Zhenhua Duan
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
Research Subjects
Zhenhua Duan has a total of 100 co-authors. Here are the top co-authors by number of shared publications:
- Cong Tian
Xidian University, Xi'an, China
49 shared publications - Nan Zhang
Xidian University
21 shared publications Cong Tian
Institute of Computing Theory and Technology and the State Key Laboratory of Integrated Services Networks, School of Computer Science and Technology, Xidian University, Xi’an, China
13 shared publicationsNan Zhang
Xidian University
13 shared publications- Bin Yu
Xidian University
9 shared publications - Xu Lu
Xidian University
8 shared publications
Latest Publications
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Parallel Combinational Equivalence Checking through Factored Form Sharing
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessParallel Combinational Equivalence Checking through Factored Form Sharing
- Open Access
- 24 July 2026
- Published by Springer Nature
- p. 284-306
Upper Bound for the Determinization of Emerson-Lei Automata: A One-Fin Approach
Emerson-Lei automata, which allow arbitrary Boolean combinations of $$\texttt{Fin}$$ and $$\texttt{Inf}$$ acceptance conditions, provide a unifying framework for $$\omega $$ -automata but pose significant challenges for determinization. The previous best algorithm relies on a transformation that introduces an exponential blow-up in the state space before determinization even begins. We present a new determinization algorithm that completely bypasses this bottleneck. Our key insight is that each disjunct of an Emerson-Lei condition in DNF corresponds directly to a one-Fin automaton —a restricted form of Streett automaton whose structure enables more efficient determinization via H-Safra trees. By exploiting this connection, we establish an upper bound of $$ 2^{O\!\big (3^{|\alpha |/3} \cdot (n \log n + n|\alpha | \log |\alpha |)\big )} $$ where n is the number of states and $$|\alpha |$$ is the acceptance condition size. This improves the exponent over the previous best bound by a factor of $$2^{2|\alpha |}/3^{|\alpha |/3}$$ , an exponential improvement in the acceptance condition complexity.- 24 July 2026
- Published by Springer Nature
- p. 284-306
Open AccessUpper Bound for the Determinization of Emerson-Lei Automata: A One-Fin Approach
Emerson-Lei automata, which allow arbitrary Boolean combinations of $$\texttt{Fin}$$ and $$\texttt{Inf}$$ acceptance conditions, provide a unifying framework for $$\omega $$ -automata but pose significant challenges for determinization. The previous best algorithm relies on a transformation that introduces an exponential blow-up in the state space before determinization even begins. We present a new determinization algorithm that completely bypasses this bottleneck. Our key insight is that each disjunct of an Emerson-Lei condition in DNF corresponds directly to a one-Fin automaton —a restricted form of Streett automaton whose structure enables more efficient determinization via H-Safra trees. By exploiting this connection, we establish an upper bound of $$ 2^{O\!\big (3^{|\alpha |/3} \cdot (n \log n + n|\alpha | \log |\alpha |)\big )} $$ where n is the number of states and $$|\alpha |$$ is the acceptance condition size. This improves the exponent over the previous best bound by a factor of $$2^{2|\alpha |}/3^{|\alpha |/3}$$ , an exponential improvement in the acceptance condition complexity. - Open Access
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 4666-4689
RAT: Retrieval-Augmented Testing of Certificate Revocation List Parsers in TLS Implementations
Transport Layer Security (TLS) implementations form the backbone of countless software systems, spanning web browsers, email clients, cloud services, and Internet of Things software, by enabling secure authentication and encrypted communication. However, their reliability hinges on the integrity of components like the Certificate Revocation List (CRL), which revokes compromised certificates to prevent attackers from exploiting expired or unauthorized credentials. Despite the CRL's critical role, CRL parsers, which decode CRL data for validation, remain overlooked in security research, exposing TLS-dependent software to potential threats. To address this gap, we introduce Retrieval-Augmented Testing (RAT), a framework powered by Large Language Models (LLMs), to systematically evaluate CRL parsers in mainstream TLS implementations such as OpenSSL, GnuTLS, and wolfSSL. RAT begins with leveraging an LLM to retrieve historical bug reports and cross-reference them with Request for Comments (RFC) 5280 specifications, generating structured test cases via an Abstract Syntax Notation One-aware mutation engine. These test cases are then fed into CRL parsers, and RAT employs an LLM to normalize their outputs. By analyzing these normalized results, RAT detects discrepancies and uncovers latent risks in CRL parsers. Our work makes the following contributions: (1) Unlike prior work focusing on certificate validation, this is the first study to systematically assess CRL parsers; (2) We propose RAT, a novel testing framework that leverages an LLM to integrate insights from retrieved bug reports and RFC 5280, enabling automated test-case generation; and (3) We have implemented an open-source prototype of RAT and experiments uncovered 23 new bugs, features, enhancements, fixes in commits, and x509s, demonstrating RAT's potential to strengthen the reliability and security of CRL parsers in TLS implementations.
