Wenbo Zhang
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
Wenbo Zhang has a total of 67 co-authors. Here are the top co-authors by number of shared publications:
- Heng Wu
Institute of Software, Chinese Academy of Sciences, Beijing, China
7 shared publications Hua Zhong
State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences, Beijing, China
5 shared publicationsJun Wei
State Key Laboratory of Computer Science, Beijing, 100190, China
5 shared publicationsTao Huang
Chinese Academy of Sciences
5 shared publications- Yuewen Wu
Institute of Software, Chinese Academy of Sciences, Beijing, China
3 shared publications - Beihong Jin
Institute of Software, Chinese Academy of Sciences, Beijing, China
2 shared publications
Latest Publications
- 1 November 2025
- Published by Springer Nature in Journal of Computer Science and Technology
- Vol. 40 (6) , 1608-1625
vTensor-Based GPU Memory Management for Edge Deep Learning Training
- 1 November 2025
- Published by Springer Nature in Journal of Computer Science and Technology
- Vol. 40 (6) , 1608-1625
vTensor-Based GPU Memory Management for Edge Deep Learning Training
- Open Access
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 541-552
Scheduling based on Block Features for Concurrent Inference with Unseen DNN Models on GPU
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 541-552
Open AccessScheduling based on Block Features for Concurrent Inference with Unseen DNN Models on GPU
- Open Access
- 3 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 56-68
ETS: Deep Learning Training Iteration Time Prediction based on Execution Trace Sliding Window
Deep learning (DL) has become essential across various computer science domains. Accurately predicting iteration time for DL models in diverse cloud data center environments is critical for making high-quality scheduling decisions. Existing approaches neglect the sequential features inherent in the runtime execution, leading to issues such as overlooking DL framework overhead and struggling to handle diverse sizes of DL models, resulting in either low accuracy or slow convergence of the prediction model. This paper introduces ETS, a novel iteration time prediction method utilizing execution trace sliding windows. Our observation reveals that DL models exhibit a highly sequential runtime execution nature. Building upon this insight, we leverage sliding windows to extract a novel type of sequential features from the runtime execution trace. These features comprehensively capture DL framework overhead and address the diversity challenge in DL model sizes. By combining a best-practice method to train a prediction model, we achieve high accuracy and rapid convergence simultaneously. Experimental validation on over 14,000 DL model configurations demonstrates ETS's effectiveness in predicting the iteration time of DL models, achieving a mere 5.9% prediction error with a training time at the 10-minute level, and improving scheduling outcomes by reducing job completion time by 17%.
- 3 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 56-68
Open AccessETS: Deep Learning Training Iteration Time Prediction based on Execution Trace Sliding Window
Deep learning (DL) has become essential across various computer science domains. Accurately predicting iteration time for DL models in diverse cloud data center environments is critical for making high-quality scheduling decisions. Existing approaches neglect the sequential features inherent in the runtime execution, leading to issues such as overlooking DL framework overhead and struggling to handle diverse sizes of DL models, resulting in either low accuracy or slow convergence of the prediction model. This paper introduces ETS, a novel iteration time prediction method utilizing execution trace sliding windows. Our observation reveals that DL models exhibit a highly sequential runtime execution nature. Building upon this insight, we leverage sliding windows to extract a novel type of sequential features from the runtime execution trace. These features comprehensively capture DL framework overhead and address the diversity challenge in DL model sizes. By combining a best-practice method to train a prediction model, we achieve high accuracy and rapid convergence simultaneously. Experimental validation on over 14,000 DL model configurations demonstrates ETS's effectiveness in predicting the iteration time of DL models, achieving a mere 5.9% prediction error with a training time at the 10-minute level, and improving scheduling outcomes by reducing job completion time by 17%.
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 17 (2) , 482-496
CSTL: Compositional Signal Temporal Logic for Adaptive Edge Service Monitoring
Edge service monitoring is essential to guarantee the healthy of service compositions at runtime. Current techniques focus mostly on the monitoring of atomic edge services, but they are inadequate for that of inter- and composite services. Besides, constraints to be monitored are usually pre-specified, although certain parameters may have to be adapted online according to the execution context. To address these challenges, this paper formulates the problem of edge service monitoring as the interpretation of temporal constraints and time-dependent QoS constraints upon intra-, inter-, and composite services. Leveraging our proposed Compositional Signal Temporal Logic (CSTL) with extended compositional modalities and online parameter settings, an adaptive monitoring mechanism is developed, where constraints are converted to CSTL formulae, and QoS variations and temporal violations are interpreted qualitatively and quantitatively at runtime. Extensive experiments are conducted upon publicly-available datasets, and evaluation results show that CSTL performs better than baseline techniques in terms of expressiveness, applicability, and robustness.
