PRACE 2023 HPC Excellence Award goes to Professor Petros Koumoutsakos

Prof. Koumoutsakos has produced, throughout his career, reference quality simulations for in fluid mechanics including unsteady separated flows, biological flows and nanofluidics. In addition to simulations, a notable aspect of his work is state of the art flow control and optimization, through novel AI algorithms including stochastic optimization, neural networks, Bayesian inference and, more recently, reinforcement learning (RL).

Prof. Koumoutsakos pioneered the use of Artificial Intelligence (AI) in Fluid Mechanics at the Center of Turbulence Research at Stanford University in the early 90’s where he introduced deep neural to reconstruct the flow field of a turbulent flow using wall only information. He worked in the field of AI and Digital Twins well before they became as popular as they are today. With his students and collaborators he made foundational work on developing stochastic optimization, Bayesian inference and their multiple interfaces with learning algorithms. His group has been heavily involved in HPC, not only in the simulation of large-scale fluid mechanics systems across scales but also more recently in creating potent, open source, scalable software (KORALI) for the deployment of Machine Learning (ML) algorithms as surrogates for Bayesian inference in Uncertainty Quantification.

Among Prof. Koumoutsakos’ ground-breaking works are his contributions over the last five years towards coupling advanced machine learning with state-of-the-art scientific computing for the forecasting of complex systems, the development of closure models for coarse grained models of Partial Differential Equations and learning of effective dynamics of complex multiscale processes. He introduced the scientific Multi-Agent Deep Reinforcement Learning (sciMAD-RL) algorithm that allows for computational elements to serve a dual role as discretization points of governing equations and as learning agents for closures and error correction. He has showcased the advantage of sciMAD-RL methods over supervised learning algorithms in challenging turbulent flows. This automated discovery of closure models was shown to generalise across grid sizes and previously unseen geometries using limited data and predict flows at unprecedented Reynolds numbers. These findings have introduced exciting prospects for turbulence modeling, spanning from applications in aerodynamics to climate modeling.

Prof. Koumoutsakos is awarded for the research presented, in particular, in the following papers:

• • G. Novati, H. L. de Laroussilhe, and P. Koumoutsakos, “Automating turbulence modelling by multi-agent reinforcement learning,” Nat. Mach. Intell., 2021
  • J. H. Bae and P. Koumoutsakos, “Scientific multi-agent reinforcement learning for wall-models of turbulent flows,” Nat. Commun., vol. 13, iss. 1, 2022
  • P. R. Vlachas, G. Arampatzis, C. Uhler, and P. Koumoutsakos, “Multiscale simulations of complex systems by learning their effective dynamics,” Nat. Mach. Intell., 2022.

Efficient use of large HPC facilities and resources is central in Prof. Koumoutsakos work as for most scientists active in fluid mechanics simulation. The method he developed is particularly suited for HPC architectures as it provides a dual role to computational elements as learning agents and discretization points. This increases the computational load for each element in a deterministic (due to discretization) and a stochastic (due to learning) fashion. Also these algorithmic developments are generalizable beyond fluid flow towards advancing simulation capabilities for complex systems.

“I am deeply grateful to PRACE and the entire HPC community, not only for this incredible honour, but also for decades of amazing, sustained support of the research in my group. Throughout my career, I have been blessed with exceptional students and collaborators and this award belongs to each and every one of them.”

Professor Petros Koumoutsakos

“Petros Koumoutsakos’s nomination to the Award was amongst a series of other excellent candidates. The Committee was impressed by the quality and broad impact of Prof. Koumoutsakos’s work and acknowledges his major contribution in improving methodology and algorithms in addition to pure HPC performance.”

Laura Grigori, Chair of the HPC Excellence Award Selection Committee