Da Teng
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
Da Teng has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Shizhuo Deng
Northeastern University
5 shared publications Dongyue Chen
2 shared publications- Dongyue Chen
Northeastern University
2 shared publications Hao Wang
2 shared publications- Jiaqi Chen
Beijing Institute of Technology, Beijing, China
2 shared publications Tong Jia
2 shared publications
Latest Publications
- Open Access
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 407-415
Ensemble CLIPs: Effective Zero-shot Classification with Hundreds of Multi-modal CLIPs
- 30 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 407-415
Open AccessEnsemble CLIPs: Effective Zero-shot Classification with Hundreds of Multi-modal CLIPs
- Open Access
- 3 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
SpikMamba: When SNN meets Mamba in Event-based Human Action Recognition
- 3 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessSpikMamba: When SNN meets Mamba in Event-based Human Action Recognition
- 23 July 2024
- Published by Elsevier in Neurocomputing
- Vol. 602
APPN: An Attention-based Pseudo-label Propagation Network for few-shot learning with noisy labels
- 23 July 2024
- Published by Elsevier in Neurocomputing
- Vol. 602
APPN: An Attention-based Pseudo-label Propagation Network for few-shot learning with noisy labels
- 26 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (11) , 6318-6328
LHAR: Lightweight Human Activity Recognition on Knowledge Distillation
Sensor-based Human Activity Recognition (HAR) is widely used in daily life and is the basic-level bridge to virtual healthcare in the metaverse. The current challenge is the low recognition accuracy for personalized users on smart wearable devices. The limited resource cannot support large deep learning models updated locally. Besides, integrating and transmitting sensor data to the cloud would reduce the efficiency. Considering the tradeoff between performance and complexity, we propose a Lightweight Human Activity Recognition (LHAR) framework. In LHAR, we combine the cross-people HAR task with the lightweight model task. LHAR framework is designed on the teacher-student architecture and the student network consists of multiple depthwise separable convolution layers to achieve fewer parameters. The dark knowledge distilled from the complex teacher model enhances the generalization ability of LHAR. To achieve effective knowledge distillation, we propose two optimization methods. Firstly, we train the teacher model by ensemble learning to promote teacher performance. Secondly, a multi-channel data augmentation method is proposed for the diversity of the dataset, which is a plug-in operation for the ensemble teacher model. In the experiments, we compare LHAR with state-of-art models in comparison evaluation, ablation study and the hyperparameter analysis, which proves the better performance of LHAR in efficiency and effectiveness.
- 26 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (11) , 6318-6328
LHAR: Lightweight Human Activity Recognition on Knowledge Distillation
Sensor-based Human Activity Recognition (HAR) is widely used in daily life and is the basic-level bridge to virtual healthcare in the metaverse. The current challenge is the low recognition accuracy for personalized users on smart wearable devices. The limited resource cannot support large deep learning models updated locally. Besides, integrating and transmitting sensor data to the cloud would reduce the efficiency. Considering the tradeoff between performance and complexity, we propose a Lightweight Human Activity Recognition (LHAR) framework. In LHAR, we combine the cross-people HAR task with the lightweight model task. LHAR framework is designed on the teacher-student architecture and the student network consists of multiple depthwise separable convolution layers to achieve fewer parameters. The dark knowledge distilled from the complex teacher model enhances the generalization ability of LHAR. To achieve effective knowledge distillation, we propose two optimization methods. Firstly, we train the teacher model by ensemble learning to promote teacher performance. Secondly, a multi-channel data augmentation method is proposed for the diversity of the dataset, which is a plug-in operation for the ensemble teacher model. In the experiments, we compare LHAR with state-of-art models in comparison evaluation, ablation study and the hyperparameter analysis, which proves the better performance of LHAR in efficiency and effectiveness.