- 17 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 17 (2) , 482-496
CSTL: Compositional Signal Temporal Logic for Adaptive Edge Service Monitoring
Edge service monitoring is essential to guarantee the healthy of service compositions at runtime. Current techniques focus mostly on the monitoring of atomic edge services, but they are inadequate for that of inter- and composite services. Besides, constraints to be monitored are usually pre-specified, although certain parameters may have to be adapted online according to the execution context. To address these challenges, this paper formulates the problem of edge service monitoring as the interpretation of temporal constraints and time-dependent QoS constraints upon intra-, inter-, and composite services. Leveraging our proposed Compositional Signal Temporal Logic (CSTL) with extended compositional modalities and online parameter settings, an adaptive monitoring mechanism is developed, where constraints are converted to CSTL formulae, and QoS variations and temporal violations are interpreted qualitatively and quantitatively at runtime. Extensive experiments are conducted upon publicly-available datasets, and evaluation results show that CSTL performs better than baseline techniques in terms of expressiveness, applicability, and robustness.
- 20 November 2023
- Published by Springer Nature
- p. 275-290
2DPChain: Orchestrating Transactions in Order-Execute Blockchain to Exploit Intra-batch and Inter-batch Parallelism
- 20 November 2023
- Published by Springer Nature
- p. 275-290
2DPChain: Orchestrating Transactions in Order-Execute Blockchain to Exploit Intra-batch and Inter-batch Parallelism
- Open Access
- 7 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 605-614
SPLIT: QoS-Aware DNN Inference on Shared GPU via Evenly-Sized Model Splitting
- 7 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 605-614
Open AccessSPLIT: QoS-Aware DNN Inference on Shared GPU via Evenly-Sized Model Splitting
- 14 April 2023
- Published by Springer Nature
- p. 483-498
InstantChain: Enhancing Order-Execute Blockchain Systems for Latency-Sensitive Applications
- 14 April 2023
- Published by Springer Nature
- p. 483-498
InstantChain: Enhancing Order-Execute Blockchain Systems for Latency-Sensitive Applications
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 72 (8) , 2224-2236
Hydra: Deadline-Aware and Efficiency-Oriented Scheduling for Deep Learning Jobs on Heterogeneous GPUs
With the rapid proliferation of deep learning (DL) jobs running on heterogeneous GPUs, scheduling DL jobs to meet various scheduling requirements, such as meeting deadlines and reducing job completion time (JCT), is critical. Unfortunately, existing efficiency-oriented and deadline-aware efforts are still rudimentary. They lack the capability of scheduling jobs to meet deadline requirements while reducing total JCT, especially when the jobs have various execution times on heterogeneous GPUs. Therefore, we present Hydra, a novel quantitative cost comparison approach, to address this scheduling issue. Here, the cost represents the total JCT plus a dynamic penalty calculated from the total tardiness (i.e., the delay time of exceeding the deadline) of all jobs. Hydra adopts a sampling approach that exploits the inherent iterative periodicity of DL jobs to estimate job execution times accurately on heterogeneous GPUs. Then, Hydra considers various combinations of job sequences and GPUs to obtain the minimized cost by leveraging an efficient branch-and-bound algorithm. Finally, the results of evaluation experiments on Alibaba traces show that Hydra can reduce total tardiness by 85.8% while reducing total JCT as much as possible, compared with state-of-the-art efforts.
- 3 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 72 (8) , 2224-2236
Hydra: Deadline-Aware and Efficiency-Oriented Scheduling for Deep Learning Jobs on Heterogeneous GPUs
With the rapid proliferation of deep learning (DL) jobs running on heterogeneous GPUs, scheduling DL jobs to meet various scheduling requirements, such as meeting deadlines and reducing job completion time (JCT), is critical. Unfortunately, existing efficiency-oriented and deadline-aware efforts are still rudimentary. They lack the capability of scheduling jobs to meet deadline requirements while reducing total JCT, especially when the jobs have various execution times on heterogeneous GPUs. Therefore, we present Hydra, a novel quantitative cost comparison approach, to address this scheduling issue. Here, the cost represents the total JCT plus a dynamic penalty calculated from the total tardiness (i.e., the delay time of exceeding the deadline) of all jobs. Hydra adopts a sampling approach that exploits the inherent iterative periodicity of DL jobs to estimate job execution times accurately on heterogeneous GPUs. Then, Hydra considers various combinations of job sequences and GPUs to obtain the minimized cost by leveraging an efficient branch-and-bound algorithm. Finally, the results of evaluation experiments on Alibaba traces show that Hydra can reduce total tardiness by 85.8% while reducing total JCT as much as possible, compared with state-of-the-art efforts.
