Wenyu Qu
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
Wenyu Qu has a total of 161 co-authors. Here are the top co-authors by number of shared publications:
- Xiulong Liu
Tianjin University, Tianjin, China
21 shared publications - Tie Qiu
Northeastern University, Shenyang, China
15 shared publications - Xinyu Tong
Tianjin University, Tianjin, China
13 shared publications - Laiping Zhao
Tianjin University
12 shared publications - Xiaobo Zhou
Tianjin University
10 shared publications - Xin Xie
Tianjin University
10 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25  (11) , 21953-21969
MRPO: An Evolutionary Game-Based Multiple Round Probabilistic Oracle for BIoV
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25  (11) , 21953-21969
MRPO: An Evolutionary Game-Based Multiple Round Probabilistic Oracle for BIoV
- Open Access
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10  (3) , 1-33
TSGen: A Framework for Indoor Trajectory-Semantic Data Synthesis
Smart homes are evolving from passive response to proactive service, where understanding the semantic information in user trajectories is essential. However, existing tracking systems output only sparse location sequences, insufficient for higher-level semantic reasoning. The core bottleneck lies in the lack of large-scale annotated datasets, while traditional data collection is costly and raises privacy concerns. We present TSGen, which formalizes indoor trajectory-semantic data synthesis as a structured decomposition problem. We first demonstrate that naive end-to-end LLM approaches are fundamentally limited for this task, then propose a hybrid workflow combining LLM-driven behavior simulation, physics-constrained trajectory synthesis, and a self-refining prompt architecture that evolves from runtime errors, to construct the T-S dataset (14,000 hours of trajectory-semantic pairs). We also implement a hierarchical processing pipeline as an initial baseline for semantic understanding, establishing reference benchmarks for future research. Evaluations cover data quality (user studies, diversity analysis, ablation studies) and baseline performance. On QA tasks over synthetic data and real-world deployments, our baseline significantly outperforms direct LLM methods while reducing computational costs. The open-sourced dataset and codebase provide foundational resources for this emerging research area 1 .
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10  (3) , 1-33
Open AccessTSGen: A Framework for Indoor Trajectory-Semantic Data Synthesis
Smart homes are evolving from passive response to proactive service, where understanding the semantic information in user trajectories is essential. However, existing tracking systems output only sparse location sequences, insufficient for higher-level semantic reasoning. The core bottleneck lies in the lack of large-scale annotated datasets, while traditional data collection is costly and raises privacy concerns. We present TSGen, which formalizes indoor trajectory-semantic data synthesis as a structured decomposition problem. We first demonstrate that naive end-to-end LLM approaches are fundamentally limited for this task, then propose a hybrid workflow combining LLM-driven behavior simulation, physics-constrained trajectory synthesis, and a self-refining prompt architecture that evolves from runtime errors, to construct the T-S dataset (14,000 hours of trajectory-semantic pairs). We also implement a hierarchical processing pipeline as an initial baseline for semantic understanding, establishing reference benchmarks for future research. Evaluations cover data quality (user studies, diversity analysis, ablation studies) and baseline performance. On QA tasks over synthetic data and real-world deployments, our baseline significantly outperforms direct LLM methods while reducing computational costs. The open-sourced dataset and codebase provide foundational resources for this emerging research area 1 .
- Open Access
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10  (3) , 1-29
UbiTrack: A Layout-Robust Neural System for Device-Free WiFi Tracking
Neural network-based device-free Wi-Fi tracking achieves superior accuracy over theory-driven approaches, yet its practical utility is severely limited by poor generalization across different device layouts. Existing models implicitly learn site-specific correlations, necessitating labor-intensive data recollection and retraining whenever the device configuration changes. To overcome this dependency, we propose UbiTrack , a generalized framework capable of robust tracking across diverse device layouts. Our core innovation lies in explicitly modeling the geometric constraints between device positions and signal propagation within the learning process. Instead of treating the layout as a black box, we formulate the tracking problem as a layout-aware mapping. We design a novel neural network architecture that incorporates geometric graph constraints into the analysis of temporal signal features, enabling the model to learn the intrinsic physical interactions between device topology and user trajectories. To facilitate the training of this data-hungry framework, we employ a progressive training strategy leveraging synthetic data derived from theory-driven principles. Extensive real-world experiments using commercial off-the-shelf (COTS) Wi-Fi devices validate that UbiTrack effectively decouples tracking performance from environmental setups. The system achieves median tracking errors of 0.56m in fixed layouts and 0.39m in variable layouts, paving the way for scalable and robust wireless sensing.
