Birds of a Feather:
Artificial Intelligence and Performance Analysis/Optimization
Birds of a Feather
AI/Machine Learning/Deep Learning
Performance Analysis and Optimization
Programming Models & Languages
TimeTuesday, June 26th2:45pm - 3:45pm
DescriptionArtificial intelligence techniques are progressing at a phenomenal pace, and have shown enormous potential for the analysis of high-dimensional data and the solution of very complex optimization problems. Performance analytics/visualization of parallel codes and optimization of same (including compilation and auto-tuning) will be able to profit from adapting proven AI techniques by achieving a greater degree of automation & user-friendliness, higher efficiency and improved results.
Conversely, AI applications can benefit significantly from applying rigorous parallel performance analysis and optimization techniques. The complexity of use cases and the data volumes involved continuously increase, outgrowing the capabilities of single-node systems in particular for the extremely compute-intensive training phases.
This BoF will bring together experts from the HPC performance analytics and optimization community, end-users of both HPC and AI applications, and AI/ML framework and algorithm developers to learn about recent results in both topical areas and discuss which approaches will best enable progress to be made.
The technical topics will include (i) the use of AI/ML to analyze parallel performance data or create insightful visualizations, (ii) the use of AI/ML to drive optimizations of parallel applications in a compilation or auto-tuning setting, and (iii) insights into fundamental performance properties of AI algorithms and ways to apply the proven performance analytics and optimization techniques developed in HPC.
This BoF will build on two similar events at ISC 2017 and SC17. Presenters include Jesus Labarta/Judit Gimenez (BSC), Felix Wolf (TU Darmstadt), Kristian Kersting (TU Darmstadt),
Wolfgang Nagel/Holger Brunst (TU Dresden), and Martin Schulz (TU Muenchen).