Scalable Empirical Dynamic Modeling with Parallel Computing and Approximate k-NN Search

Empirical Dynamic Modeling (EDM) is a mathematical framework for modeling and predicting non-linear time series data. Although EDM increasingly adopted in various research fields, its application to large-scale data has been limited due high computational cost. This article presents kEDM, high-performance implementation of analyzing datasets. kEDM adopts the Kokkos performance-portable programming model efficiently run on both CPU GPU while sharing single code base. We also conduct hardware-specific optimization performance-critical kernels. achieved up 6.58× speedup pairwise causal inference real-world biology datasets compared an existing implementation. Furthermore, we integrate multiple approximate k-NN search algorithms into enable analysis extremely large that were intractable with conventional based exhaustive search. EDM-based forecast enhanced demonstrated 790× Simplex projection less than 1% increase MAPE.