- 30 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Software Engineering
- Vol. 3 (FSE) , 4666-4689
Open AccessRAT: Retrieval-Augmented Testing of Certificate Revocation List Parsers in TLS Implementations
Transport Layer Security (TLS) implementations form the backbone of countless software systems, spanning web browsers, email clients, cloud services, and Internet of Things software, by enabling secure authentication and encrypted communication. However, their reliability hinges on the integrity of components like the Certificate Revocation List (CRL), which revokes compromised certificates to prevent attackers from exploiting expired or unauthorized credentials. Despite the CRL's critical role, CRL parsers, which decode CRL data for validation, remain overlooked in security research, exposing TLS-dependent software to potential threats. To address this gap, we introduce Retrieval-Augmented Testing (RAT), a framework powered by Large Language Models (LLMs), to systematically evaluate CRL parsers in mainstream TLS implementations such as OpenSSL, GnuTLS, and wolfSSL. RAT begins with leveraging an LLM to retrieve historical bug reports and cross-reference them with Request for Comments (RFC) 5280 specifications, generating structured test cases via an Abstract Syntax Notation One-aware mutation engine. These test cases are then fed into CRL parsers, and RAT employs an LLM to normalize their outputs. By analyzing these normalized results, RAT detects discrepancies and uncovers latent risks in CRL parsers. Our work makes the following contributions: (1) Unlike prior work focusing on certificate validation, this is the first study to systematically assess CRL parsers; (2) We propose RAT, a novel testing framework that leverages an LLM to integrate insights from retrieved bug reports and RFC 5280, enabling automated test-case generation; and (3) We have implemented an open-source prototype of RAT and experiments uncovered 23 new bugs, features, enhancements, fixes in commits, and x509s, demonstrating RAT's potential to strengthen the reliability and security of CRL parsers in TLS implementations.
- Open Access
- 13 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-24
Enhanced Logic Rewriting with Intra- and Inter-graph Parallelism
Logic rewriting, as a critical and time-consuming task in synthesis, is widely employed in the integrated circuit (IC) design flow because it has the unique advantages of high optimization and independence from technology. However, existing solutions either employ locks to guarantee safe inter-node parallelism (at the cost of limiting parallelism), or parallelize the sub-procedures of rewriting for individual nodes without adequately considering logical sharing (at the cost of inevitably decreasing quality). In this article, we present DACPara 2.0, a fast, enhanced, and easily extensible parallel framework for high-quality logic rewriting. Our key insight is that, due to their characteristics, different large-scale circuits should adopt different parallel mechanisms to process this task enabling significant improvements in parallelism and scalability. In this spirit, for those with complex logic, we propose a divide-and-conquer parallel approach to exploit intra-graph parallelism (i.e., parallelism among nodes and their sub-procedures within the same And-Inverter Graph (AIG)), which separates three substages and redesigns them using dynamic global information. In this process, the nodes in an AIG are executed bottom-up in a level-wise parallel fashion. On the other hand, in the case of heavily pipelined industrial designs where each pipeline stage is represented as different copies of the same design, we propose a conflict-free sub-AIGs parallel approach featuring an ingenious fanout-based partitioning strategy to exploit inter-graph parallelism (i.e., parallelism between independent sub-AIGs). Experiments show that DACPara 2.0 using 40 CPU physical cores achieves 52.86×/42.25× speedup in rewriting/total runtime compared to logic rewriting in ABC, and 3.27×/2.61× speedup over the state-of-the-art CPU parallel method, with extremely comparable quality of result. Also, for all large-scale circuits with complex logic, DACPara 2.0 can achieve a 0.4% improvement in quality compared to the state-of-the-art GPU accelerated method.
