1 of 8

Infrastructure Session Summary

Joe Osborn, Gabriel Perdue, Yihui Ren, Malachi Schram

AI4EIC October 2022

2 of 8

Session Summary

  • One tutorial
  • Four talks
    • Foundational models - Svitlana Volkova
    • AI/ML hardware co-design - Frank Liu
    • ML with FPGAs - Nhan Tran
    • AI/ML with HPC - Balint Joo
  • Panel discussion on infrastructure for EIC
    • Jin Huang
    • Tia Miceli
    • Mike Williams
    • Shinjae Yoo

3 of 8

MLFlow Tutorial

4 of 8

Foundational Models

Svitlana Volkova (PNNL)

How do we integrate new compute environments? Examples from industry and others

5 of 8

AI/ML Hardware co-design

  • ML moving towards the edge
  • Infrastructures needed to accomplish this at the EIC?
  • SNS example already deployed

Frank Liu (ORNL)

6 of 8

ML with FPGAs

  • Fast ML for Science workshop
    • https://indico.cern.ch/e/fml2022
  • New ways of thinking about developing models
  • Considering new computing architectures and technologies

​

Nhan Tran (FNAL)

7 of 8

AI/ML for HPC

  • AI/ML applications on Summit and Frontier
  • How can we integrate these, and other HPC resources, into EIC workflows?
  • Opportunities for compute cycles exist - are there applications? How do we build up to this?

Balint Joo (ORNL)

8 of 8

Panel Discussion

  • Panel discussion consisted of many future pointing questions and discussion:
  • Some themes from the discussion (in no particular order):
  • How do we build up both hardware and person-power infrastructure (and maintain them)?
  • Lessons learned from other facilities such as the LHC - applicable to the EIC?
  • How to design infrastructure that can tolerate rapidly evolving landscapes?
  • Importance of cross disciplinary teams
  • Accessibility of compute resources needs to be broader

​