Guanzhou Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Guanzhou Zhu has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Dong Zhao
Beijing University of Posts and Telecommunications
7 shared publications - Huadong Ma
Beijing University of Posts and Telecommunications
7 shared publications - Zhengyuan Zhang
School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China
5 shared publications - Zilong Wang
School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China
3 shared publications - Chunliang Li
State Key Laboratory of Networking and Switching Technology, School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications - Yixuan Song
State Key Laboratory of Network and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications
Latest Publications
- Open Access
- 15 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (2) , 1-35
PointRing: Enabling Calibration-Free Pointing Interaction in Smart Homes with a UWB-IMU Smart Ring
Pointing provides a direct and intuitive mechanism for selecting and controlling devices in multi-device smart homes. However, existing wrist-worn approaches often suffer from high deployment overhead and degraded reliability due to orientation drift and repeated recalibration. We present PointRing , a smart ring that fuses Ultra-Wideband (UWB) and Inertial Measurement Unit (IMU) sensing to enable calibration-free, drift-resistant pointing-based device selection. PointRing introduces two core components: TRA (Trajectory-based Ray Alignment), a trajectory-driven fusion method that infers and refines pointing rays from UWB localization and IMU motion traces without relying on absolute attitude calibration, and SnapMap , a lightweight deployment tool that uses smartphone-captured RGB-D images to register three UWB anchors and device locations for rapid deployment and redeployment. Across a controlled laboratory and a real home environment with six devices, PointRing achieves an overall device selection accuracy of 94.86%. Full deployment takes 117-165 s, while single-device redeployment takes 15-19 s. A user study with 43 participants further characterizes the deployment workflow and interaction tradeoffs, and reveals that PointRing effectively complements commercial smart speakers in typical home scenarios—particularly for spatial disambiguation and fine-grained control.
- 15 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (2) , 1-35
Open AccessPointRing: Enabling Calibration-Free Pointing Interaction in Smart Homes with a UWB-IMU Smart Ring
Pointing provides a direct and intuitive mechanism for selecting and controlling devices in multi-device smart homes. However, existing wrist-worn approaches often suffer from high deployment overhead and degraded reliability due to orientation drift and repeated recalibration. We present PointRing , a smart ring that fuses Ultra-Wideband (UWB) and Inertial Measurement Unit (IMU) sensing to enable calibration-free, drift-resistant pointing-based device selection. PointRing introduces two core components: TRA (Trajectory-based Ray Alignment), a trajectory-driven fusion method that infers and refines pointing rays from UWB localization and IMU motion traces without relying on absolute attitude calibration, and SnapMap , a lightweight deployment tool that uses smartphone-captured RGB-D images to register three UWB anchors and device locations for rapid deployment and redeployment. Across a controlled laboratory and a real home environment with six devices, PointRing achieves an overall device selection accuracy of 94.86%. Full deployment takes 117-165 s, while single-device redeployment takes 15-19 s. A user study with 43 participants further characterizes the deployment workflow and interaction tradeoffs, and reveals that PointRing effectively complements commercial smart speakers in typical home scenarios—particularly for spatial disambiguation and fine-grained control.
- 10 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12764-12779
PFHAR: Practically Adopting Multi-Modal Foundation Model for Human Activity Recognition Through Edge-Cloud Collaborative Learning
Multi-modal human activity recognition (HAR) is a key technology for a wide range of applications and has received widespread attention in recent years. However, the difficulty of achieving generalizability in multi-modal sensing models, combined with heterogeneous and unlabeled downstream data, significantly hinders their broader adoption. In this work, we propose PFHAR, a unified framework for practically adopting multi-modal foundation HAR model to target user groups. PFHAR uses a novel dynamic masked contrastive learning method to pre-train a foundation model on various heterogeneous public HAR datasets, ensuring strong generalizability across different modal combinations. It then adopts semi-supervised edge-cloud collaborative learning to fine-tune the pre-trained model with heterogeneous and unlabeled local data, adapting it for the target user group. Our evaluations on public and self-collected datasets demonstrate that PFHAR significantly outperforms SOTA baselines in both the pre-training and edge-cloud collaborative fine-tuning stages.
