Zhi Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Zhi Wang has a total of 101 co-authors. Here are the top co-authors by number of shared publications:
- Wanpeng Li
University of Liverpool
8 shared publications - Chunbo Liu
Information Security Evaluation Center, Civil Aviation University of China, Tianjin 300300, China
6 shared publications - Shuangshuang Liu
DISSec, College of Cyber Science, Nankai University, Tianjin, China
4 shared publications - Xiaoyu Cheng
Nankai University
4 shared publications - Zhaojun Gu
Information Security Evaluation Center, Civil Aviation University of China, Tianjin 300300, China
4 shared publications - Haotong Liu
College of Cryptology and Cyber Science, Nankai University, Tianjin, China
3 shared publications
Latest Publications
- Open Access
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7429-7431
POSTER: Beyond Pickle-Free: Loading-Boundary Failures in Safetensors
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 7429-7431
Open AccessPOSTER: Beyond Pickle-Free: Loading-Boundary Failures in Safetensors
- Open Access
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4866-4880
As Seasons Change: Temporal Feature Powers Malware Classification
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4866-4880
Open AccessAs Seasons Change: Temporal Feature Powers Malware Classification
- 19 July 2026
- Published by Springer Nature
- p. 13-24
PerCheck: An AST-Based Framework for Detecting Over-Privilege in Chrome Extensions
- 19 July 2026
- Published by Springer Nature
- p. 13-24
PerCheck: An AST-Based Framework for Detecting Over-Privilege in Chrome Extensions
- 13 July 2026
- Published by Springer Nature
- p. 27-38
A Hierarchical Adaptive Belief Rule Base for Chrome Extension Risk Assessment
- 13 July 2026
- Published by Springer Nature
- p. 27-38
A Hierarchical Adaptive Belief Rule Base for Chrome Extension Risk Assessment
- Open Access
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 3793-3815
DECODE: Dynamic Exploration for Constraint-Guided Vulnerability Discovery in Deep Learning Operators
The security and robustness of deep learning (DL) frameworks are vital, as vulnerabilities in low-level operator implementations can lead to serious reliability and security risks. While testing has proven effective in uncovering such flaws, existing techniques struggle to accurately capture the complex input constraints required by DL operators, resulting in low test coverage and missed bugs. To address this, we propose DECODE, a fully automated framework that performs efficient and precise constraint extraction through dynamic analysis. DECODE models operator-specific input requirements by observing valid execution traces and exploring constraint relationships. It then uses these refined constraints to generate high-quality test inputs capable of exposing memory error vulnerabilities. We evaluated DECODE on two widely used DL frameworks - TensorFlow and PyTorch - where it uncovered 96 bugs (54 in TensorFlow and 42 in PyTorch), 82 of which have been confirmed by developers. Compared to state-of-the-art tools, DECODE detected 41, 75, and 87 more bugs than IvySyn, DocTer, and DeepREL, respectively, demonstrating its superior effectiveness in uncovering previously undetected vulnerabilities.
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 3793-3815
Open AccessDECODE: Dynamic Exploration for Constraint-Guided Vulnerability Discovery in Deep Learning Operators
The security and robustness of deep learning (DL) frameworks are vital, as vulnerabilities in low-level operator implementations can lead to serious reliability and security risks. While testing has proven effective in uncovering such flaws, existing techniques struggle to accurately capture the complex input constraints required by DL operators, resulting in low test coverage and missed bugs. To address this, we propose DECODE, a fully automated framework that performs efficient and precise constraint extraction through dynamic analysis. DECODE models operator-specific input requirements by observing valid execution traces and exploring constraint relationships. It then uses these refined constraints to generate high-quality test inputs capable of exposing memory error vulnerabilities. We evaluated DECODE on two widely used DL frameworks - TensorFlow and PyTorch - where it uncovered 96 bugs (54 in TensorFlow and 42 in PyTorch), 82 of which have been confirmed by developers. Compared to state-of-the-art tools, DECODE detected 41, 75, and 87 more bugs than IvySyn, DocTer, and DeepREL, respectively, demonstrating its superior effectiveness in uncovering previously undetected vulnerabilities.
- Open Access
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1200-1212
CRX-ray: Large-Scale Detection of API Key Leakage in Browser Extensions
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1200-1212
Open AccessCRX-ray: Large-Scale Detection of API Key Leakage in Browser Extensions
- 21 April 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (9)
A Belief Rule Base Method for Modeling Latent States with Uncertainty in Chrome Extension Risk Assessment
Chrome extensions, as popular browser functionality enhancement tools, often pose privacy risks due to permission-related issues. To address this, we propose a novel permission-based extension evaluation framework: RRIBRB-ExtPA (Rule Reduction Based Interval Belief Rule Base for Extension Permission-based Assessment, RRIBRB-ExtPA). This framework integrates our improved RRIBRB model. It reduces the dimension of precondition attributes. Rule combination and activation mechanisms are also optimized. Together, these improvements mitigate the rule explosion problem inherent in BRB models. Additionally, we introduce an overprivilege detection method to identify permission abuse behaviors in extensions, enhancing overall risk assessment capabilities. Experimental results demonstrate that RRIBRB-ExtPA maintains an accuracy rate of 98.6% even with an 87.5\% reduction in the number of rules, while exhibiting stable performance under small-sample conditions. This indicates that RRIBRB-ExtPA is a viable permission-based extension risk assessment framework.