- 13 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 31 (6) , 1-24
Open AccessEnhanced Logic Rewriting with Intra- and Inter-graph Parallelism
Logic rewriting, as a critical and time-consuming task in synthesis, is widely employed in the integrated circuit (IC) design flow because it has the unique advantages of high optimization and independence from technology. However, existing solutions either employ locks to guarantee safe inter-node parallelism (at the cost of limiting parallelism), or parallelize the sub-procedures of rewriting for individual nodes without adequately considering logical sharing (at the cost of inevitably decreasing quality). In this article, we present DACPara 2.0, a fast, enhanced, and easily extensible parallel framework for high-quality logic rewriting. Our key insight is that, due to their characteristics, different large-scale circuits should adopt different parallel mechanisms to process this task enabling significant improvements in parallelism and scalability. In this spirit, for those with complex logic, we propose a divide-and-conquer parallel approach to exploit intra-graph parallelism (i.e., parallelism among nodes and their sub-procedures within the same And-Inverter Graph (AIG)), which separates three substages and redesigns them using dynamic global information. In this process, the nodes in an AIG are executed bottom-up in a level-wise parallel fashion. On the other hand, in the case of heavily pipelined industrial designs where each pipeline stage is represented as different copies of the same design, we propose a conflict-free sub-AIGs parallel approach featuring an ingenious fanout-based partitioning strategy to exploit inter-graph parallelism (i.e., parallelism between independent sub-AIGs). Experiments show that DACPara 2.0 using 40 CPU physical cores achieves 52.86×/42.25× speedup in rewriting/total runtime compared to logic rewriting in ABC, and 3.27×/2.61× speedup over the state-of-the-art CPU parallel method, with extremely comparable quality of result. Also, for all large-scale circuits with complex logic, DACPara 2.0 can achieve a 0.4% improvement in quality compared to the state-of-the-art GPU accelerated method.
- Open Access
- 11 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 35 (9) , 1-33
AutoTestForge: A Multidimensional Automated Testing Framework for Natural Language Processing Models
In recent years, the application of behavioral testing in Natural Language Processing (NLP) model evaluation has experienced substantial growth. However, existing methods are restricted by reliance on manual labor and the limited scope of capability assessment. To address these limitations, we introduce AutoTestForge, an automated and multidimensional testing framework for NLP models. Through the integration of Large Language Models (LLMs) to automatically generate test templates and instantiate them, manual involvement is significantly reduced. Additionally, a mechanism for validating test case labels based on differential testing is proposed, which makes use of a multi-model voting system to guarantee the quality of test cases. The framework expands the test suite across three dimensions: taxonomy, fairness, and robustness, offering a comprehensive evaluation of the capabilities of NLP models. This expansion enables a in-depth and thorough assessment of the models, providing valuable insights into their strengths and weaknesses. A comprehensive evaluation across the sentiment analysis (SA) task and semantic textual similarity (STS) task demonstrates that AutoTestForge consistently outperforms existing datasets and testing tools with higher failure rates (an average of \(32.35\%\) for SA and \(31.61\%\) for STS). Moreover, different generation strategies exhibit stable effectiveness with failure rates ranging from \(25.77\%-38.04\%\) .- 11 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
- Vol. 35 (9) , 1-33
Open AccessAutoTestForge: A Multidimensional Automated Testing Framework for Natural Language Processing Models
In recent years, the application of behavioral testing in Natural Language Processing (NLP) model evaluation has experienced substantial growth. However, existing methods are restricted by reliance on manual labor and the limited scope of capability assessment. To address these limitations, we introduce AutoTestForge, an automated and multidimensional testing framework for NLP models. Through the integration of Large Language Models (LLMs) to automatically generate test templates and instantiate them, manual involvement is significantly reduced. Additionally, a mechanism for validating test case labels based on differential testing is proposed, which makes use of a multi-model voting system to guarantee the quality of test cases. The framework expands the test suite across three dimensions: taxonomy, fairness, and robustness, offering a comprehensive evaluation of the capabilities of NLP models. This expansion enables a in-depth and thorough assessment of the models, providing valuable insights into their strengths and weaknesses. A comprehensive evaluation across the sentiment analysis (SA) task and semantic textual similarity (STS) task demonstrates that AutoTestForge consistently outperforms existing datasets and testing tools with higher failure rates (an average of \(32.35\%\) for SA and \(31.61\%\) for STS). Moreover, different generation strategies exhibit stable effectiveness with failure rates ranging from \(25.77\%-38.04\%\) . - 4 June 2025
