Chen Ma
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
Chen Ma has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Qi Xuan
Zhejiang University of Technology
4 shared publications Li Chen
School of Software BNRist Tsinghua University Beijing China
3 shared publications- Hanzhe Yu
Institute of Cyberspace Security, Zhejiang University of Technology, Hangzhou 310023, China
2 shared publications Junhai Yong
Tsinghua University & BNRist, Beijing, China
2 shared publicationsShuyu Cheng
JQ Investments 200122 Shanghai China
2 shared publicationsXinjie Xu
Institute of Cyberspace Security Zhejiang University of Technology 310023 Hangzhou China
2 shared publications
Latest Publications
- Open Access
Query-Efficient Hard-Label Attack: A Prior-Guided Adam Ray Search Optimization
Deep neural networks are vulnerable to adversarial examples, even in hard-label black-box settings where only the top-1 prediction is available. To address the challenges of high-dimensional optimization under limited query budgets, we propose two query-efficient attack methods: Adam-OPT, which integrates Adam-based adaptive optimization into the ray-search framework to stabilize and accelerate zeroth-order gradient updates; Prior-Adam-OPT, which further incorporates transfer-based priors from surrogate models to enhance gradient estimation. Adam-OPT leverages historical gradient information and per-parameter adaptive updates to improve convergence, while Prior-Adam-OPT constructs a prior-guided orthogonal search basis that combines surrogate and random directions, enhancing both gradient accuracy and query efficiency. Our approach demonstrates superior performance across CIFAR-10, ImageNet, and zero-shot CLIP models, consistently reducing perturbation magnitudes and improving attack efficiency compared to state-of-the-art hard-label attacks. Ablation studies highlight the importance of the number of vectors used for gradient estimation and the quality of surrogate models, showing that combining adaptive optimization with transfer-based priors provides a scalable and robust framework for generating high-quality adversarial examples in challenging black-box scenarios.
Open AccessQuery-Efficient Hard-Label Attack: A Prior-Guided Adam Ray Search Optimization
Deep neural networks are vulnerable to adversarial examples, even in hard-label black-box settings where only the top-1 prediction is available. To address the challenges of high-dimensional optimization under limited query budgets, we propose two query-efficient attack methods: Adam-OPT, which integrates Adam-based adaptive optimization into the ray-search framework to stabilize and accelerate zeroth-order gradient updates; Prior-Adam-OPT, which further incorporates transfer-based priors from surrogate models to enhance gradient estimation. Adam-OPT leverages historical gradient information and per-parameter adaptive updates to improve convergence, while Prior-Adam-OPT constructs a prior-guided orthogonal search basis that combines surrogate and random directions, enhancing both gradient accuracy and query efficiency. Our approach demonstrates superior performance across CIFAR-10, ImageNet, and zero-shot CLIP models, consistently reducing perturbation magnitudes and improving attack efficiency compared to state-of-the-art hard-label attacks. Ablation studies highlight the importance of the number of vectors used for gradient estimation and the quality of surrogate models, showing that combining adaptive optimization with transfer-based priors provides a scalable and robust framework for generating high-quality adversarial examples in challenging black-box scenarios.
- Open Access
Improving the Convergence Rate of Ray Search Optimization for Query-Efficient Hard-Label Attacks
In hard-label black-box adversarial attacks, where only the top-1 predicted label is accessible, the prohibitive query complexity poses a major obstacle to practical deployment. In this paper, we focus on optimizing a representative class of attacks that search for the optimal ray direction yielding the minimum $\ell_2$-norm perturbation required to move a benign image into the adversarial region. Inspired by Nesterov's Accelerated Gradient (NAG), we propose a momentum-based algorithm, ARS-OPT, which proactively estimates the gradient with respect to a future ray direction inferred from accumulated momentum. We provide a theoretical analysis of its convergence behavior, showing that ARS-OPT enables more accurate directional updates and achieves faster, more stable optimization. To further accelerate convergence, we incorporate surrogate-model priors into ARS-OPT's gradient estimation, resulting in PARS-OPT with enhanced performance. The superiority of our approach is supported by theoretical guarantees under standard assumptions. Extensive experiments on ImageNet and CIFAR-10 demonstrate that our method surpasses 13 state-of-the-art approaches in query efficiency.