- 21 April 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (9)
A Belief Rule Base Method for Modeling Latent States with Uncertainty in Chrome Extension Risk Assessment
Chrome extensions, as popular browser functionality enhancement tools, often pose privacy risks due to permission-related issues. To address this, we propose a novel permission-based extension evaluation framework: RRIBRB-ExtPA (Rule Reduction Based Interval Belief Rule Base for Extension Permission-based Assessment, RRIBRB-ExtPA). This framework integrates our improved RRIBRB model. It reduces the dimension of precondition attributes. Rule combination and activation mechanisms are also optimized. Together, these improvements mitigate the rule explosion problem inherent in BRB models. Additionally, we introduce an overprivilege detection method to identify permission abuse behaviors in extensions, enhancing overall risk assessment capabilities. Experimental results demonstrate that RRIBRB-ExtPA maintains an accuracy rate of 98.6% even with an 87.5\% reduction in the number of rules, while exhibiting stable performance under small-sample conditions. This indicates that RRIBRB-ExtPA is a viable permission-based extension risk assessment framework.
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 3673-3685
An Empirical Study on Static Application Security Testing (SAST) Tools for Python
Python is currently the most popular programming language and ensuring the security of Python applications has become a critical concern. Static Application Security Testing (SAST) tools have been introduced to address this need, claiming to support a wide range of Common Weakness Enumerations (CWEs). However, the ability of these tools to detect real-world vulnerabilities in Python programs has not been comprehensively evaluated. In this paper, we selected eight SAST tools from 117 existing ones based on well-designed criteria. Based on the synthetic and real-world dataset, we evaluated and compared these SAST tools from different perspectives including effectiveness and efficiency. Our results reveal significant limitations in current SAST tools: although perform well on the synthetic dataset, a single tool detects no more than 40% of the vulnerabilities in our real-world dataset. Even when aggregating the outputs of all evaluated tools, only 66.7% of the real-world vulnerabilities are identified. To further understand these shortcomings, we performed a root cause analysis of the detection results and identified useful insights for both SAST tool designers and users, focusing on tool development, evaluation, and selection.
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 3673-3685
Open AccessAn Empirical Study on Static Application Security Testing (SAST) Tools for Python
Python is currently the most popular programming language and ensuring the security of Python applications has become a critical concern. Static Application Security Testing (SAST) tools have been introduced to address this need, claiming to support a wide range of Common Weakness Enumerations (CWEs). However, the ability of these tools to detect real-world vulnerabilities in Python programs has not been comprehensively evaluated. In this paper, we selected eight SAST tools from 117 existing ones based on well-designed criteria. Based on the synthetic and real-world dataset, we evaluated and compared these SAST tools from different perspectives including effectiveness and efficiency. Our results reveal significant limitations in current SAST tools: although perform well on the synthetic dataset, a single tool detects no more than 40% of the vulnerabilities in our real-world dataset. Even when aggregating the outputs of all evaluated tools, only 66.7% of the real-world vulnerabilities are identified. To further understand these shortcomings, we performed a root cause analysis of the detection results and identified useful insights for both SAST tool designers and users, focusing on tool development, evaluation, and selection.
- Open Access
LogICL: Distilling LLM Reasoning to Bridge the Semantic Gap in Cross-Domain Log Anomaly Detection
Effective log anomaly detection is critical to sustaining reliability in large-scale IT infrastructures. Transformer-based models require substantial resources and labeled data, exacerbating the cold-start problem in target domains where logs are scarce. Existing cross-domain methods leverage source logs but struggle with generalization due to reliance on surface lexical similarity, failing to capture latent semantic equivalence amid structural divergences. To address this, we propose LogICL, a framework distilling Large Language Model (LLM) reasoning into a lightweight encoder for cross-domain anomaly detection. During training, LogICL constructs a delta matrix measuring the utility of demonstrations selected via Maximal Marginal Relevance relative to zero-shot inference. The encoder is optimized via a multi-objective loss comprising an ICL-Guided term that aligns representations based on reasoning assistance utility, maximum mean discrepancy for domain alignment, and supervised contrastive loss. At inference, the optimized encoder retrieves reasoning-aware demonstrations using semantic similarity and delta scores, enabling frozen-LLM in-context learning with Chain-of-Thought for accurate and interpretable detection. Experiments on few-shot and zero-shot cross-domain benchmarks confirm LogICL achieves state-of-the-art performance across heterogeneous systems. Further analysis via visualizations and case studies confirms LogICL bridges the semantic gap beyond surface lexical similarity, effectively capturing latent semantic equivalence for rapid deployment.