- 30 September 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10  (3) , 1-29
Open AccessUbiTrack: A Layout-Robust Neural System for Device-Free WiFi Tracking
Neural network-based device-free Wi-Fi tracking achieves superior accuracy over theory-driven approaches, yet its practical utility is severely limited by poor generalization across different device layouts. Existing models implicitly learn site-specific correlations, necessitating labor-intensive data recollection and retraining whenever the device configuration changes. To overcome this dependency, we propose UbiTrack , a generalized framework capable of robust tracking across diverse device layouts. Our core innovation lies in explicitly modeling the geometric constraints between device positions and signal propagation within the learning process. Instead of treating the layout as a black box, we formulate the tracking problem as a layout-aware mapping. We design a novel neural network architecture that incorporates geometric graph constraints into the analysis of temporal signal features, enabling the model to learn the intrinsic physical interactions between device topology and user trajectories. To facilitate the training of this data-hungry framework, we employ a progressive training strategy leveraging synthetic data derived from theory-driven principles. Extensive real-world experiments using commercial off-the-shelf (COTS) Wi-Fi devices validate that UbiTrack effectively decouples tracking performance from environmental setups. The system achieves median tracking errors of 0.56m in fixed layouts and 0.39m in variable layouts, paving the way for scalable and robust wireless sensing.
- 20 July 2026
- Published by Springer Nature
- p. 379-397
DiVeR: Differential and Verifiable Data Dissemination for Multi-path Payment Routing
- 20 July 2026
- Published by Springer Nature
- p. 379-397
DiVeR: Differential and Verifiable Data Dissemination for Multi-path Payment Routing
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 75  (7) , 2305-2321
RollShard: Atomic Multi-Shard Transactions via Verifiable Stateless Off-Chain Processing
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 75  (7) , 2305-2321
RollShard: Atomic Multi-Shard Transactions via Verifiable Stateless Off-Chain Processing
- Open Access
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 144-159
iRoute: Local Routing Table-based Workflow Management in Serverless Computing
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 144-159
Open AccessiRoute: Local Routing Table-based Workflow Management in Serverless Computing
- Open Access
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 423-438
PARD: Enhancing Goodput for Inference Pipeline via Proactive Request Dropping
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 423-438
Open AccessPARD: Enhancing Goodput for Inference Pipeline via Proactive Request Dropping
- Open Access
PARD: Enhancing Goodput for Inference Pipeline via Proactive Request Dropping
Modern deep neural network (DNN) applications integrate multiple DNN models into inference pipelines with stringent latency requirements for customized tasks. To mitigate extensive request timeouts caused by accumulation, systems for inference pipelines commonly drop a subset of requests so the remaining ones can satisfy latency constraints. Since it is commonly believed that request dropping adversely affects goodput, existing systems only drop requests when they have to, which we call reactive dropping. However, this reactive policy can not maintain high goodput, as it neither makes timely dropping decisions nor identifies the proper set of requests to drop, leading to issues of dropping requests too late or dropping the wrong set of requests. We propose that the inference system should proactively drop certain requests in advance to enhance the goodput across the entire workload. To achieve this, we design an inference system PARD. It enhances goodput with timely and precise dropping decisions by integrating a proactive dropping method that decides when to drop requests using runtime information of the inference pipeline, and an adaptive request priority mechanism that selects which specific requests to drop based on remaining latency budgets and workload intensity. Evaluation on a cluster of 64 GPUs over real-world workloads shows that PARD achieves $16\%$-$176\%$ higher goodput than the state of the art while reducing the drop rate and wasted computation resources by $1.6\times$-$17\times$ and $1.5\times$-$62\times$ respectively.Open AccessPARD: Enhancing Goodput for Inference Pipeline via Proactive Request Dropping
