The Argonne Leadership Computing Facility’s (ALCF) mission is to accelerate major scientific discoveries and engineering breakthroughs for humanity by designing and providing world-leading computing facilities in partnership with the computational science community. We help researchers solve some of the world’s largest and most complex problems with our unique combination of supercomputing resources and computational science expertise.

The ALCF has an opening for a postdoctoral position in data management targeting AI applications at scale. This Postdoc will join the AL/ML group, a vibrant multidisciplinary team of scientists and High-Performance Computing (HPC) engineers. In the AL/ML group, we work at the forefront of HPC to push scientific boundaries, carrying out research and development in state-of-the-art data management, machine learning and statistics techniques. With the advancement of Exascale systems and the variety of novel AI hardware designed to accelerate both training and inference, the ALCF is studying the application of these techniques to a variety of our science applications, including but not limited to: Computational Chemistry, Plasma Physics, High Energy Physics, analysis of Light Source data such as that from the Advanced Photon Source, Biology, Astronomy, and other science disciplines.

This Postdoc will have a rare chance to work on exascale supercomputing systems and novel AI hardware to help solve significant real-world problems using machine learning and deep learning. ALCF researchers work in a highly collaborative environment involving science application teams, academia, and industry, as well as other national labs and agencies, to solve some of the world’s largest and most complex problems in science and engineering.

Objective:

• The goal for this postdoctoral position to work on development and scaling of the data infrastructure and software for AI applications on supercomputing systems and AI testbed systems.
• The postdoc will work on multimodal data management for science applications, focusing on integrating, organizing, and analyzing diverse data types to accelerate scientific discovery and innovation.
• A primary goal will be to research and develop data management infrastructure leveraging Aurora’s storage system, DAOS, aDistributed Asynchronous Object Storage system,to meet the needs of AI-driven applications.
• Another goal would be to evaluate vector databases and retrieval-augmented generation (RAG) on supercomputers to address the requirements of AI fine-tuning and inference applications.
• The candidate will work with diverse applications and on diverse systems, including object stores, memory-X and filesystems, to research and develop strategies to evaluate, profile and optimize AI applications at scale.
• Another key objective is to help us with design of future systems and data-management solutions to meet needs of our applications.

Benefit to ALCF:

This postdoc position will help ALCF evaluate and integrate data infrastructure needed to better facilitate AI models, including training, fine-tuning and inferencing, at scale. It will help us better understand and improve DAOS to meet the needs of AI-driven science applications. We expect the postdoc to help prototype, benchmark, and evaluate strategies to better support these workloads for Aurora.