Open AccessLogICL: Distilling LLM Reasoning to Bridge the Semantic Gap in Cross-Domain Log Anomaly Detection
Effective log anomaly detection is critical to sustaining reliability in large-scale IT infrastructures. Transformer-based models require substantial resources and labeled data, exacerbating the cold-start problem in target domains where logs are scarce. Existing cross-domain methods leverage source logs but struggle with generalization due to reliance on surface lexical similarity, failing to capture latent semantic equivalence amid structural divergences. To address this, we propose LogICL, a framework distilling Large Language Model (LLM) reasoning into a lightweight encoder for cross-domain anomaly detection. During training, LogICL constructs a delta matrix measuring the utility of demonstrations selected via Maximal Marginal Relevance relative to zero-shot inference. The encoder is optimized via a multi-objective loss comprising an ICL-Guided term that aligns representations based on reasoning assistance utility, maximum mean discrepancy for domain alignment, and supervised contrastive loss. At inference, the optimized encoder retrieves reasoning-aware demonstrations using semantic similarity and delta scores, enabling frozen-LLM in-context learning with Chain-of-Thought for accurate and interpretable detection. Experiments on few-shot and zero-shot cross-domain benchmarks confirm LogICL achieves state-of-the-art performance across heterogeneous systems. Further analysis via visualizations and case studies confirms LogICL bridges the semantic gap beyond surface lexical similarity, effectively capturing latent semantic equivalence for rapid deployment.
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (4) , 7680-7694
Enhanced Authentication and Key Agreement Protocol for VANETs
Vehicular ad hoc networks (VANETs) promote the rapid development of intelligent transportation by connecting vehicles, road infrastructures, and other devices through wireless communication. However, due to their highly dynamic nature, high-speed vehicle movement, frequent changes in network topology, and device heterogeneity, VANETs face more complex security challenges than general wireless networks, such as man-in-the-middle (MITM) attacks, forged nodes, data tampering, and privacy leakage. To tackle these challenges, authentication and key agreement protocols play a crucial role in ensuring communication security and privacy protection. Awais et al. proposed a novel authentication and key agreement protocol. Although they claim their scheme can resist a wide range of common attacks, we have found that the protocol cannot resist stolen smart card attacks, stolen verifier attacks, and temporary random number leakage attacks. In this article, we comprehensively analyze the security vulnerabilities present in the Awais et al. protocol and introduce an enhanced authentication and key agreement protocol. We demonstrate the enhanced protocol’s security through formal (random oracle model (ROM), ProVerif, and Burrows–Abadi–Needham (BAN) logic) and informal security analysis. In addition, the experimental results show that the time cost of this protocol is 1.3477 ms, which is significantly lower than the 1.9339 ms of Awais et al.’s protocol and other related protocols; in terms of the communication cost, this protocol is only 3742 bits, which is also better than the 3904 bits of Awais et al.’s protocol and other schemes. Comprehensive comparison shows that our proposed protocol has significant advantages in performance and efficiency while guaranteeing security.
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (4) , 7680-7694
Enhanced Authentication and Key Agreement Protocol for VANETs
Vehicular ad hoc networks (VANETs) promote the rapid development of intelligent transportation by connecting vehicles, road infrastructures, and other devices through wireless communication. However, due to their highly dynamic nature, high-speed vehicle movement, frequent changes in network topology, and device heterogeneity, VANETs face more complex security challenges than general wireless networks, such as man-in-the-middle (MITM) attacks, forged nodes, data tampering, and privacy leakage. To tackle these challenges, authentication and key agreement protocols play a crucial role in ensuring communication security and privacy protection. Awais et al. proposed a novel authentication and key agreement protocol. Although they claim their scheme can resist a wide range of common attacks, we have found that the protocol cannot resist stolen smart card attacks, stolen verifier attacks, and temporary random number leakage attacks. In this article, we comprehensively analyze the security vulnerabilities present in the Awais et al. protocol and introduce an enhanced authentication and key agreement protocol. We demonstrate the enhanced protocol’s security through formal (random oracle model (ROM), ProVerif, and Burrows–Abadi–Needham (BAN) logic) and informal security analysis. In addition, the experimental results show that the time cost of this protocol is 1.3477 ms, which is significantly lower than the 1.9339 ms of Awais et al.’s protocol and other related protocols; in terms of the communication cost, this protocol is only 3742 bits, which is also better than the 3904 bits of Awais et al.’s protocol and other schemes. Comprehensive comparison shows that our proposed protocol has significant advantages in performance and efficiency while guaranteeing security.