A distributed memory architecture implementation of the False Nearest Neighbors method based on distribution of dimensions

The False Nearest Neighbors (FNN) method is particularly relevant in several fields of science and engineering (medicine, economics, oceanography, biological systems, etc.). In some of these applications, it is important to give results within a reasonable time scale, so the execution time of the FNN method has to be reduced. This paper describes two parallel implementations of the FNN method based on the distribution of embedding dimensions for distributed memory architectures. A “Single-Program, Multiple Data” (SPMD) paradigm is employed using a simple data decomposition approach where each processor runs the same program but acts on a different subset of the data. The computationally intensive part of the method lies mainly in the neighbor search and this task is therefore parallelized and executed using 4 to 64 processors. The accuracy and performance of the two parallel approaches are then assessed and compared to the best sequential implementation of the FNN method which appears in the TISEAN project. The results indicate that the two parallel approaches, when the method is run using 64 processors on the MareNostrum supercomputer, are between 17 and 37 times faster than the sequential one. Efficiency is between 26% and 59%.