A New State of MIIND

Multiple Interacting Instantiations of Neuronal Dynamics (MIIND) [1] is a software suite for the simulation of neuronal circuits and large-scale networks. MIIND models at the level of the population. It is very easy to model circuits of populations and visualize the dynamics. Whole circuits can now be embedded in large hierarchical structures so that networks of circuits with considerable spatial complexity can be created. The activation of the networks can be visualized in 2D and 3D. Although MIIND is a C++ framework, with the associated performance benefits, much of its functionality can be accessed in Python. Uniquely, MIIND offers the possibility to simulate neural populations with population density techniques.

Conventional population simulation techniques require a large number of neurons to be simulated and then to calculate relevant population statistics by spike counting and averaging. With population density techniques these averages are simulated directly, bypassing the simulation of individual neurons. This leads to remarkably efficient simulations. MIIND is the only publicly available repository of population density algorithms. Recently, we have managed to include networks of leaky-integrate-and-fire neurons with balanced excitation-inhibition. In the diffusion approximation this entails the solution of Fokker-Planck equations, but importantly, our algorithms are also applicable outside of the diffusion regime. At present MIIND is a powerful tool for the simulation of large networks of spiking neuron populations and the presentation of their results.

MIIND is an open source framework for computational and cognitive neuroscience, which exposes application programming interfaces for the creation of neural models of cognition through design patterns. This allows rapid application development through reuse of contributed components to create complex neural models. We will demonstrate a number of applications developed with MIIND to show coupled systems of Wilson-Cowan and Fokker-Planck equations. The demonstrations are shown from within a graphical interface to a dynamical model of visual attention. The software allows the time-course of neural activity to be visualised as a complete model, as cortical circuits, and as individual neurons or neural populations.

References

1. de Kamps M, Baier V, Drever J, Dietz M, Mösenlechner L and van der Velde F. The State of MIIND. Neural Networks, 21(8), 1164 – 1181, (2008).