GNN-RL: Revolutionizing HPC Scheduling with AI
Presentation by Kyrian Adimora on the groundbreaking approach to HPC scheduling using Graph Neural Networks and Reinforcement Learning.
Presentation Summary
In the ever-evolving landscape of High-Performance Computing (HPC), efficient resource management remains a critical challenge. Kyrian Adimora’s presentation introduces a groundbreaking approach to HPC scheduling that harnesses the power of Graph Neural Networks (GNNs) and Reinforcement Learning (RL) to dramatically improve system performance.
Presentation on October 23rd at 5:30PM in Learned 1136.
Key Highlights
- Unveiling the limitations of traditional scheduling algorithms in modern HPC environments
- Introduction to a novel GNN-RL hybrid model for intelligent, adaptive scheduling
- Demonstration of significant improvements in resource utilization, job throughput, and makespan reduction
- Real-world application using Argonne Leadership Computing Facility workload data
- Future directions in AI-driven HPC optimization
This research represents a significant leap forward in HPC resource management, promising to accelerate scientific discoveries across various fields, from climate modeling to drug discovery.
Images from the Presentation
Overview of HPC Scheduling
GNN-RL Convergence Training Loss Analysis
Comparison of Traditional vs. AI-driven Scheduling Algorithms
Presenter Bio
Kyrian C. Adimora
Student Member, IEEE & ACM
Ph.D. Candidate, Computer Science
University of Kansas
Kyrian Adimora is currently a Ph.D. candidate in Computer Science at the University of Kansas, where he focuses on the intersection of High-Performance Computing (HPC) and Artificial Intelligence. His academic journey began with a Bachelor’s degree in Computer Engineering, awarded with First Class Honors from Michael Okpara University of Agriculture, Umudike, Nigeria. He furthered his studies by obtaining a Master’s degree in Computer Engineering from the same institution. Additionally, he holds a Postgraduate Diploma in Information Management Technology from the Federal University of Technology, Owerri, and a Higher National Diploma with Distinction in Electrical Electronics Engineering from Federal Polytechnics Nekede.
Adimora has served as a Lecturer at Michael Okpara University of Agriculture, Umudike, where he contributed to the academic development of students in engineering disciplines. He also held the position of Director of ICT at Eastern Palm University (now Kingsley Ozumba Mbadiwe University) from 2018 to 2019, where he managed the university’s information and communication technology initiatives.
His research interests include High-Performance Computing, Artificial Intelligence, Machine Learning, Cloud Computing, and IoT/Mobile Security.
Adimora is an active member of the IEEE Computer Society, Association for Computing Machinery (ACM), National Society of Black Engineers (NSBE), American Association for the Advancement of Science (AAAS), and the Council for the Regulation of Engineering in Nigeria (COREN). Notably, he has been nominated as the IEEE Kansas City Education Activity Chair and serves as a reviewer for multiple high-impact journals in his field.