- 10 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12764-12779
PFHAR: Practically Adopting Multi-Modal Foundation Model for Human Activity Recognition Through Edge-Cloud Collaborative Learning
Multi-modal human activity recognition (HAR) is a key technology for a wide range of applications and has received widespread attention in recent years. However, the difficulty of achieving generalizability in multi-modal sensing models, combined with heterogeneous and unlabeled downstream data, significantly hinders their broader adoption. In this work, we propose PFHAR, a unified framework for practically adopting multi-modal foundation HAR model to target user groups. PFHAR uses a novel dynamic masked contrastive learning method to pre-train a foundation model on various heterogeneous public HAR datasets, ensuring strong generalizability across different modal combinations. It then adopts semi-supervised edge-cloud collaborative learning to fine-tune the pre-trained model with heterogeneous and unlabeled local data, adapting it for the target user group. Our evaluations on public and self-collected datasets demonstrate that PFHAR significantly outperforms SOTA baselines in both the pre-training and edge-cloud collaborative fine-tuning stages.
- Open Access
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-26
MASTER: A Multi-modal Foundation Model for Human Activity Recognition
Multi-modal sensing has become crucial in Human Activity Recognition (HAR) due to its ability to combine data from diverse sensors. However, challenges arise in recognizing various activities in different scenes using multi-modal data from different positions and devices, due to dynamic combinations of modal inputs, data heterogeneity, and scarcity of labeled data. To tackle these challenges, we propose MASTER, a multi-modal foundation model specifically designed for HAR. MASTER introduces a masked-data modeling-based self-supervised pre-training method, enabling the model to learn from unlabeled data and adapt to dynamic combinations of modal inputs. Moreover, it incorporates a few-shot alignment mechanism to facilitate adaptation to different activities, scenes, positions, and devices. Through the pre-training and fine-tuning on 7 multi-modal HAR datasets, MASTER currently supports, but is not limited to, 8 modalities (ACC, Gyro, mmWave, WiFi, Skeleton, Lidar, Infrared, and RGB) and 45 human activities. The results demonstrate that MASTER achieves the highest accuracy with minimal labeled data across various situations, surpassing alternative solutions.
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-26
Open AccessMASTER: A Multi-modal Foundation Model for Human Activity Recognition
Multi-modal sensing has become crucial in Human Activity Recognition (HAR) due to its ability to combine data from diverse sensors. However, challenges arise in recognizing various activities in different scenes using multi-modal data from different positions and devices, due to dynamic combinations of modal inputs, data heterogeneity, and scarcity of labeled data. To tackle these challenges, we propose MASTER, a multi-modal foundation model specifically designed for HAR. MASTER introduces a masked-data modeling-based self-supervised pre-training method, enabling the model to learn from unlabeled data and adapt to dynamic combinations of modal inputs. Moreover, it incorporates a few-shot alignment mechanism to facilitate adaptation to different activities, scenes, positions, and devices. Through the pre-training and fine-tuning on 7 multi-modal HAR datasets, MASTER currently supports, but is not limited to, 8 modalities (ACC, Gyro, mmWave, WiFi, Skeleton, Lidar, Infrared, and RGB) and 45 human activities. The results demonstrate that MASTER achieves the highest accuracy with minimal labeled data across various situations, surpassing alternative solutions.
- Open Access
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-32
Semi-supervised Contrastive Learning for Reliable Sleep Staging with Small Labeled Photoplethysmography Data
Sleep monitoring is essential for assessing the quality of the body's rest. In medicine, it involves collecting polysomnography (PSG) data in hospitals, along with expert annotations. For convenient sleep monitoring, some studies on wearable devices train models using labeled photoplethysmography (PPG) data to classify multiple sleep stages. However, the difficulty of collecting labeled PPG data limits the application of these studies. To address the impaired model performance caused by the limited labeled PPG data, we develop a sleep staging model that simultaneously inputs sleep-related features and raw data to leverage their advantages in high task correlation and rich information. Furthermore, we design a semi-supervised contrastive learning method, FASL, which adjusts the loss function based on a soft label confidence coefficient to utilize the majority of unlabeled data and enhance the model's generalization ability. Finally, we apply semi-supervised contrastive learning to the features before the Sequence Encoder to mitigate the impact of individual variations. Compared to the best semi-supervised baseline, our approach demonstrates apparent performance improvements on two publicly available datasets and one self-collected dataset. To our knowledge, this work is the first to investigate a semi-supervised sleep monitoring method using PPG data, providing guidance on how to leverage unlabeled PPG data for sleep staging. The code and self-collected dataset used in this study are publicly available at https://github.com/QiWangXPY/FASL.