- Published by Elsevier in Neural Networks
- Vol. 190, 107656
Intra-head pruning for vision transformers via inter-layer dimension relationship modeling
- 4 June 2025
- Published by Elsevier in Neural Networks
- Vol. 190, 107656
Intra-head pruning for vision transformers via inter-layer dimension relationship modeling
- Open Access
ParMod: A Parallel and Modular Framework for Learning Non-Markovian Tasks
The commonly used Reinforcement Learning (RL) model, MDPs (Markov Decision Processes), has a basic premise that rewards depend on the current state and action only. However, many real-world tasks are non-Markovian, which has long-term memory and dependency. The reward sparseness problem is further amplified in non-Markovian scenarios. Hence learning a non-Markovian task (NMT) is inherently more difficult than learning a Markovian one. In this paper, we propose a novel \textbf{Par}allel and \textbf{Mod}ular RL framework, ParMod, specifically for learning NMTs specified by temporal logic. With the aid of formal techniques, the NMT is modulaized into a series of sub-tasks based on the automaton structure (equivalent to its temporal logic counterpart). On this basis, sub-tasks will be trained by a group of agents in a parallel fashion, with one agent handling one sub-task. Besides parallel training, the core of ParMod lies in: a flexible classification method for modularizing the NMT, and an effective reward shaping method for improving the sample efficiency. A comprehensive evaluation is conducted on several challenging benchmark problems with respect to various metrics. The experimental results show that ParMod achieves superior performance over other relevant studies. Our work thus provides a good synergy among RL, NMT and temporal logic.
Open AccessParMod: A Parallel and Modular Framework for Learning Non-Markovian Tasks
The commonly used Reinforcement Learning (RL) model, MDPs (Markov Decision Processes), has a basic premise that rewards depend on the current state and action only. However, many real-world tasks are non-Markovian, which has long-term memory and dependency. The reward sparseness problem is further amplified in non-Markovian scenarios. Hence learning a non-Markovian task (NMT) is inherently more difficult than learning a Markovian one. In this paper, we propose a novel \textbf{Par}allel and \textbf{Mod}ular RL framework, ParMod, specifically for learning NMTs specified by temporal logic. With the aid of formal techniques, the NMT is modulaized into a series of sub-tasks based on the automaton structure (equivalent to its temporal logic counterpart). On this basis, sub-tasks will be trained by a group of agents in a parallel fashion, with one agent handling one sub-task. Besides parallel training, the core of ParMod lies in: a flexible classification method for modularizing the NMT, and an effective reward shaping method for improving the sample efficiency. A comprehensive evaluation is conducted on several challenging benchmark problems with respect to various metrics. The experimental results show that ParMod achieves superior performance over other relevant studies. Our work thus provides a good synergy among RL, NMT and temporal logic.
- 11 October 2024
- Published by Elsevier in Journal of Systems Architecture
- Vol. 156
Verifiable privacy-preserving semantic retrieval scheme in the edge computing
- 11 October 2024
- Published by Elsevier in Journal of Systems Architecture
- Vol. 156
Verifiable privacy-preserving semantic retrieval scheme in the edge computing
- 7 September 2024
- Published by Elsevier in Neural Networks
- Vol. 180, 106708
Intermediate-grained kernel elements pruning with structured sparsity
- 7 September 2024
- Published by Elsevier in Neural Networks
- Vol. 180, 106708
Intermediate-grained kernel elements pruning with structured sparsity
- 14 July 2024
- Published by Springer Nature
- p. 58-72
An Approach to Improving Reliability of Parallel Graph Computation
- 14 July 2024
- Published by Springer Nature
- p. 58-72
An Approach to Improving Reliability of Parallel Graph Computation