Haoyang Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Haoyang Wang has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Xiaoya Fan
Northwestern Polytechnical University
4 shared publications - Chuxi Li
Northwestern Polytechnical University
2 shared publications - Jianfeng An
Northwestern Polytechnical University, Xi'an, China, China
2 shared publications - Kaijie Feng
School of Computer Science, Northwestern Polytechnical University, Xian, China
2 shared publications - Meng Zhang
School of Computer Science, Northwestern Polytechnical University, Xi’an, China
2 shared publications - Shengbing Zhang
School of Computer Science, Northwestern Polytechnical University, Xi’an, China
2 shared publications
Latest Publications
- 6 February 2026
- Published by Springer Nature
- p. 247-260
Enabling Efficient Distributed Graph Neural Network Acceleration with Near Memory Processing
- 6 February 2026
- Published by Springer Nature
- p. 247-260
Enabling Efficient Distributed Graph Neural Network Acceleration with Near Memory Processing
- 14 January 2025
- Published by Elsevier in Neurocomputing
- Vol. 623
CSDSE: An efficient design space exploration framework for deep neural network accelerator based on cooperative search
- 14 January 2025
- Published by Elsevier in Neurocomputing
- Vol. 623
CSDSE: An efficient design space exploration framework for deep neural network accelerator based on cooperative search
- 5 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Embedded Systems Letters
- Vol. 16 (4) , 345-348
Energy-Efficient Personalized Federated Continual Learning on Edge
Federated learning (FL) on the edge devices must support continual learning (CL) to handle continuously evolving the data and perform the model training in an energy-efficient manner to accommodate the devices with limited computational and energy resources. This letter proposes an energy-efficient personalized federated CL (FCL) framework for the edge devices. The network structure on each device is divided into parts for retaining old knowledge and learning new knowledge, training only part of the model to reduce overhead. A data-free parameter selection approach selects important parameters from the trained model to retain old knowledge. During new task learning, a federated search method determines a resource-adaptive personalized model structure for each device. Experimental results demonstrate that our method can effectively support FCL in an energy-efficient manner on the edge devices.
- 5 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Embedded Systems Letters
- Vol. 16 (4) , 345-348
Energy-Efficient Personalized Federated Continual Learning on Edge
Federated learning (FL) on the edge devices must support continual learning (CL) to handle continuously evolving the data and perform the model training in an energy-efficient manner to accommodate the devices with limited computational and energy resources. This letter proposes an energy-efficient personalized federated CL (FCL) framework for the edge devices. The network structure on each device is divided into parts for retaining old knowledge and learning new knowledge, training only part of the model to reduce overhead. A data-free parameter selection approach selects important parameters from the trained model to retain old knowledge. During new task learning, a federated search method determines a resource-adaptive personalized model structure for each device. Experimental results demonstrate that our method can effectively support FCL in an energy-efficient manner on the edge devices.
- 6 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (11) , 3997-4008
NDPGNN: A Near-Data Processing Architecture for GNN Training and Inference Acceleration
Graph neural networks (GNNs) require a large number of fine-grained memory accesses, which results in inefficient use of bandwidth resources. In this article, we introduce a near-data processing architecture tailored for GNN acceleration, named NDPGNN. NDPGNN provides different operating modes to meet the acceleration needs of various GNN frameworks while ensuring the configurability and scalability of the system. NDPGNN takes advantage of data locality characteristics to repeatedly distribute and utilize data, thereby reducing memory access requirements, and further improving memory access efficiency by combining a subgraph sparse node scheduling strategy with intermediate result reuse. We use data packaging to provide a higher effective data ratio for long-distance data transmission, thereby improving the utilization of the system’s limited bandwidth resources. Compared with the previous method, NDPGNN brings 5.68 times improvement in system performance while reducing energy consumption overhead by 8.49 times.
- 6 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (11) , 3997-4008
NDPGNN: A Near-Data Processing Architecture for GNN Training and Inference Acceleration
Graph neural networks (GNNs) require a large number of fine-grained memory accesses, which results in inefficient use of bandwidth resources. In this article, we introduce a near-data processing architecture tailored for GNN acceleration, named NDPGNN. NDPGNN provides different operating modes to meet the acceleration needs of various GNN frameworks while ensuring the configurability and scalability of the system. NDPGNN takes advantage of data locality characteristics to repeatedly distribute and utilize data, thereby reducing memory access requirements, and further improving memory access efficiency by combining a subgraph sparse node scheduling strategy with intermediate result reuse. We use data packaging to provide a higher effective data ratio for long-distance data transmission, thereby improving the utilization of the system’s limited bandwidth resources. Compared with the previous method, NDPGNN brings 5.68 times improvement in system performance while reducing energy consumption overhead by 8.49 times.
- 27 February 2024
- Published by Springer Nature
- p. 1-23
CSDSE: Apply Cooperative Search to Solve the Exploration-Exploitation Dilemma of Design Space Exploration
- 27 February 2024
- Published by Springer Nature
- p. 1-23
CSDSE: Apply Cooperative Search to Solve the Exploration-Exploitation Dilemma of Design Space Exploration