Miind
Miind Documentation

News (8 January 2018)!

We have introduced support for two-dimensional population densities in a new MIIND version (1.04). This requires a new tutorial, and the MIIND tar ball version 1.04 (or a recent clone of the git repository). The old code (miind-1.03) will remain available. For that version you must use the old tutorial, but unless you are an existing MIIND user, don't use the 1.03 version (or below).

The code can handle a large number of 2D neuronal models, such as adaptive-exponential-integrate-and-fire, conductance-based, Fitzhugh-Nagumo. Each model can be run in the same way, using a MeshAlgorithm. The tutorial describes how. When defining a model, you need two kinds of files: a model file and a number of mat files. For some neural models these files are available:

Introduction

MIIND is a simulator for modeling circuits of neural populations, with an emphasis on population density techniques, co-funded by the Human Brain Project.

MIIND addresses large neuronal networks at the population level, rather than at that of individual levels. Population density techniques retain a close correspondence to spiking neuron models: informally, if you use population density techniques you should get simulation results that are close to that of spiking neuron models, but for less overhead. The correspondence between population density techniques and groups of spiking neurons is much more rigorous than for rate based models. Conceptually, we are close to the DIPDE simulator developed by the Allen Brain Institute, but we provide different algorithms. Long term, we expect a common front end for DIPDE and MIIND.

Below, we will provide an example. We assume that you have some familiarity with using neural simulators such as NEST or BRIAN, or else that you've worked with rate based models such as Wilson-Cowan dynamics.

In case of a bug, please raise an issue. (Press on the green button), or send an email to M.deK.nosp@m.amps.nosp@m.@leed.nosp@m.s.ac.nosp@m..uk

Below, we will provide an example. We assume that you have some familiarity with using neural simulators such as NEST or BRIAN, or else that you've worked with rate based models such as Wilson-Cowan dynamics.

Example 1: a one dimensional density

dense.png

The image shows a density profile: consider a population of simulated neurons, measure the potential of each neuron and represent them in a histogram. The markers are the result of a NEST simulation of 10000 leaky-integrate-and-fire neurons, the horizontal axis represents the membrane potential. The solid curve is calculated by MIIND. The entire population is calculated by a single density function. From this, one can calculate population averaged quantities, such as the firing rate. Consider the spike raster of the simulation:

spikeraster.png

This corresponds closely to the firing rate calculated from the density function:

rate.png

MIIND provides a simulation framework for neural simulations. It focusses on population level descriptions of neural dynamics, i.e. it does not provide a simulator for individual neurons, but models population activity directly. It provides support for simple neural mass models, but focusses strongly on so-called population density techniques.

To get a feeling for the simulator and its capabilities, go to the Example page.

To run the simulator, go to the Workflow page, after installation is completed. Make sure you have had a glance at the Example page.

Download

The latest tar bal can be found here. Install on Unix platforms is straightforward, using cmake. The procedure is explained in Installation. You can checkout a snapshot of the latest code with:

git clone git://git.code.sf.net/p/miind/git miind-git

In anticipation of a move to GitHub, we push our commits to a mirror repository there: the MIIND GitHub page

The components are Open Source software under a BSD licence.

Licence

MIIND is free and Open Source. It is distributed under a BSD license (see The License for MIIND)

Installation

Use cmake and make. For more details see Installation.

Dependencies

For more details see Dependencies

the latest developments.

There is a now tutorial. MIIND version 1.04 was released on 9 January 2018. It supports 2D density models.

MIIND version 1.03 was released om 9 February 2016. It contains more efficient leaky-integrate-and-fire support, and an extended example section.

MIIND version 1.02 was released on 19 September 2015. It contains a major bug fix.

MIIND version 1.01 was released on 11 March 2015. Appart from bug fixes it contains a visualization tool to monitor progress of a running simulation (including the evolution of the densities!).

MIIND version 1.0 was released 25 January 2015. It is now solely dedicated to population density techniques and neural mass models. Important new features include:

Many new features are in developement. These include:

The old neural network code, including a HMAX implementation is still available, but will no longer be maintained.

WIKI

The MIIND wiki. The WIKI contains more details about the extra work that needs to be done to install MIIND under windows.

hbp.png