I had the pleasure of being invited to give a talk to the Scientific Machine Learning Research Group at the Rutherford Appleton Laboratory. My talk focuses on performance considerations and trade-offs of emerging AI workflows from the I/O sub-system perspective.

Abstract: Scientific applications are currently increasing the use of surrogate models to replace expensive kernels and are including digital twins in their simulations. This creates complex workflows which have a mixture of I/O from different components: simulation, AI models, and analytics and which often require conflicting optimizations. In this talk, we will take a look at the performance of a few anticipated AI workflows and discuss ways of avoiding breaking the I/O subsystem during their deployment.

Please feel free to contact me if you’d like a copy of the slides.

Related paper:
Understanding and Leveraging the I/O Patterns of Emerging Machine Learning Analytics
A. Gainaru, D. Ganyushin, B. Xie, T. Kurc, J. Saltz, S. Oral, N. Podhorszki, F. Poeschel, A. Huebl, S. Klasky
Driving Scientific and Engineering Discoveries Through the Integration of Experiment, Big Data, and Modeling and Simulation
21st Smoky Mountains Computational Sciences and Engineering, SMC 2021, Virtual Event, October 18-20, 2021
Full article: https://www.osti.gov/servlets/purl/1860590