- 3 September 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (3) , 1-32
Open AccessSemi-supervised Contrastive Learning for Reliable Sleep Staging with Small Labeled Photoplethysmography Data
Sleep monitoring is essential for assessing the quality of the body's rest. In medicine, it involves collecting polysomnography (PSG) data in hospitals, along with expert annotations. For convenient sleep monitoring, some studies on wearable devices train models using labeled photoplethysmography (PPG) data to classify multiple sleep stages. However, the difficulty of collecting labeled PPG data limits the application of these studies. To address the impaired model performance caused by the limited labeled PPG data, we develop a sleep staging model that simultaneously inputs sleep-related features and raw data to leverage their advantages in high task correlation and rich information. Furthermore, we design a semi-supervised contrastive learning method, FASL, which adjusts the loss function based on a soft label confidence coefficient to utilize the majority of unlabeled data and enhance the model's generalization ability. Finally, we apply semi-supervised contrastive learning to the features before the Sequence Encoder to mitigate the impact of individual variations. Compared to the best semi-supervised baseline, our approach demonstrates apparent performance improvements on two publicly available datasets and one self-collected dataset. To our knowledge, this work is the first to investigate a semi-supervised sleep monitoring method using PPG data, providing guidance on how to leverage unlabeled PPG data for sleep staging. The code and self-collected dataset used in this study are publicly available at https://github.com/QiWangXPY/FASL.
- Open Access
- 9 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (2) , 1-24
RingBP: Towards Continuous, Comfortable, and Generalized Blood Pressure Monitoring Using a Smart Ring
The utilization of wearable devices with integrated photoplethysmography (PPG) sensors for continuous and comfortable Arterial Blood Pressure (ABP) monitoring has gained popularity. However, our analysis on a large-scale self-collected dataset reveals a significant challenge: approximately 56.7% of individuals experience the absence of the dicrotic notch (a vital morphology feature) in wrist-PPG data, while 23.8% face the same issue in finger-PPG data, which poses a big obstacle to generalized ABP monitoring. To address this challenge, we propose a morphology-independent ABP estimation algorithm based on the Frank-Starling law. We also devise techniques to address calibration and PPG data drift issues commonly faced in continuous ABP monitoring. Further, a system called RingBP is developed towards continuous, comfortable, and generalized ABP monitoring using a smart ring. We design and implement a smart ring prototype and conduct extensive experiments involving 85 participants, demonstrating its superiority over other solutions.
- 9 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (2) , 1-24
Open AccessRingBP: Towards Continuous, Comfortable, and Generalized Blood Pressure Monitoring Using a Smart Ring
The utilization of wearable devices with integrated photoplethysmography (PPG) sensors for continuous and comfortable Arterial Blood Pressure (ABP) monitoring has gained popularity. However, our analysis on a large-scale self-collected dataset reveals a significant challenge: approximately 56.7% of individuals experience the absence of the dicrotic notch (a vital morphology feature) in wrist-PPG data, while 23.8% face the same issue in finger-PPG data, which poses a big obstacle to generalized ABP monitoring. To address this challenge, we propose a morphology-independent ABP estimation algorithm based on the Frank-Starling law. We also devise techniques to address calibration and PPG data drift issues commonly faced in continuous ABP monitoring. Further, a system called RingBP is developed towards continuous, comfortable, and generalized ABP monitoring using a smart ring. We design and implement a smart ring prototype and conduct extensive experiments involving 85 participants, demonstrating its superiority over other solutions.
