Graph Convolutional Network for a GPU

 
Project contacts: Loc Hoang (l_hoang@utexas.edu) Roshan Dathathri (roshan@cs.utexas.edu)
Project description: Project description: GPUs have become popular platform for improving the performance of graph analytical systems. CUDA is the popular language for developing applications for GPUs. There are many graph analytics systems [1, 2] for GPUs that are implemented using CUDA. Graph neural networks [3] are an emerging research area in which graph analytics are combined with deep neural networks for the tasks such as vertex classification and link prediction. Many systems exist for both CPUs and GPUs [4, 5, 6] that implement graph convolutional networks (GCN) [7], a type of GNN. However, it is not clear if their performance can be further improved. In this project, you will implement GCN using CUDA for a single GPU. You can implement using the data structures in the Galois library [8]. The goal should be to beat the performance of existing implementations.

Hardware:

• Bridges cluster P100 nodes: 1 machines each with 2 NVIDIA P100 GPUs.
• Internal Machine located in UT called Tuxedo: 1 machine with 4 K80 GPUs and 2 GTX 1080 GPUs.

Project deliverables and deadlines:

• (Nov 6) A clear statement in English describing your project proposal.
• (Nov 13) An understanding of the project and source code for implementing your ideas and a collection of the baseline performance results of GCN on the hardware you’re using.
• (Dec 6) An implementation of GCN for a GPU.
• (Dec 6) A project report, written like an ACM conference paper, that summarizes the work you did.

Papers:
[1] Sreepathi Pai, Keshav Pingali: A compiler for throughput optimization of graph algorithms on GPUs. OOPSLA 2016: 1-19


[2] Gluon: a communication-optimizing substrate for distributed heterogeneous graph analytics, PLDI’18, https://dl.acm.org/citation.cfm?id=3192404

[3] M. Gori, G. Monfardini, F. Scarselli. A new model for learning in graph domains. IEEE Conference on Neural Networks 2005.


[4] M. Wang et al. Deep Graph Library: Towards Efficient and Scalable Deep Learning on Graphs. ICLR 2019 Workshop.


[5] Z. Jia, S. Lin, M. Gao, M. Zaharia, Alex Aiken. Improving the Accuracy, Scalability, and Performance of Graph Neural Networks with Roc. MLSys 2020.


[6] Pytorch Geometric. https://pytorch-geometric.readthedocs.io/en/latest/


[7] T. Kipf, M. Welling. Semi-Supervised Classification with Graph Convolutional Networks. ICLR 2017.


[8] Galois Library. https://github.com/IntelligentSoftwareSystems/Galois