) contains temporal information and evolves over time. Learning and reasoning about the representation of temporal knowledge graph are more difficult. Training or fine-tuning LLMs for temporal graph related tasks incurs significant computational overhead and requires the design of prompts. Conducting tasks of TKG does not necessarily require such complex work. Therefore, we explore temporal knowledge graph completion (TKGC) based on pre-trained "small" language model. We propose TKG-BERT by applying BERT for temporal knowledge graph completion and classification. Specifically, We introduce three ways to model temporal knowledge in TKG-BERT: vanilla knowledge embedding (Van.), explicit time modeling (Exp.) and implicit time modeling (Imp.). TKG-BERT(Van.) only adopts static knowledge without embedding time information; TKG-BERT(Exp.) embeds timestamp in quadruple explicitly; TKG-BERT(Imp.) models time implicitly, by dividing the training set and testing set in chronological order. We conduct experiments on ICEWS14 and ICEWS05-15, which are two public temporal knowledge graph datasets. Various experiments of temporal knowledge graph completion and classification tasks show the effectiveness of pre-trained language model for TKG completion. We also compare the performance of TKG-BERT accross different time modeling way and proportion of training set.
- 20 November 2024
- Published by MDPI AG in Preprints.org
Open AccessPre-Trained Language Model for Temporal Knowledge Graph Completion
Large Language Models (LLMs) have been proven remarkable in natural language processing, which prompts numerous works on knowledge extraction, knowledge fusion, knowledge representation, and knowledge completion. Existing works mainly focus on static multi-relational knowledge graph (KG). Unlike static knowledge graph, temporal knowledge graph (TKG) contains temporal information and evolves over time. Learning and reasoning about the representation of temporal knowledge graph are more difficult. Training or fine-tuning LLMs for temporal graph related tasks incurs significant computational overhead and requires the design of prompts. Conducting tasks of TKG does not necessarily require such complex work. Therefore, we explore temporal knowledge graph completion (TKGC) based on pre-trained "small" language model. We propose TKG-BERT by applying BERT for temporal knowledge graph completion and classification. Specifically, We introduce three ways to model temporal knowledge in TKG-BERT: vanilla knowledge embedding (Van.), explicit time modeling (Exp.) and implicit time modeling (Imp.). TKG-BERT(Van.) only adopts static knowledge without embedding time information; TKG-BERT(Exp.) embeds timestamp in quadruple explicitly; TKG-BERT(Imp.) models time implicitly, by dividing the training set and testing set in chronological order. We conduct experiments on ICEWS14 and ICEWS05-15, which are two public temporal knowledge graph datasets. Various experiments of temporal knowledge graph completion and classification tasks show the effectiveness of pre-trained language model for TKG completion. We also compare the performance of TKG-BERT accross different time modeling way and proportion of training set.
- 10 July 2024
- Published by Springer Nature in World Wide Web
- Vol. 27 (4) , 1-25
MIM: A multiple integration model for intrusion detection on imbalanced samples
- 10 July 2024
- Published by Springer Nature in World Wide Web
- Vol. 27 (4) , 1-25
MIM: A multiple integration model for intrusion detection on imbalanced samples
- 5 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (2) , 756-767
E2EGI: End-to-End Gradient Inversion in Federated Learning
A plethora of healthcare data is produced every day due to the proliferation of prominent technologies such as Internet of Medical Things (IoMT). Digital-driven smart devices like wearable watches, wristbands and bracelets are utilized extensively in modern healthcare applications. Mining valuable information from the data distributed at the owners' level is useful, but it is challenging to preserve data privacy. Federated learning (FL) has swiftly surged in popularity due to its efficacy in dealing privacy vulnerabilities. Recent studies have demonstrated that Gradient Inversion Attack (GIA) can reconstruct the input data by leaked gradients, previous work demonstrated the achievement of GIA in very limited scenarios, such as the label repetition rate of the target sample being low and batch sizes being smaller than 48. In this paper, a novel method of End-to-End Gradient Inversion (E2EGI) is proposed. Compared to the state-of-the-art method, E2EGI's Minimum Loss Combinatorial Optimization (MLCO) has the ability to realize reconstructed samples with higher similarity, and the Distributed Gradient Inversion algorithm can implement GIA with batch sizes of 8 to 256 on deep network models (such as ResNet-50) and ImageNet datasets. A new Label Reconstruction algorithm is developed that relies only on the gradient information of the target model, which can achieve a label reconstruction accuracy of 81% in one batch sample with a label repetition rate of 96%, a 27% improvement over the state-of-the-art method. This proposed work can underpin data security assessments for healthcare federated learning. Code is available at https://github.com/zhaohuali/E2EGI .
- 5 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (2) , 756-767
E2EGI: End-to-End Gradient Inversion in Federated Learning
A plethora of healthcare data is produced every day due to the proliferation of prominent technologies such as Internet of Medical Things (IoMT). Digital-driven smart devices like wearable watches, wristbands and bracelets are utilized extensively in modern healthcare applications. Mining valuable information from the data distributed at the owners' level is useful, but it is challenging to preserve data privacy. Federated learning (FL) has swiftly surged in popularity due to its efficacy in dealing privacy vulnerabilities. Recent studies have demonstrated that Gradient Inversion Attack (GIA) can reconstruct the input data by leaked gradients, previous work demonstrated the achievement of GIA in very limited scenarios, such as the label repetition rate of the target sample being low and batch sizes being smaller than 48. In this paper, a novel method of End-to-End Gradient Inversion (E2EGI) is proposed. Compared to the state-of-the-art method, E2EGI's Minimum Loss Combinatorial Optimization (MLCO) has the ability to realize reconstructed samples with higher similarity, and the Distributed Gradient Inversion algorithm can implement GIA with batch sizes of 8 to 256 on deep network models (such as ResNet-50) and ImageNet datasets. A new Label Reconstruction algorithm is developed that relies only on the gradient information of the target model, which can achieve a label reconstruction accuracy of 81% in one batch sample with a label repetition rate of 96%, a 27% improvement over the state-of-the-art method. This proposed work can underpin data security assessments for healthcare federated learning. Code is available at https://github.com/zhaohuali/E2EGI .