Taking Compute to the Data: Our use of HPCMP Platforms in a Machine Learning Workflow

It is no secret that machine learning requires significant amounts of data and compute. The DoD has large amounts of data but access to compute can be difficult to obtain. This talk will review our use of classified HPCMP platforms to train and evaluate machine learning models to achieve mission goals.

Specifically, we will discuss our lessons learned using Singularity on HPCMP GPU nodes to:

  • Train machine learning models on multiple machine learning architectures
  • Consistently evaluate mission data on our trained models
  • Collaborate between multiple team members to train and evaluate models

PRESENTER

Churilla, Matthew
churilla@sei.cmu.edu
412-268-8891

Carnegie Mellon University / Software Engineering Institute

CO-AUTHOR

Brown, Bryan
bpbrown@sei.cmu.edu

CATEGORY

Artificial Intelligence / Machine Learning usage for HPC Applications

SYSTEMS USED

classified

SECRET

Yes