skip to main content

Scaling Graph Neural Networks to 1000s of GPUs

Project

Project Details

Program
Computer Science
Field of Study
Machine Learning
Division
Computer, Electrical and Mathematical Sciences and Engineering
Center Affiliation
Extreme Computing Research Center

Project Description

Graph Neural Networks (GNNs) are a special type of deep neural networks that deal with graphs, instead of the more traditional images. GNNs are used in a variety of applications, from recommendation systems, to social networks, to computer security, to biological networks. The common characteristic is that graphs tend to be large and complex; therefore both training and inference require significant processing power. The goal of this project is to scale GNN training to thousands of GPUs. We will target our new supercomputer, Shaheen III, which is projected to include 2800 Nvidia Hopper super-chips than combine a CPU with a H100 GPU https://www.nextplatform.com/2022/09/26/kaust-hpe-shaheen-iii-supercomputer We will use the latest frameworks, such as Microsoft DeepSpeed, and we will target very large graphs.

About the Researcher

Panagiotis Kalnis
Professor, Computer Science
Computer, Electrical and Mathematical Science and Engineering Division

Affiliations

Education Profile

  • PhD Hong Kong University of Science and Technology, 2002
  • MS University of Patras, GreeceBS University of Patras, Greece, 1997

Research Interests

a€‹Professor Kalnis's research interests are in Databases and Information management. Specifically, he is interested in: Database outsourcing and cloud computing, mobile computing, Peer-to-Peer, OLAP, data warehouses, spatial-temporal and high-dimensional databases, GIS, Security - Privacy a- Anonymity.

Selected Publications

  • Tao Y., Yi K., Sheng C., Kalnis P., ""Efficient and accurate nearest neighbor and closest pair search in high dimensional space"", (to appear) ACM transactions on Data Base Systems (ACM TODS), 2010.
  • Ghinita G., Kalnis P., Tao Y., ""Anonymous publication of sensitive transactional data"", (to appear) IEEE Transactions on Knowledge and Data Engineering (IEEE TKDE), 2010.
  • Yiu M.L., Ghinita G., Jensen CS., Kalnis P., ""Enabling search services on outsourced private spatial data"", (to appear) Very Large Data Base Journal (VLDBJ), 2010.
  • Ghinita G., Zhao K., Papadias D., Kalnis P., ""A reciprocal framework for spatial K-anonymity"", Information Systems, 35(3), 299-314, 2010.
  • Ghinita G., Karras P., Kalnis P., Mamoulis N., ""A framework for efficient data anonymization under privacy and accuracy constraints"", (to appear) ACM Transactions on Data Base Systems (ACM TODS), 2009.
  • Zhu Z., Kalnis P., Bakiras S., ""DCMP: A Distributed Cycle Minimization Protocol for Peer-to-Peer Networks"", IEEE Transactions on Parallel and Distributed Systems (IEEE TPDS), 19(3), 363-377, 2008.

Desired Project Deliverables

- Tensorflow or PyTorch - based implementation - Project report

Recommended Student Background

Computer Science
Machine Learning