Wenrui Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Wenrui Zhang has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Shenda Hong
National Institute of Health Data Science, Peking University, Beijing, China
5 shared publications - Shijia Geng
HeartVoice Medical Technology, Hefei, China
4 shared publications - Dong Chen
Huawei Technologies Co., Ltd., Beijing, China
3 shared publications - Guodong Wei
HeartVoice Medical Technology, Hefei, China
2 shared publications - Jie Wang
Huawei Technologies Co., Ltd, Beijing, China
2 shared publications - Xinxin Di
Department of Electrocardiology, The First Affiliated Hospital of University of Science and Technology of China, Hefei, China
2 shared publications
Latest Publications
- 9 June 2026
- Published by American Physical Society (APS) in Physical Review Letters
- Vol. 137 (4)
Program-synthesis-driven auto-design of universal unitary operators
- 9 June 2026
- Published by American Physical Society (APS) in Physical Review Letters
- Vol. 137 (4)
Program-synthesis-driven auto-design of universal unitary operators
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 94-105
EvoC2Rust: A Skeleton-guided Framework for Project-Level C-to-Rust Translation
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 94-105
Open AccessEvoC2Rust: A Skeleton-guided Framework for Project-Level C-to-Rust Translation
- Open Access
- 9 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 12-20
CG-Bench: Can Language Models Assist Call Graph Construction in the Real World?
- 9 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 12-20
Open AccessCG-Bench: Can Language Models Assist Call Graph Construction in the Real World?
- 9 December 2023
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (8) , 4047-4058
Cardiac arrhythmia classification with rejection of ECG recordings based on uncertainty estimation from deep neural networks
- 9 December 2023
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (8) , 4047-4058
Cardiac arrhythmia classification with rejection of ECG recordings based on uncertainty estimation from deep neural networks
- 21 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (11) , 16129-16138
Self-Supervised Time Series Representation Learning via Cross Reconstruction Transformer
Since labeled samples are typically scarce in real-world scenarios, self-supervised representation learning in time series is critical. Existing approaches mainly employ the contrastive learning framework, which automatically learns to understand similar and dissimilar data pairs. However, they are constrained by the request for cumbersome sampling policies and prior knowledge of constructing pairs. Also, few works have focused on effectively modeling temporal-spectral correlations to improve the capacity of representations. In this article, we propose the cross reconstruction transformer (CRT) to solve the aforementioned issues. CRT achieves time series representation learning through a cross-domain dropping-reconstruction task. Specifically, we obtain the frequency domain of the time series via the fast Fourier transform (FFT) and randomly drop certain patches in both time and frequency domains. Dropping is employed to maximally preserve the global context while masking leads to the distribution shift. Then a Transformer architecture is utilized to adequately discover the cross-domain correlations between temporal and spectral information through reconstructing data in both domains, which is called Dropped Temporal-Spectral Modeling. To discriminate the representations in global latent space, we propose instance discrimination constraint (IDC) to reduce the mutual information between different time series samples and sharpen the decision boundaries. Additionally, a specified curriculum learning (CL) strategy is employed to improve the robustness during the pretraining phase, which progressively increases the dropping ratio in the training process. We conduct extensive experiments to evaluate the effectiveness of the proposed method on multiple real-world datasets. Results show that CRT consistently achieves the best performance over existing methods by 2%–9%. The code is publicly available at https://github.com/BobZwr/Cross-Reconstruction-Transformer .
- 21 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (11) , 16129-16138
Self-Supervised Time Series Representation Learning via Cross Reconstruction Transformer
Since labeled samples are typically scarce in real-world scenarios, self-supervised representation learning in time series is critical. Existing approaches mainly employ the contrastive learning framework, which automatically learns to understand similar and dissimilar data pairs. However, they are constrained by the request for cumbersome sampling policies and prior knowledge of constructing pairs. Also, few works have focused on effectively modeling temporal-spectral correlations to improve the capacity of representations. In this article, we propose the cross reconstruction transformer (CRT) to solve the aforementioned issues. CRT achieves time series representation learning through a cross-domain dropping-reconstruction task. Specifically, we obtain the frequency domain of the time series via the fast Fourier transform (FFT) and randomly drop certain patches in both time and frequency domains. Dropping is employed to maximally preserve the global context while masking leads to the distribution shift. Then a Transformer architecture is utilized to adequately discover the cross-domain correlations between temporal and spectral information through reconstructing data in both domains, which is called Dropped Temporal-Spectral Modeling. To discriminate the representations in global latent space, we propose instance discrimination constraint (IDC) to reduce the mutual information between different time series samples and sharpen the decision boundaries. Additionally, a specified curriculum learning (CL) strategy is employed to improve the robustness during the pretraining phase, which progressively increases the dropping ratio in the training process. We conduct extensive experiments to evaluate the effectiveness of the proposed method on multiple real-world datasets. Results show that CRT consistently achieves the best performance over existing methods by 2%–9%. The code is publicly available at https://github.com/BobZwr/Cross-Reconstruction-Transformer .
