Distributed dynamics analysis of spiking neural network simulations

In order increase the chance of the adoption of new parallelization strategies and algorithms in the computational neuroscience community, one needs first to quantify qualitatively and quantitatively their impact on the almost chaotic dynamic behavior of neural microcircuit simulations. The analysis of this dynamics behavior is computationally expensive, as a large number of simulation scenarios need to be considered and high-dimensional data sets have to be analyzed. In this work, we proposed an improvement of the dimensionality reduction approach considered in the dynamics analysis methodology. Furthermore, by adding support for distributed processing with MPI to the existing OpenMP-based spiking neural network simulator, we enable the processing of such scenarios on high-performance computing resources. The preliminary results are presented in the paper. They have been obtained at the simulation of small-world neural microcircuits with hybrid parallelization on a Linux cluster with 64 cores. As such, the approach undertaken here provides partial solutions to both challenges outlined above.