- Open Access
- 21 December 2022
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 6 (4) , 1-27
Embracing Consumer-level UWB-equipped Devices for Fine-grained Wireless Sensing
RF sensing has been actively exploited in the past few years to enable novel IoT applications. Among different wireless technologies, WiFi-based sensing is most popular owing to the pervasiveness of WiFi infrastructure. However, one critical issue associated with WiFi sensing is that the information required for sensing can not be obtained from consumer-level devices such as smartphones or smart watches. The commonly-seen WiFi devices in our everyday lives actually can not be utilized for sensing. Instead, dedicated hardware with a specific WiFi card (e.g., Intel 5300) needs to be used for WiFi sensing. This paper involves Ultra-Wideband (UWB) into the ecosystem of RF sensing and makes RF sensing work on consumer-level hardware such as smartphones and smart watches for the first time. We propose a series of methods to realize UWB sensing on consumer-level electronics without any hardware modification. By leveraging fine-grained human respiration monitoring as the application example, we demonstrate that the achieved performance on consumer-level electronics is comparable to that achieved using dedicated UWB hardware. We show that UWB sensing hosted on consumer-level electronics is able to achieve fine granularity, robustness against interference and also multi-target sensing, pushing RF sensing one step towards real-life adoption.
- 21 December 2022
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 6 (4) , 1-27
Open AccessEmbracing Consumer-level UWB-equipped Devices for Fine-grained Wireless Sensing
RF sensing has been actively exploited in the past few years to enable novel IoT applications. Among different wireless technologies, WiFi-based sensing is most popular owing to the pervasiveness of WiFi infrastructure. However, one critical issue associated with WiFi sensing is that the information required for sensing can not be obtained from consumer-level devices such as smartphones or smart watches. The commonly-seen WiFi devices in our everyday lives actually can not be utilized for sensing. Instead, dedicated hardware with a specific WiFi card (e.g., Intel 5300) needs to be used for WiFi sensing. This paper involves Ultra-Wideband (UWB) into the ecosystem of RF sensing and makes RF sensing work on consumer-level hardware such as smartphones and smart watches for the first time. We propose a series of methods to realize UWB sensing on consumer-level electronics without any hardware modification. By leveraging fine-grained human respiration monitoring as the application example, we demonstrate that the achieved performance on consumer-level electronics is comparable to that achieved using dedicated UWB hardware. We show that UWB sensing hosted on consumer-level electronics is able to achieve fine granularity, robustness against interference and also multi-target sensing, pushing RF sensing one step towards real-life adoption.
- Open Access
- 21 December 2022
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 6 (4) , 1-25
ECG-grained Cardiac Monitoring Using UWB Signals
With the development of wireless sensing, researchers have proposed many contactless vital sign monitoring systems, which can be used to monitor respiration rates, heart rates, cardiac cycles and etc. However, these vital signs are ones of coarse granularity, so they are less helpful in the diagnosis of cardiovascular diseases (CVDs). Considering that electrocardiogram (ECG) is an important evidence base for the diagnoses of CVDs, we propose to generate ECGs from ultra-wideband (UWB) signals in a contactless manner as a fine-grained cardiac monitoring solution. Specifically, we analyze the properties of UWB signals containing heartbeats and respiration, and design two complementary heartbeat signal restoration methods to perfectly recover heartbeat signal variation. To establish the mapping between the mechanical activity of the heart sensed by UWB devices and the electrical activity of the heart recorded in ECGs, we construct a conditional generative adversarial network to encode the mapping between mechanical activity and electrical activity and propose a contrastive learning strategy to reduce the interference from noise in UWB signals. We build the corresponding cardiac monitoring system named RF-ECG and conduct extensive experiments using about 120,000 heartbeats from more than 40 participants. The experimental results show that the ECGs generated by RF-ECG have good performance in both ECG intervals and morphology compared with the ground truth. Moreover, diseases such as tachycardia/bradycardia, sinus arrhythmia, and premature contractions can be diagnosed from the ECGs generated by our RF-ECG.
- 21 December 2022
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 6 (4) , 1-25
Open AccessECG-grained Cardiac Monitoring Using UWB Signals
With the development of wireless sensing, researchers have proposed many contactless vital sign monitoring systems, which can be used to monitor respiration rates, heart rates, cardiac cycles and etc. However, these vital signs are ones of coarse granularity, so they are less helpful in the diagnosis of cardiovascular diseases (CVDs). Considering that electrocardiogram (ECG) is an important evidence base for the diagnoses of CVDs, we propose to generate ECGs from ultra-wideband (UWB) signals in a contactless manner as a fine-grained cardiac monitoring solution. Specifically, we analyze the properties of UWB signals containing heartbeats and respiration, and design two complementary heartbeat signal restoration methods to perfectly recover heartbeat signal variation. To establish the mapping between the mechanical activity of the heart sensed by UWB devices and the electrical activity of the heart recorded in ECGs, we construct a conditional generative adversarial network to encode the mapping between mechanical activity and electrical activity and propose a contrastive learning strategy to reduce the interference from noise in UWB signals. We build the corresponding cardiac monitoring system named RF-ECG and conduct extensive experiments using about 120,000 heartbeats from more than 40 participants. The experimental results show that the ECGs generated by RF-ECG have good performance in both ECG intervals and morphology compared with the ground truth. Moreover, diseases such as tachycardia/bradycardia, sinus arrhythmia, and premature contractions can be diagnosed from the ECGs generated by our RF-ECG.