Thursday, 27th of February 2020, 12:00 – 1:00

Is zettascale computing possible before exascale platforms?

SR1, ICT Building,
Technikerstraße 21a, 6020 Innsbruck

Shantenu Jha
Rutgers University


We outline the vision of “Learning Everywhere” which captures the possibility and impact of learning methods coupled to traditional HPC methods. We discuss (i) the potential promise of the effective performance improvements for traditional HPC simulations that ML for HPC (MLforHPC) provides; (ii) identify and survey recent problems that have either been directly impacted or could benefit from MLforHPC, and (iii) Provide a taxonomy of the modes and methods by which MLforHPC can impact computational science. We identify three MLforHPC scenarios: MLinHPC, MLoutHPC and MLaroundHPC. We will discuss how learning methods and HPC simulations are being integrated, and provide representative examples. We will discuss specific applications and software systems developed for ML driven MD simulations. We also identify a spectrum of challenges and requirements to deliver on the potential impact and stress that it requires both new cyberinfrastructure and new application developments; large gains are not seen by just optimizing the environment.


Nach oben scrollen