DC Field | Value | Language |
---|---|---|
dc.contributor.author | Kwon, Miryeong | ko |
dc.contributor.author | Gouk, Donghyun | ko |
dc.contributor.author | Lee, Sangwon | ko |
dc.contributor.author | Jung, Myoungsoo | ko |
dc.date.accessioned | 2022-08-24T03:00:35Z | - |
dc.date.available | 2022-08-24T03:00:35Z | - |
dc.date.created | 2022-04-20 | - |
dc.date.issued | 2022-05-09 | - |
dc.identifier.citation | 13rd Annual Non-Volatile Memories Workshop (NVMW), 2022 | - |
dc.identifier.uri | http://hdl.handle.net/10203/298050 | - |
dc.description.abstract | Graph neural networks (GNNs) process large-scale graphs consisting of a hundred billion edges, which exhibit much higher accuracy in a variety of prediction tasks. However, as GNNs are engaged with a large set of graphs and embedding data on storage, they suffer from heavy I/O accesses and irregular computation. We propose a novel deep learning framework on large graphs, HolisticGNN, that provides an easy-to-use, near-storage inference infrastructure for fast, energy-efficient GNN processing. To achieve the best end-to-end latency and high energy efficiency, HolisticGNN allows users to implement various GNN algorithms and directly executes them where the data exist in a holistic manner. We fabricate HolisticGNN's hardware RTL and implement its software on an FPGA-based computational SSD (CSSD). Our empirical evaluations show that the inference time of HolisticGNN outperforms GNN inference services using high-performance GPU by 7.1x while reducing energy consumption by 33.2x, on average. | - |
dc.language | English | - |
dc.publisher | University of California, San Diego | - |
dc.title | HolisticGNN: Geometric Deep Learning Engines for Computational SSDs | - |
dc.type | Conference | - |
dc.type.rims | CONF | - |
dc.citation.publicationname | 13rd Annual Non-Volatile Memories Workshop (NVMW), 2022 | - |
dc.identifier.conferencecountry | US | - |
dc.identifier.conferencelocation | UC San Diego | - |
dc.contributor.localauthor | Jung, Myoungsoo | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.