Hailong Ma
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hailong Ma has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
Tao Hu
Information Engineering University
4 shared publications- Tao Hu
Information Engineering University, Zhengzhou, China
4 shared publications - Yanze Qu
Information Engineering University, Zhengzhou, China
4 shared publications Liang Wang
National Digital Switching System Engineering and Technology Research Center, PLA Strategic Support Force Information Engineering University, Zhengzhou, China
3 shared publications- Peng Yi
Information Engineering University, Zhengzhou, China
3 shared publications - Xinlei Liu
Information Engineering University, Zhengzhou, China
3 shared publications
Latest Publications
- Open Access
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 5373-5387
SymSan: Mitigating Malware Embedding in Large Language Models via Parameter Space Symmetry
- 14 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 5373-5387
Open AccessSymSan: Mitigating Malware Embedding in Large Language Models via Parameter Space Symmetry
- Open Access
- 22 October 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (21) , 10446
FedSTIP-Tuning: Federated Black-Box Prompt Tuning with Aggregate-Only Feedback and Common-Candidate Optimization
Federated black-box prompt tuning enables organizations to adapt a proprietary frozen language model without centralizing their labeled data, but local search-state aggregation does not directly define a shared pooled-data objective, and client-specific feedback can expose private signals. We propose FedSTIP-Tuning, which composes a three-party permutation-based protected forward propagation with a common-candidate population and pairwise-masked loss aggregation in a cross-silo setting. All participating clients evaluate the same prompt candidates, and the coordinator ranks their sample-weighted aggregate losses, thereby optimizing the empirical objective over the fixed union of client shards under full participation. Under the stated semi-honest and non-collusion assumptions, the aggregation layer hides each released loss vector’s client-level decomposition, while the aggregate values and execution metadata remain observable to their authorized parties. On RoBERTa-large, FedSTIP-Tuning exceeds the strongest evaluated FedBPT configuration by 6.40, 2.66, and 7.60 percentage points on AG News, SST-2, and DBpedia, respectively, under matched forward budgets. Its protected forward preserves all candidate rankings across 600 paired generations and is 861× and 957× faster than a generic interactive two-party computation reference on the RoBERTa-base warm and cold paths, and 694× and 578× on RoBERTa-large.
- 22 October 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (21) , 10446
Open AccessFedSTIP-Tuning: Federated Black-Box Prompt Tuning with Aggregate-Only Feedback and Common-Candidate Optimization
Federated black-box prompt tuning enables organizations to adapt a proprietary frozen language model without centralizing their labeled data, but local search-state aggregation does not directly define a shared pooled-data objective, and client-specific feedback can expose private signals. We propose FedSTIP-Tuning, which composes a three-party permutation-based protected forward propagation with a common-candidate population and pairwise-masked loss aggregation in a cross-silo setting. All participating clients evaluate the same prompt candidates, and the coordinator ranks their sample-weighted aggregate losses, thereby optimizing the empirical objective over the fixed union of client shards under full participation. Under the stated semi-honest and non-collusion assumptions, the aggregation layer hides each released loss vector’s client-level decomposition, while the aggregate values and execution metadata remain observable to their authorized parties. On RoBERTa-large, FedSTIP-Tuning exceeds the strongest evaluated FedBPT configuration by 6.40, 2.66, and 7.60 percentage points on AG News, SST-2, and DBpedia, respectively, under matched forward budgets. Its protected forward preserves all candidate rankings across 600 paired generations and is 861× and 957× faster than a generic interactive two-party computation reference on the RoBERTa-base warm and cold paths, and 694× and 578× on RoBERTa-large.
- Open Access
- 11 August 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 204
SepSpace: payload-free open-set network traffic classification via supervised contrastive representation learning and prototype-radius rejection
Network traffic classification (NTC) increasingly needs to go beyond predefined categories and identify traffic from previously unseen classes, namely the open-set recognition problem in NTC. This problem remains challenging because encrypted traffic limits the practicality of payload-dependent features, while existing methods still lack representation learning tailored to open-set discrimination. This study aims to develop a payload-free open-set NTC framework that constructs an embedding space suitable for both known-class classification and unseen-class rejection. To this end, we propose SepSpace, which models network flows using meta-feature sequences. SepSpace introduces supervised contrastive learning to construct a discriminative embedding space with compact intra-class structures and separated inter-class boundaries. Combined with a prototype-radius rejection mechanism, SepSpace achieves effective unseen-class rejection while maintaining known-class classification performance. Meanwhile, to comprehensively evaluate open-set performance, we introduce a class-holdout evaluation protocol and conduct experiments on four benchmark datasets together with the conventional cross-dataset protocol. Experimental results demonstrate that, under the class-holdout protocol, SepSpace achieves the highest average area under the receiver operating characteristic curve (AUROC) of 0.9015 and open-set classification rate (OSCR) of 0.6351 among the evaluated methods; under the cross-dataset protocol, it obtains a higher average AUROC of 0.9396. The implementation and datasets of SepSpace are available at https://github.com/BITbla/SepSpace .
