Postdoctoral Researcher - AI for High Performance Computing

Investigate novel ways to leverage AI techniques in traditional HPC workloads

Modern scientific research has become heavily reliant on large-scale computation. Whether we consider physics, climatology, pharmaceutics, or engineering, all these disciplines frequently use large simulation and/or data-processing frameworks. Given their compute-intensive nature, these frameworks use special purpose high performance computing software and hardware solutions to enable maximum scalability. The resulting high-performance computing (HPC) field has been an established component of scientific research for many years.

An interesting side effect of the increasing availability of data and compute has been the development of new AI capabilities. The recent machine learning revolution has leveraged the enormous amounts of available compute to achieve new levels of performance on a variety of tasks. Machine learning models have already revolutionized computer vision, natural language processing, speech recognition and other data processing tasks. They are now also finding their way to scientific simulation, bio-medical data analysis and other traditional HPC workloads.

Novel machine learning models offer a data-driven alternative to the principled simulation models often used in traditional scientific computing. Using experimental or simulated data, machine learning models can be trained to emulate the behavior of partial or entire simulators. Often these data-driven models produce results orders of magnitude faster than running full simulations. These alternative machine learning models have opened new avenues of scientific research, by allowing scientists to simulate and optimize models at scales that were previously thought impossible. Very recent developments, like the success of Google’s AlphaFold protein folding simulations, indicate that we are only seeing the beginning of this trend.

The goal of this research project is the development of methods to enable better AI - HPC synergy. During the project, the successful candidate will investigate novel ways to leverage machine learning models in traditional HPC workloads. This will include dynamic use of machine learning models to replace (part of) standard HPC workloads, as well as the use of data driven analytics to exploit computational patterns and optimize the use of compute infrastructure. The final framework will achieve better scalability and use of resources by using hybrid AI/HPC methods.

This project is an initiative of the Compute Systems Architecture Unit (CSA). The CSA unit researches emerging workloads and their performance on large-scale supercomputer architectures for next-generation Artificial Intelligence (AI) and high-performance computing (HPC) applications. The team is responsible for algorithm research, runtime management innovations, performance modeling, architecture simulation and prototyping for these future applications and the future systems to execute them, to reach multiple orders of magnitude better performance, energy-efficiency, and total-cost-of-ownership.