Open AccessImproving the Convergence Rate of Ray Search Optimization for Query-Efficient Hard-Label Attacks
In hard-label black-box adversarial attacks, where only the top-1 predicted label is accessible, the prohibitive query complexity poses a major obstacle to practical deployment. In this paper, we focus on optimizing a representative class of attacks that search for the optimal ray direction yielding the minimum $\ell_2$-norm perturbation required to move a benign image into the adversarial region. Inspired by Nesterov's Accelerated Gradient (NAG), we propose a momentum-based algorithm, ARS-OPT, which proactively estimates the gradient with respect to a future ray direction inferred from accumulated momentum. We provide a theoretical analysis of its convergence behavior, showing that ARS-OPT enables more accurate directional updates and achieves faster, more stable optimization. To further accelerate convergence, we incorporate surrogate-model priors into ARS-OPT's gradient estimation, resulting in PARS-OPT with enhanced performance. The superiority of our approach is supported by theoretical guarantees under standard assumptions. Extensive experiments on ImageNet and CIFAR-10 demonstrate that our method surpasses 13 state-of-the-art approaches in query efficiency.
- 9 November 2025
- Published by Elsevier in Information Fusion
- Vol. 127
Generative recommender systems: A comprehensive survey on model, framework, and application
- 9 November 2025
- Published by Elsevier in Information Fusion
- Vol. 127
Generative recommender systems: A comprehensive survey on model, framework, and application
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 11394-11403
RealHD: A High-Quality Dataset for Robust Detection of State-of-the-Art AI-Generated Images
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 11394-11403
Open AccessRealHD: A High-Quality Dataset for Robust Detection of State-of-the-Art AI-Generated Images
- Open Access
Boosting Ray Search Procedure of Hard-label Attacks with Transfer-based Priors
One of the most practical and challenging types of black-box adversarial attacks is the hard-label attack, where only the top-1 predicted label is available. One effective approach is to search for the optimal ray direction from the benign image that minimizes the $\ell_p$-norm distance to the adversarial region. The unique advantage of this approach is that it transforms the hard-label attack into a continuous optimization problem. The objective function value is the ray's radius, which can be obtained via binary search at a high query cost. Existing methods use a "sign trick" in gradient estimation to reduce the number of queries. In this paper, we theoretically analyze the quality of this gradient estimation and propose a novel prior-guided approach to improve ray search efficiency both theoretically and empirically. Specifically, we utilize the transfer-based priors from surrogate models, and our gradient estimators appropriately integrate them by approximating the projection of the true gradient onto the subspace spanned by these priors and random directions, in a query-efficient manner. We theoretically derive the expected cosine similarities between the obtained gradient estimators and the true gradient, and demonstrate the improvement achieved by incorporating priors. Extensive experiments on the ImageNet and CIFAR-10 datasets show that our approach significantly outperforms 11 state-of-the-art methods in terms of query efficiency.
Open AccessBoosting Ray Search Procedure of Hard-label Attacks with Transfer-based Priors
One of the most practical and challenging types of black-box adversarial attacks is the hard-label attack, where only the top-1 predicted label is available. One effective approach is to search for the optimal ray direction from the benign image that minimizes the $\ell_p$-norm distance to the adversarial region. The unique advantage of this approach is that it transforms the hard-label attack into a continuous optimization problem. The objective function value is the ray's radius, which can be obtained via binary search at a high query cost. Existing methods use a "sign trick" in gradient estimation to reduce the number of queries. In this paper, we theoretically analyze the quality of this gradient estimation and propose a novel prior-guided approach to improve ray search efficiency both theoretically and empirically. Specifically, we utilize the transfer-based priors from surrogate models, and our gradient estimators appropriately integrate them by approximating the projection of the true gradient onto the subspace spanned by these priors and random directions, in a query-efficient manner. We theoretically derive the expected cosine similarities between the obtained gradient estimators and the true gradient, and demonstrate the improvement achieved by incorporating priors. Extensive experiments on the ImageNet and CIFAR-10 datasets show that our approach significantly outperforms 11 state-of-the-art methods in terms of query efficiency.
