What is it about?
Optimization problems on big graphs (such as minimum vertex cover, maximum cut, traveling salesman problems) appear in many fields including social sciences, power systems, chemistry, and bioinformatics. This work builds a high performance parallel reinforcement-learning framework with a generic programming interface to solve large-scale graph optimization problems on distributed many GPUs.
Featured Image
Photo by Nana Dua on Unsplash
Why is it important?
This work maps large graphs across distributed GPUs, and uses spatial parallelism and data parallelism to achieve unprecedented high performance (18X faster than existing ML systems). This work also conducts in-depth analysis of parallel efficiency and memory cost, and shows that the designed new algorithms and RL system are scalable on numerous distributed GPUs.
Perspectives
We hope this work will provide a foundation and "playground" for researchers to practice, invent, and create innovative reinforcement learning methods and GNN models to solve real-world NP-hard optimization problems in a faster and more effective way.
Prof. Fengguang Song
Indiana University Purdue University at Indianapolis
Read the Original
This page is a summary of: A Distributed-GPU Deep Reinforcement Learning System for Solving Large Graph Optimization Problems, ACM Transactions on Parallel Computing, June 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3589188.
You can read the full text:
Resources
Contributors
The following have contributed to this page