- 15 June 2023
- Published by Elsevier in Knowledge-Based Systems
- Vol. 276
Self-relation attention networks for weakly supervised few-shot activity recognition
- 15 June 2023
- Published by Elsevier in Knowledge-Based Systems
- Vol. 276
Self-relation attention networks for weakly supervised few-shot activity recognition
- 2 May 2023
- Published by SPIE-Intl Soc Optical Eng
- Vol. 12642, 78-99
Research and application of optical network unit of intelligent electric meter based on gigabit-capable passive optical network
As intelligent grid has been fully promoted in recent years, electric information acquisition system has become an important factor restricting its development. In addition to the development of user-side service functions, the communication mode of the acquisition system has also become crucial. In view of the current problems of low data transmission efficiency, inadequate data acquisition and low real-time monitoring efficiency of intelligent meters, this paper proposes a method for collecting and returning electric power data based on an operator's fiber optic network. The method adds an embedded fiber network unit module to the intelligent meter, encodes electric power data and other parameters of the intelligent meter according to the DL/T645 and Q/GDW376.1 protocols of the automatic meter reading system, and realizes the meter reading function, high-speed broadband access network function, and broadband Internet access function for users through the TCP/IP protocol of fiber communication technology and the existing power line carrier technology. The experimental results show that the device has good operating status, high data acquisition success rate, and satisfactory speed test results. This method provides a new direction for realizing the informatization and automation of the intelligent grid, and provides a new way for user broadband access through combination with power line carrier.
- 2 May 2023
- Published by SPIE-Intl Soc Optical Eng
- Vol. 12642, 78-99
Research and application of optical network unit of intelligent electric meter based on gigabit-capable passive optical network
As intelligent grid has been fully promoted in recent years, electric information acquisition system has become an important factor restricting its development. In addition to the development of user-side service functions, the communication mode of the acquisition system has also become crucial. In view of the current problems of low data transmission efficiency, inadequate data acquisition and low real-time monitoring efficiency of intelligent meters, this paper proposes a method for collecting and returning electric power data based on an operator's fiber optic network. The method adds an embedded fiber network unit module to the intelligent meter, encodes electric power data and other parameters of the intelligent meter according to the DL/T645 and Q/GDW376.1 protocols of the automatic meter reading system, and realizes the meter reading function, high-speed broadband access network function, and broadband Internet access function for users through the TCP/IP protocol of fiber communication technology and the existing power line carrier technology. The experimental results show that the device has good operating status, high data acquisition success rate, and satisfactory speed test results. This method provides a new direction for realizing the informatization and automation of the intelligent grid, and provides a new way for user broadband access through combination with power line carrier.
- 1 October 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21619646,p. 1-4
NQETT: Node Quality Adjusted ETT for Wireless Mesh Networks
Compared with other routing metrics, such as RTT, PktPair, ETT has become the tendency for routing selecting in wireless mesh networks (WMNs). However, in ETT, the node quality is not considered. In this paper, the time during which the packet cached in the relay node is defined as node quality, while this caching time is for the node contending for the channel successfully to transmit it. To calculate the node quality, the average time between two nearby successfully transmission (TBT) is proposed. Then a new metric NQETT is proposed to describe the status of the networks more exactly. Simulation results show that the average throughput of our approach is 17% more than the original AODV and 9% more than ETT at least.
- 1 October 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- No. 21619646,p. 1-4
NQETT: Node Quality Adjusted ETT for Wireless Mesh Networks
Compared with other routing metrics, such as RTT, PktPair, ETT has become the tendency for routing selecting in wireless mesh networks (WMNs). However, in ETT, the node quality is not considered. In this paper, the time during which the packet cached in the relay node is defined as node quality, while this caching time is for the node contending for the channel successfully to transmit it. To calculate the node quality, the average time between two nearby successfully transmission (TBT) is proposed. Then a new metric NQETT is proposed to describe the status of the networks more exactly. Simulation results show that the average throughput of our approach is 17% more than the original AODV and 9% more than ETT at least.