Challenges in Scaling Empirical Dynamic Modeling#
Keichi Takahashi (Tohoku Univ.)
Abstract#
Empirical Dynamic Modeling (EDM) is an emerging data-driven framework for analyzing and predicting non-linear dynamical systems. Although researchers have successfully applied EDM in various fields such as ecology, neuroscience, and geophysics, analyzing large-scale datasets using EDM was intractable due to the lack of a high-performance implementation. To tackle this situation, we have been developing a new high-performance implementation of EDM. This talk will present our ongoing efforts in optimizing EDM for HPC systems, including algorithmic optimizations based on approximation algorithms and implementational optimizations utilizing GPUs and Vector Engines.