Graph Neural Networks Automated Design and Deployment on Device-Edge Co-Inference Systems
Abstract: The key to device-edge co-inference paradigm is to partition models into computation-friendly and computation-intensive parts across the device and the edge, respectively. However, for Graph Neural Networks (GNNs), we find that simply partitioning without altering their structures can hardly achieve the full potential of the co-inference paradigm due to various computational-communication overheads of GNN operations over heterogeneous devices. We present GCoDE, the first automatic framework for GNN that innovatively Co-designs the architecture search and the mapping of each operation on Device-Edge hierarchies. GCoDE abstracts the device communication process into an explicit operation and fuses the search of architecture and the operations mapping in a unified space for joint-optimization. Also, the performance-awareness approach, utilized in the constraint-based search process of GCoDE, enables effective evaluation of architecture efficiency in diverse heterogeneous systems. We implement the co-inference engine and runtime dispatcher in GCoDE to enhance the deployment efficiency. Experimental results show that GCoDE can achieve up to $44.9\times$ speedup and $98.2\%$ energy reduction compared to existing approaches across various applications and system configurations.
- Yawei Li et al. Towards efficient graph convolutional networks for point cloud handling. In Proceedings of ICCV, 2021.
- Lanning Wei et al. Neural architecture search for GNN-based graph classification. ACM TOIS, 2023.
- Shih-Yuan Yu et al. Scene-graph augmented data-driven risk assessment of autonomous vehicle decisions. IEEE TITS, 2021.
- Feiyu Chen et al. Multivariate, multi-frequency and multimodal: Rethinking graph neural networks for emotion recognition in conversation. In Proceedings of CVPR, 2023.
- Yongan Zhang et al. G-CoS: GNN-accelerator co-search towards both better accuracy and efficiency. In Proceedings of ICCAD, 2021.
- Ao Zhou et al. Hardware-aware graph neural network automated design for edge computing platforms. In Proceddings of DAC, 2023.
- Jingyi Li et al. Roulette: A semantic privacy-preserving device-edge collaborative inference framework for deep learning classification tasks. IEEE TMC, 2023.
- Jiawei Shao et al. BRANCHY-GNN: A device-edge co-inference framework for efficient point cloud processing. In Proceddings of ICASSP, 2021.
- Yue Wang et al. Dynamic graph CNN for learning on point clouds. ACM TOG, 2019.
- Zhirong Wu et al. 3D ShapeNets: A deep representation for volumetric shapes. In Proceedings of CVPR, 2015.
- Mohanad Odema et al. LENS: Layer distribution enabled neural architecture search in edge-cloud hierarchies. In Proceddings of DAC, 2021.
- Mohanad Odema et al. MaGNAS: A mapping-aware graph neural architecture search framework for heterogeneous MPSoC deployment. ACM TECS, 2023.
- Hadjer Benmeziane et al. A comprehensive survey on hardware-aware neural architecture search. arXiv preprint arXiv:2101.09336, 2021.
- Kaicheng Yu et al. Evaluating the search phase of neural architecture search. In Proceedings of ICLR, 2020.
- Keyulu Xu et al. How powerful are graph neural networks? In Proceedings of ICLR, 2019.
- Junxian Huang et al. A close examination of performance and power characteristics of 4G LTE networks. In Proceedings of MobiSys, 2012.
- Python socket. [Online]. Available: https://docs.python.org/3/library/socket.html.
- zlib tool. [Online]. Available: https://www.zlib.net.
- Yufeng Zhang et al. Every document owns its structure: Inductive text classification via graph neural networks. In Proceedings of ACL, 2020.
- Matthias Fey et al. Fast graph representation learning with PyTorch Geometric. In Proceedings of ICLR, 2019.
- NVIDIA. Jetson TX2. [Online]. Available: https://www.nvidia.com.
- Raspberry Pi 4B. [Online]. Available: https://www.raspberrypi.com.
- NVIDIA. GeForce GTX1060. [Online]. Available: https://www.nvidia.com.
- Intel. Core i7-7700 Processor. [Online]. Available: https://www.intel.com.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Collections
Sign up for free to add this paper to one or more collections.