Scaling of a biophysical neocortical attractor model with Parallel NEURON

This work entails scaling a biophysical model of the neocortex using parallel NEURON [1] while running on a Cray XE6 and a Blue Gene / P in virtual node mode. Previous scaling experiments have been done with the SPLIT simulator on the Blue Gene / L with a similar neocortical model [2]. We chose a biophysical model of medium complexity based on the Hodgkin-Huxley formalism because this provides the capability of exploring the effects of psychotropic drugs as well as the oscillatory effects of cortical microcircuits and globally correlated network activity. Neocortical simulations were performed to determine both strong (fixed network size, increasing cores) and weak (increasing network size, fixed load per core) scaling with two variations of a square necortical patch of hypercolumns and internal minicolumns. The first simpler variation consists of minicolumns with 20 layer 2/3 pyramidal cells, 2 basket cells and 5 layer 4 pyramidal cells and has orthogonally stored memory patterns, encoded with long-range excitatory connections between individual minicolumns across hypercolumns. The second more complex variation has an additional 2 regular spiking non-pyramidal (RSNP) interneurons per minicolumn for encoding inhibitory long-range connections between hypercolumns. The long-range connections are implemented as terminal clusters with a single projection innervating a single minicolumn with multiple excitatory connections on the L2/3 pyramidal cells and surrounding minicolumns with multiple inhibitory connections on the RSNP interneurons. Patterns are encoded as sparse and randomly overlapping.

Simulations were performed with both single patches of increasing area and cascades of multiple patches with feed-forward and feed-backward projections. Individual simulations consisted of stimulation and completion of a single memory pattern within 1 second of cortical activity. Preliminary results show near linear speedups of the computational part of the simulation, but degradation of file I/O performance as the number of cores increase. Since each core writes out spiking activity after the simulation, the performance decline may be due to the ratio of core to I/O nodes and the large number of output files. With this performance analysis, further work will include measuring and scaling memory storage capacity with the described second variation of the biophysical neocortical model.

References
1. Carnevale, N.T. and Hines, M.L., The NEURON Book, Cambridge University Press (2006)
2. M. Djurfeldt, M. Lundqvist, C. Johansson, M. Rehn, Ö. Ekeberg, A. Lansner, Brain-scale simulation of the neocortex on the IBM Blue Gene/L supercomputer, IBM Journal of Research and Development, Volume 52, Number 1/2, Page 31 (2008)