- Open Access
- 27 September 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 7 (3) , 1-26
Combining Smart Speaker and Smart Meter to Infer Your Residential Power Usage by Self-supervised Cross-modal Learning
Energy disaggregation is a key enabling technology for residential power usage monitoring, which benefits various applications such as carbon emission monitoring and human activity recognition. However, existing methods are difficult to balance the accuracy and usage burden (device costs, data labeling and prior knowledge). As the high penetration of smart speakers offers a low-cost way for sound-assisted residential power usage monitoring, this work aims to combine a smart speaker and a smart meter in a house to liberate the system from a high usage burden. However, it is still challenging to extract and leverage the consistent/complementary information (two types of relationships between acoustic and power features) from acoustic and power data without data labeling or prior knowledge. To this end, we design COMFORT, a cross-modality system for self-supervised power usage monitoring, including (i) a cross-modality learning component to automatically learn the consistent and complementary information, and (ii) a cross-modality inference component to utilize the consistent and complementary information. We implement and evaluate COMFORT with a self-collected dataset from six houses in 14 days, demonstrating that COMFORT finds the most appliances (98%), improves the appliance recognition performance in F-measure by at least 41.1%, and reduces the Mean Absolute Error (MAE) of energy disaggregation by at least 30.4% over other alternative solutions.
- 27 September 2023
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 7 (3) , 1-26
Open AccessCombining Smart Speaker and Smart Meter to Infer Your Residential Power Usage by Self-supervised Cross-modal Learning
Energy disaggregation is a key enabling technology for residential power usage monitoring, which benefits various applications such as carbon emission monitoring and human activity recognition. However, existing methods are difficult to balance the accuracy and usage burden (device costs, data labeling and prior knowledge). As the high penetration of smart speakers offers a low-cost way for sound-assisted residential power usage monitoring, this work aims to combine a smart speaker and a smart meter in a house to liberate the system from a high usage burden. However, it is still challenging to extract and leverage the consistent/complementary information (two types of relationships between acoustic and power features) from acoustic and power data without data labeling or prior knowledge. To this end, we design COMFORT, a cross-modality system for self-supervised power usage monitoring, including (i) a cross-modality learning component to automatically learn the consistent and complementary information, and (ii) a cross-modality inference component to utilize the consistent and complementary information. We implement and evaluate COMFORT with a self-collected dataset from six houses in 14 days, demonstrating that COMFORT finds the most appliances (98%), improves the appliance recognition performance in F-measure by at least 41.1%, and reduces the Mean Absolute Error (MAE) of energy disaggregation by at least 30.4% over other alternative solutions.
- 26 March 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 23 (7) , 7244-7257
MePark: Using Meters as Sensors for Citywide On-Street Parking Availability Prediction
Real-time parking availability prediction is of great value to optimize the on-street parking resource utilization and improve traffic conditions, while the expensive costs of the existing parking availability sensing systems have limited their large-scale applications in more cities and areas. This paper presents the MePark system to predict real-time citywide on-street parking availability at fine-grained temporal level based on the readily accessible parking meter transactions data and other context data, together with the parking events data reported from a limited number of specially deployed sensors. We design an iterative mechanism to effectively integrate the aggregated inflow prediction and individual parking duration prediction for adequately exploiting the transactions data. Meanwhile, we extract discriminative features from the multi-source data, and combine the multiple-graph convolutional neural network (MGCN) and the long short-term memory (LSTM) network for capturing complex spatio-temporal correlations. The extensive experimental results based on a four-month real-world on-street parking dataset in Shenzhen, China demonstrate the advantages of our approach over various baselines.
- 26 March 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 23 (7) , 7244-7257
MePark: Using Meters as Sensors for Citywide On-Street Parking Availability Prediction
Real-time parking availability prediction is of great value to optimize the on-street parking resource utilization and improve traffic conditions, while the expensive costs of the existing parking availability sensing systems have limited their large-scale applications in more cities and areas. This paper presents the MePark system to predict real-time citywide on-street parking availability at fine-grained temporal level based on the readily accessible parking meter transactions data and other context data, together with the parking events data reported from a limited number of specially deployed sensors. We design an iterative mechanism to effectively integrate the aggregated inflow prediction and individual parking duration prediction for adequately exploiting the transactions data. Meanwhile, we extract discriminative features from the multi-source data, and combine the multiple-graph convolutional neural network (MGCN) and the long short-term memory (LSTM) network for capturing complex spatio-temporal correlations. The extensive experimental results based on a four-month real-world on-street parking dataset in Shenzhen, China demonstrate the advantages of our approach over various baselines.