- 5 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (5) , 6430-6450
ADERLNet: Adaptive Denoising Enhancement Representation Learning for Low-Latency and High-Accurate Target Detection on SAR Sensors
Target detection on synthetic aperture radar (SAR) imagery is critical for several military and civilian applications. However, the coherent imaging mechanism of SAR will inevitably generate speckle noise, reducing the signal-to-noise ratio of SAR imagery, resulting in the loss of certain important information related to texture and thus degrading the accuracy of SAR image target detection. In addition, learning irrelevant features, such as natural clutter, increases the false alarm rate. To address these issues, this paper proposes a high-accurate and low-latency SAR target detection network (ADERLNet) to enhance the ability of the target detection model to represent valid features from strongly noisy images and irrelevant features. Specifically, the deep target detection architecture incorporates soft thresholding functions as nonlinear transformation layers, which can eliminate noisy and irrelevant features by forcing these interference features to be zeros. Second, given that manually setting the appropriate thresholds in the soft thresholding functions is challenging, the thresholds in ADERLNet are automatically determined to avoid substantial signal processing expertise by integrating specially designed sub-networks as trainable modules. Finally, the classification and regression losses are redesigned to offer the model stronger generalization ability by balancing the positive and negative samples and handling the impact of possible low-quality samples in the training set. Additionally, the structural re-parameterization method is employed to balance accuracy and speed. Experimental results on three widely used datasets for SAR target detection, including SSDD, MSAR-1.0, and SAR-Ship-Dataset, reveal that ADERLNet maximally improves average precision (AP) by 2.7%, 3.3%, and 1.8%, respectively, and far outperforms other methods on detection accuracy while maintaining excellent real-time performance, fully highlighting the superiority of ADERLNet. The trained models and source codes will be available at https://github.com/yangyahu-1994/ADERLNet.
- 5 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (5) , 6430-6450
ADERLNet: Adaptive Denoising Enhancement Representation Learning for Low-Latency and High-Accurate Target Detection on SAR Sensors
Target detection on synthetic aperture radar (SAR) imagery is critical for several military and civilian applications. However, the coherent imaging mechanism of SAR will inevitably generate speckle noise, reducing the signal-to-noise ratio of SAR imagery, resulting in the loss of certain important information related to texture and thus degrading the accuracy of SAR image target detection. In addition, learning irrelevant features, such as natural clutter, increases the false alarm rate. To address these issues, this paper proposes a high-accurate and low-latency SAR target detection network (ADERLNet) to enhance the ability of the target detection model to represent valid features from strongly noisy images and irrelevant features. Specifically, the deep target detection architecture incorporates soft thresholding functions as nonlinear transformation layers, which can eliminate noisy and irrelevant features by forcing these interference features to be zeros. Second, given that manually setting the appropriate thresholds in the soft thresholding functions is challenging, the thresholds in ADERLNet are automatically determined to avoid substantial signal processing expertise by integrating specially designed sub-networks as trainable modules. Finally, the classification and regression losses are redesigned to offer the model stronger generalization ability by balancing the positive and negative samples and handling the impact of possible low-quality samples in the training set. Additionally, the structural re-parameterization method is employed to balance accuracy and speed. Experimental results on three widely used datasets for SAR target detection, including SSDD, MSAR-1.0, and SAR-Ship-Dataset, reveal that ADERLNet maximally improves average precision (AP) by 2.7%, 3.3%, and 1.8%, respectively, and far outperforms other methods on detection accuracy while maintaining excellent real-time performance, fully highlighting the superiority of ADERLNet. The trained models and source codes will be available at https://github.com/yangyahu-1994/ADERLNet.