Modern deep neural network (DNN) applications integrate multiple DNN models into inference pipelines with stringent latency requirements for customized tasks. To mitigate extensive request timeouts caused by accumulation, systems for inference pipelines commonly drop a subset of requests so the remaining ones can satisfy latency constraints. Since it is commonly believed that request dropping adversely affects goodput, existing systems only drop requests when they have to, which we call reactive dropping. However, this reactive policy can not maintain high goodput, as it neither makes timely dropping decisions nor identifies the proper set of requests to drop, leading to issues of dropping requests too late or dropping the wrong set of requests. We propose that the inference system should proactively drop certain requests in advance to enhance the goodput across the entire workload. To achieve this, we design an inference system PARD. It enhances goodput with timely and precise dropping decisions by integrating a proactive dropping method that decides when to drop requests using runtime information of the inference pipeline, and an adaptive request priority mechanism that selects which specific requests to drop based on remaining latency budgets and workload intensity. Evaluation on a cluster of 64 GPUs over real-world workloads shows that PARD achieves $16\%$-$176\%$ higher goodput than the state of the art while reducing the drop rate and wasted computation resources by $1.6\times$-$17\times$ and $1.5\times$-$62\times$ respectively. - 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 10616-10635
TR-RNP: An Adaptive Transmission-Reliable Relay Node Placement Strategy in Internet of Underwater Things
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 10616-10635
TR-RNP: An Adaptive Transmission-Reliable Relay Node Placement Strategy in Internet of Underwater Things
- 30 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25  (6) , 8413-8431
CATS: Toward Accurate Device-Free Tracking by Quantifying the Sensing Confidence
Wireless device-free tracking has become a cornerstone of human-centric sensing services owing to its ubiquity and convenience. With the development of smart IoT devices such as smart speakers and voice assistants, integrating Wi-Fi with acoustic signals has opened up new opportunities for multi-modal tracking. Nevertheless, achieving accurate tracking in real-world environments remains challenging, primarily due to two factors: (1) Wireless signals are highly susceptible to interference in complex indoor scenarios, and (2) the uneven distribution of sensing regions imposes fundamental theoretical constraints on system performance. To address these challenges, this paper introduces CATS, a Confidence-based Augmented Tracking System. The core concept of CATS lies in quantifying the confidence of the features and improving the tracking performance with it. Specifically, we design signal-level and position-level confidence to assess the impact of noise interference during deployment and the inherent theoretical limitations. Furthermore, we develop a knowledge-data joint-driven framework that efficiently integrates confidence into a dual-grained tracking network. Experimental evaluation on commercial devices demonstrates that CATS achieves high-accuracy tracking with a median error of$ 0.47\; {\rm m}$ in various complex scenarios. Compared to existing systems, it reduces the median error by 35.58% and the tail error by 49.68% on average.- 30 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25  (6) , 8413-8431
CATS: Toward Accurate Device-Free Tracking by Quantifying the Sensing Confidence
Wireless device-free tracking has become a cornerstone of human-centric sensing services owing to its ubiquity and convenience. With the development of smart IoT devices such as smart speakers and voice assistants, integrating Wi-Fi with acoustic signals has opened up new opportunities for multi-modal tracking. Nevertheless, achieving accurate tracking in real-world environments remains challenging, primarily due to two factors: (1) Wireless signals are highly susceptible to interference in complex indoor scenarios, and (2) the uneven distribution of sensing regions imposes fundamental theoretical constraints on system performance. To address these challenges, this paper introduces CATS, a Confidence-based Augmented Tracking System. The core concept of CATS lies in quantifying the confidence of the features and improving the tracking performance with it. Specifically, we design signal-level and position-level confidence to assess the impact of noise interference during deployment and the inherent theoretical limitations. Furthermore, we develop a knowledge-data joint-driven framework that efficiently integrates confidence into a dual-grained tracking network. Experimental evaluation on commercial devices demonstrates that CATS achieves high-accuracy tracking with a median error of$ 0.47\; {\rm m}$ in various complex scenarios. Compared to existing systems, it reduces the median error by 35.58% and the tail error by 49.68% on average.