University of Paris-Saclay Versailles Cedex, France
In most areas of science, data production is now faster than compute capabilities. The computational modeling and data analysis associated with high-performance computing techniques are used to make these huge amounts of data effectively talk. In this talk, we highlight some challenges in the ecosystem defined by interactions among modeling, simulation and high-performance data analysis. We will then present how these challenges are addressed through several examples such as gamma ray detection in astronomy or clustering in a network of individuals and behavior anomaly detection. An innovative approach allowing the construction of adapted numerical methods for the underlying mathematics of this ecosystem as well as a parallel/distributed programming model well-suited to new and emerging high-performance architectures will be presented. Examples of experiments showing the effectiveness of this approach for high-performance linear algebra computation as well as high-performance machine learning techniques will conclude the talk.