- Open Access
Finding Optimal Tangent Points for Reducing Distortions of Hard-label Attacks
One major problem in black-box adversarial attacks is the high query complexity in the hard-label attack setting, where only the top-1 predicted label is available. In this paper, we propose a novel geometric-based approach called Tangent Attack (TA), which identifies an optimal tangent point of a virtual hemisphere located on the decision boundary to reduce the distortion of the attack. Assuming the decision boundary is locally flat, we theoretically prove that the minimum $\ell_2$ distortion can be obtained by reaching the decision boundary along the tangent line passing through such tangent point in each iteration. To improve the robustness of our method, we further propose a generalized method which replaces the hemisphere with a semi-ellipsoid to adapt to curved decision boundaries. Our approach is free of pre-training. Extensive experiments conducted on the ImageNet and CIFAR-10 datasets demonstrate that our approach can consume only a small number of queries to achieve the low-magnitude distortion. The implementation source code is released online at https://github.com/machanic/TangentAttack.Open AccessFinding Optimal Tangent Points for Reducing Distortions of Hard-label Attacks
One major problem in black-box adversarial attacks is the high query complexity in the hard-label attack setting, where only the top-1 predicted label is available. In this paper, we propose a novel geometric-based approach called Tangent Attack (TA), which identifies an optimal tangent point of a virtual hemisphere located on the decision boundary to reduce the distortion of the attack. Assuming the decision boundary is locally flat, we theoretically prove that the minimum $\ell_2$ distortion can be obtained by reaching the decision boundary along the tangent line passing through such tangent point in each iteration. To improve the robustness of our method, we further propose a generalized method which replaces the hemisphere with a semi-ellipsoid to adapt to curved decision boundaries. Our approach is free of pre-training. Extensive experiments conducted on the ImageNet and CIFAR-10 datasets demonstrate that our approach can consume only a small number of queries to achieve the low-magnitude distortion. The implementation source code is released online at https://github.com/machanic/TangentAttack. - Open Access
- 5 October 2020
- Published by HARD Publishing Company in Polish Journal of Environmental Studies
- Vol. 30 (1) , 337-350
A Comparison of Su and Lu Modeling of Hydro-Thermal Coupling Model Using Field Temperature Records
The transport of materials and energy between rivers and riparian zones are driven by lateral hyporheic exchange and plays a vital role in regulating the ecological health of rivers. Owing to the limitations of models and methods, prior research has mainly focused on vertical hyporheic exchange in river systems. Combined with the collected riparian zone temperature and water level data, a 2D hydrothermal coupling model of the riparian zone was developed using the thermal conductivity model, and the fitting effects of the Lu et al. (2007) model and Su et al. (2016) model on temperature variations in riparian zone comparatively analyzed. Then, the dynamic variation characteristics of the riparian zone temperature and the lateral hyporheic exchange recharge pattern were investigated by the constructed models. The calibrated model could reasonably illustrate the variations in the riparian zone temperature, and the sensitivity analysis revealed that the hydraulic conductivity (Ks) is the most sensitive factor to the temperature variations.
- 5 October 2020
- Published by HARD Publishing Company in Polish Journal of Environmental Studies
- Vol. 30 (1) , 337-350
Open AccessA Comparison of Su and Lu Modeling of Hydro-Thermal Coupling Model Using Field Temperature Records
The transport of materials and energy between rivers and riparian zones are driven by lateral hyporheic exchange and plays a vital role in regulating the ecological health of rivers. Owing to the limitations of models and methods, prior research has mainly focused on vertical hyporheic exchange in river systems. Combined with the collected riparian zone temperature and water level data, a 2D hydrothermal coupling model of the riparian zone was developed using the thermal conductivity model, and the fitting effects of the Lu et al. (2007) model and Su et al. (2016) model on temperature variations in riparian zone comparatively analyzed. Then, the dynamic variation characteristics of the riparian zone temperature and the lateral hyporheic exchange recharge pattern were investigated by the constructed models. The calibrated model could reasonably illustrate the variations in the riparian zone temperature, and the sensitivity analysis revealed that the hydraulic conductivity (Ks) is the most sensitive factor to the temperature variations.
- Open Access
- 15 October 2019
- Published by Association for Computing Machinery (ACM)
- p. 692-701
MetaAdvDet
- 15 October 2019
- Published by Association for Computing Machinery (ACM)
- p. 692-701
Open AccessMetaAdvDet
- 1 August 2019
- Published by Elsevier in Neurocomputing
- Vol. 355, 35-47
AU R-CNN: Encoding expert prior knowledge into R-CNN for action unit detection
- 1 August 2019
- Published by Elsevier in Neurocomputing
- Vol. 355, 35-47
AU R-CNN: Encoding expert prior knowledge into R-CNN for action unit detection