- Open Access
Self-Supervised Time Series Representation Learning via Cross Reconstruction Transformer
Unsupervised/self-supervised representation learning in time series is critical since labeled samples are usually scarce in real-world scenarios. Existing approaches mainly leverage the contrastive learning framework, which automatically learns to understand the similar and dissimilar data pairs. Nevertheless, they are restricted to the prior knowledge of constructing pairs, cumbersome sampling policy, and unstable performances when encountering sampling bias. Also, few works have focused on effectively modeling across temporal-spectral relations to extend the capacity of representations. In this paper, we aim at learning representations for time series from a new perspective and propose Cross Reconstruction Transformer (CRT) to solve the aforementioned problems in a unified way. CRT achieves time series representation learning through a cross-domain dropping-reconstruction task. Specifically, we transform time series into the frequency domain and randomly drop certain parts in both time and frequency domains. Dropping can maximally preserve the global context compared to cropping and masking. Then a transformer architecture is utilized to adequately capture the cross-domain correlations between temporal and spectral information through reconstructing data in both domains, which is called Dropped Temporal-Spectral Modeling. To discriminate the representations in global latent space, we propose Instance Discrimination Constraint to reduce the mutual information between different time series and sharpen the decision boundaries. Additionally, we propose a specified curriculum learning strategy to optimize the CRT, which progressively increases the dropping ratio in the training process.
Open AccessSelf-Supervised Time Series Representation Learning via Cross Reconstruction Transformer
Unsupervised/self-supervised representation learning in time series is critical since labeled samples are usually scarce in real-world scenarios. Existing approaches mainly leverage the contrastive learning framework, which automatically learns to understand the similar and dissimilar data pairs. Nevertheless, they are restricted to the prior knowledge of constructing pairs, cumbersome sampling policy, and unstable performances when encountering sampling bias. Also, few works have focused on effectively modeling across temporal-spectral relations to extend the capacity of representations. In this paper, we aim at learning representations for time series from a new perspective and propose Cross Reconstruction Transformer (CRT) to solve the aforementioned problems in a unified way. CRT achieves time series representation learning through a cross-domain dropping-reconstruction task. Specifically, we transform time series into the frequency domain and randomly drop certain parts in both time and frequency domains. Dropping can maximally preserve the global context compared to cropping and masking. Then a transformer architecture is utilized to adequately capture the cross-domain correlations between temporal and spectral information through reconstructing data in both domains, which is called Dropped Temporal-Spectral Modeling. To discriminate the representations in global latent space, we propose Instance Discrimination Constraint to reduce the mutual information between different time series and sharpen the decision boundaries. Additionally, we propose a specified curriculum learning strategy to optimize the CRT, which progressively increases the dropping ratio in the training process.
- Open Access
- 16 November 2022
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 22 (1) , 1-9
Estimating critical values from electrocardiogram using a deep ordinal convolutional neural network
Background: Critical values are commonly used in clinical laboratory tests to define health-related conditions of varying degrees. Knowing the values, people can quickly become aware of health risks, and the health professionals can take immediate actions and save lives. Methods: In this paper, we propose a method that extends the concept of critical value to one of the most commonly used physiological signals in the clinical environment—Electrocardiogram (ECG). We first construct a mapping from common ECG diagnostic conclusions to critical values. After that, we build a 61-layer deep convolutional neural network named CardioV, which is characterized by an ordinal classifier. Results: We conduct experiments on a large public ECG dataset, and demonstrate that CardioV achieves a mean absolute error of 0.4984 and a ROC-AUC score of 0.8735. In addition, we find that the model performs better for extreme critical values and the younger age group, while gender does not affect the performance. The ablation study confirms that the ordinal classification mechanism suits for estimating the critical values which contain ranking information. Moreover, model interpretation techniques help us discover that CardioV focuses on the characteristic ECG locations during the critical value estimation process. Conclusions: As an ordinal classifier, CardioV performs well in estimating ECG critical values that can help people quickly identify different heart conditions. We obtain ROC-AUC scores above 0.8 for all four critical value categories, and find that the extreme values (0 (no risk) and 3 (high risk)) have better model performance than the other two (1 (low risk) and 2 (medium risk)). Results also show that gender does not affect the performance, and the older age group has worse performance than the younger age group. In addition, visualization techniques reveal that the model pays more attention to characteristic ECG locations.
- 16 November 2022
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 22 (1) , 1-9
Open AccessEstimating critical values from electrocardiogram using a deep ordinal convolutional neural network
Background: Critical values are commonly used in clinical laboratory tests to define health-related conditions of varying degrees. Knowing the values, people can quickly become aware of health risks, and the health professionals can take immediate actions and save lives. Methods: In this paper, we propose a method that extends the concept of critical value to one of the most commonly used physiological signals in the clinical environment—Electrocardiogram (ECG). We first construct a mapping from common ECG diagnostic conclusions to critical values. After that, we build a 61-layer deep convolutional neural network named CardioV, which is characterized by an ordinal classifier. Results: We conduct experiments on a large public ECG dataset, and demonstrate that CardioV achieves a mean absolute error of 0.4984 and a ROC-AUC score of 0.8735. In addition, we find that the model performs better for extreme critical values and the younger age group, while gender does not affect the performance. The ablation study confirms that the ordinal classification mechanism suits for estimating the critical values which contain ranking information. Moreover, model interpretation techniques help us discover that CardioV focuses on the characteristic ECG locations during the critical value estimation process. Conclusions: As an ordinal classifier, CardioV performs well in estimating ECG critical values that can help people quickly identify different heart conditions. We obtain ROC-AUC scores above 0.8 for all four critical value categories, and find that the extreme values (0 (no risk) and 3 (high risk)) have better model performance than the other two (1 (low risk) and 2 (medium risk)). Results also show that gender does not affect the performance, and the older age group has worse performance than the younger age group. In addition, visualization techniques reveal that the model pays more attention to characteristic ECG locations.
- 23 September 2022
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 79
A simple self-supervised ECG representation learning method via manipulated temporal–spatial reverse detection
- 23 September 2022
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 79
A simple self-supervised ECG representation learning method via manipulated temporal–spatial reverse detection