- 11 August 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 204
Open AccessSepSpace: payload-free open-set network traffic classification via supervised contrastive representation learning and prototype-radius rejection
Network traffic classification (NTC) increasingly needs to go beyond predefined categories and identify traffic from previously unseen classes, namely the open-set recognition problem in NTC. This problem remains challenging because encrypted traffic limits the practicality of payload-dependent features, while existing methods still lack representation learning tailored to open-set discrimination. This study aims to develop a payload-free open-set NTC framework that constructs an embedding space suitable for both known-class classification and unseen-class rejection. To this end, we propose SepSpace, which models network flows using meta-feature sequences. SepSpace introduces supervised contrastive learning to construct a discriminative embedding space with compact intra-class structures and separated inter-class boundaries. Combined with a prototype-radius rejection mechanism, SepSpace achieves effective unseen-class rejection while maintaining known-class classification performance. Meanwhile, to comprehensively evaluate open-set performance, we introduce a class-holdout evaluation protocol and conduct experiments on four benchmark datasets together with the conventional cross-dataset protocol. Experimental results demonstrate that, under the class-holdout protocol, SepSpace achieves the highest average area under the receiver operating characteristic curve (AUROC) of 0.9015 and open-set classification rate (OSCR) of 0.6351 among the evaluated methods; under the cross-dataset protocol, it obtains a higher average AUROC of 0.9396. The implementation and datasets of SepSpace are available at https://github.com/BITbla/SepSpace .
- 14 July 2026
- Published by Springer Nature
- p. 556-567
Zero-Shot Backdoor Defense via Singular Value Truncation
- 14 July 2026
- Published by Springer Nature
- p. 556-567
Zero-Shot Backdoor Defense via Singular Value Truncation
- 13 July 2026
- Published by Springer Nature
- p. 313-324
DynGuard: Dynamic Runtime Isolation for Entangled Logical Backdoors in LLMs
- 13 July 2026
- Published by Springer Nature
- p. 313-324
DynGuard: Dynamic Runtime Isolation for Entangled Logical Backdoors in LLMs
- 12 July 2026
- Published by Springer Nature
- p. 520-536
Minimal-Overhead Backdoor Detection via Entropy-Guided Activation Substitution
- 12 July 2026
- Published by Springer Nature
- p. 520-536
Minimal-Overhead Backdoor Detection via Entropy-Guided Activation Substitution
- Open Access
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 2008-2012
Alternating Guided Training for Robust Adversarial Defense
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 2008-2012
Open AccessAlternating Guided Training for Robust Adversarial Defense
- Open Access
- 13 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (12) , 2422
A Survey on Reinforcement Learning-Driven Adversarial Sample Generation for PE Malware
Malware remains a central tool in cyberattacks, and systematic research into adversarial attack techniques targeting malware is crucial in advancing detection and defense systems that can evolve over time. Although numerous review articles already exist in this area, there is still a lack of comprehensive exploration into emerging artificial intelligence technologies such as reinforcement learning from the attacker’s perspective. To address this gap, we propose a foundational reinforcement learning (RL)-based framework for adversarial malware generation and develop a systematic evaluation methodology to dissect the internal mechanisms of generative models across multiple key dimensions, including action space design, state space representation, and reward function construction. Drawing from a comprehensive review and synthesis of the existing literature, we identify several core findings. (1) The scale of the action space directly affects the model training efficiency. Meanwhile, factors such as the action diversity, operation determinism, execution order, and modification ratio indirectly influence the quality of the generated adversarial samples. (2) Comprehensive and sensitive state feature representations can compensate for the information loss caused by binary feedback from real-world detection engines, thereby enhancing both the effectiveness and stability of attacks. (3) A multi-dimensional reward signal effectively mitigates the policy fragility associated with single-metric rewards, improving the agent’s adaptability in complex environments. (4) While the current RL frameworks applied to malware generation exhibit diverse architectures, they share a common core: the modeling of discrete action spaces and continuous state spaces. In addition, this work explores future research directions in the area of adversarial malware generation and outlines the open challenges and critical issues faced by defenders in responding to such threats. Our goal is to provide both a theoretical foundation and practical guidance for building more robust and adaptive security detection mechanisms.
