Adapting CMA-ES for the Massively Parallelized Simulation of Neurons and Neural Circuits

Tomoki Kazawa, Tetsuya Fukuda, Akihiko Goto, Daisuke Miyamoto, Stephan Shuichi Haupt, Ryohei Kanzaki

Abstract


Parameter estimation for complex or/and high dimensional models is a general problem encountered in many scientific and engineering disciplines using simulations. We succeeded in estimating biophysical parameters of single neurons from current injection data using CMA-ES with restart implementation. We expanded the CMA-ES based solver into a new implementation of mp-LMCMA-ES which is adapted to highly parallelized computation. We succeeded in the estimation of several thousand parameters related to synaptic mechanisms in a neural circuit simulation consisting of a few tens of neurons.

Keywords


K computer, Evolutionary algorithm, NEURON, MPI, MPI spawn restart, Serialize


DOI
10.12783/dtetr/amsms2019/31846

Refbacks

  • There are currently no refbacks.