PI Name: Jie Liu
PI Institution: argonne-national-laboratory
Collaborating ANL Division: mathematics-and-computer-science-mcs

Description: HPC systems are designed based data reuse, for example, in-memory caching prefetching techniques are heavily investigated to reduce any non-computational cost. Our project name is Braid, it brings in knowledge across entire workflow from data generation, continuous training, to inference. Our goal is to design flow-aware data management techniques to maximize training data reuse on HPC systems.

Testbed: Neddy, witherspoon, gpu_v100_smx2