Application of AI to HPC: Smart Simulations and System Analysis
TimeTuesday, June 23rd2:15pm - 2:45pm
DescriptionAs the use of AI techniques accelerates in the HPC field, the scientific community is starting to adopt AI and in particular Deep Learning, as means to enhance and potentially replace scientific simulations. In recent studies, data-driven AI approaches have been shown to reach an accuracy that exceeds or comes very close to traditional scientific modeling methods.
Applying AI techniques to HPC simulations can speed-up the simulation time by predicting what regions, data, and exploratory techniques are most likely to be useful and omitting the rest. These so-called Smart (or Intelligent) Simulations are an emerging topic in various scientific domains. Smart Simulations employ AI tools for tasks such as: model parameter optimization, initial condition generation and optimization, full/partial model replacement, parametrization enhancement. This talk will cover ongoing projects at Cray on enabling and supporting Smart Simulations and their applicability in fields such as CFD, Cosmology and Ocean Modelling.
The second part of the talk will focus on the more general use of AI for improving the overall functioning and utilization of HPC systems. A first example explores how AI and predictive analytics can be applied to telemetry and log data collected on Cray/HPE systems. Another use-case looks at applying methods such as NLP, unsupervised learning and correlative analysis to system snapshots for detecting anomalies.