- 13 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (12) , 2422
Open AccessA Survey on Reinforcement Learning-Driven Adversarial Sample Generation for PE Malware
Malware remains a central tool in cyberattacks, and systematic research into adversarial attack techniques targeting malware is crucial in advancing detection and defense systems that can evolve over time. Although numerous review articles already exist in this area, there is still a lack of comprehensive exploration into emerging artificial intelligence technologies such as reinforcement learning from the attacker’s perspective. To address this gap, we propose a foundational reinforcement learning (RL)-based framework for adversarial malware generation and develop a systematic evaluation methodology to dissect the internal mechanisms of generative models across multiple key dimensions, including action space design, state space representation, and reward function construction. Drawing from a comprehensive review and synthesis of the existing literature, we identify several core findings. (1) The scale of the action space directly affects the model training efficiency. Meanwhile, factors such as the action diversity, operation determinism, execution order, and modification ratio indirectly influence the quality of the generated adversarial samples. (2) Comprehensive and sensitive state feature representations can compensate for the information loss caused by binary feedback from real-world detection engines, thereby enhancing both the effectiveness and stability of attacks. (3) A multi-dimensional reward signal effectively mitigates the policy fragility associated with single-metric rewards, improving the agent’s adaptability in complex environments. (4) While the current RL frameworks applied to malware generation exhibit diverse architectures, they share a common core: the modeling of discrete action spaces and continuous state spaces. In addition, this work explores future research directions in the area of adversarial malware generation and outlines the open challenges and critical issues faced by defenders in responding to such threats. Our goal is to provide both a theoretical foundation and practical guidance for building more robust and adaptive security detection mechanisms.
- Open Access
- 15 August 2024
- Published by MDPI AG in Electronics
- Vol. 13 (16) , 3246
SMWE: A Framework for Secure and Makespan-Oriented Workflow Execution in Serverless Computing
Serverless computing is a promising paradigm that greatly simplifies cloud programming. With serverless computing, developers simply provide event-driven functions to a serverless platform, and these functions can be orchestrated as serverless workflows to accomplish complex tasks. Due to the lightweight limitation of functions, serverless workflows not only suffer from existing vulnerability-based threats but also face new security threats from the function compiling phase. In this paper, we present SMWE, a secure and makespan-oriented workflow execution framework in serverless computing. SMWE enables all life cycle protection for functions by adopting compiler shifting and running environment replacement in the serverless workflow. Furthermore, SMWE balances the tradeoff between security and makespan by carefully scheduling functions to running environments and selectively applying the secure techniques to functions. Extensive evaluations show that SMWE significantly increases the security of serverless workflows with small makespan cost.
- 15 August 2024
- Published by MDPI AG in Electronics
- Vol. 13 (16) , 3246
Open AccessSMWE: A Framework for Secure and Makespan-Oriented Workflow Execution in Serverless Computing
Serverless computing is a promising paradigm that greatly simplifies cloud programming. With serverless computing, developers simply provide event-driven functions to a serverless platform, and these functions can be orchestrated as serverless workflows to accomplish complex tasks. Due to the lightweight limitation of functions, serverless workflows not only suffer from existing vulnerability-based threats but also face new security threats from the function compiling phase. In this paper, we present SMWE, a secure and makespan-oriented workflow execution framework in serverless computing. SMWE enables all life cycle protection for functions by adopting compiler shifting and running environment replacement in the serverless workflow. Furthermore, SMWE balances the tradeoff between security and makespan by carefully scheduling functions to running environments and selectively applying the secure techniques to functions. Extensive evaluations show that SMWE significantly increases the security of serverless workflows with small makespan cost.
- Open Access
- 6 March 2023
- Published by MDPI AG in Electronics
- Vol. 12 (5) , 1255
A Network Intrusion Detection Method Based on Domain Confusion
Network intrusion detection models based on deep learning encounter problems in the migration application. The performance is not as good as expected. In this paper, a network intrusion detection method based on domain confusion is proposed to improve the migration performance of the model. A domain confusion network is designed for feature transformation based on the idea of domain adaptation, mapping the traffic data in different network environments to the same feature space. Meanwhile, a regularizer is proposed to control the information loss in the mapping process to ensure that the transformed feature obtains enough information for intrusion detection. The experiment results show that the detection performance of the model in this paper is similar to or even better than the traditional models, and the migration performance in different network environments is better than the traditional models.
- 6 March 2023
- Published by MDPI AG in Electronics
- Vol. 12 (5) , 1255
Open AccessA Network Intrusion Detection Method Based on Domain Confusion
Network intrusion detection models based on deep learning encounter problems in the migration application. The performance is not as good as expected. In this paper, a network intrusion detection method based on domain confusion is proposed to improve the migration performance of the model. A domain confusion network is designed for feature transformation based on the idea of domain adaptation, mapping the traffic data in different network environments to the same feature space. Meanwhile, a regularizer is proposed to control the information loss in the mapping process to ensure that the transformed feature obtains enough information for intrusion detection. The experiment results show that the detection performance of the model in this paper is similar to or even better than the traditional models, and the migration performance in different network environments is better than the traditional models.