contents previous next

IaFNetworkEditor

Integrate and fire networks are composed of nodes and links, where each node represents a population of cells with the same properties, and each link represents a connectivity rule. Profile or patterned (BinaryControl) inputs may be attached to cell populations.

An integrate-and-fire network is built from three types of nodes, and five possible types of links between them. These components appear at the top of the main window. Instances may be dragged off. Connections can be moved and stuck on with the left mouse button, and pulled off with the right one. Anything dragged off the window is deleted. Parameters for the active node (the last one clicked) are shown on the right and for the active link on the left.

Quantities defined here are statistical, defining such things as the projection density and repartition of cells. Actual instances of these laws, defining precisely which others each cell makes connections to, are constructed by IaFCalc.

At its simplest, a network can be made of a set of independent populations drawn from the yellow pool heap at the top left, connected by the white delta-function connections (IaFLinkDF) or pink biexponential connections (IaFLinkSC). Input can be provided with voltage profile from the heap at the top right connected to one of the populations with an orange direct electrical connection (IaFLinkGJ). Each population has a recordable field, which indicates whether the activity of these cells should be recorded for display purposes.

Delta function connections (white) deliver a brief conductance impulse to the postsyanptic cell after delay ms with propagation probability Pprop. The reversal potential is Erev and the conductance is all applied in a single timestep, but is equivalent to Gmax applied for 1ms. As with other link types, Fconn is the fraction of the target population contacted by each cell.

Biexponential connections (pink) are similar to delta function connections except that the postsynaptic cell receives an extended synaptic conductance beginning after delay, given by the difference of two exponentials with time constants Trise and Tfall. For both of these connection types, the connect field is only used if there are cells in common between the source and target populations. In that case, the default "other" value forbids self-connections, the "any" option allows them, and the "self-only" option allows only autaptic connections. This last option overrides the Fconn field and is intended primarily as a means of allowing variations in post-spike excitability by opening up a shunt on the cell that fired.

There are three further components designed to provide a hierarchical organization of the network by grouping the cells into exclusive or nonexclusive subpopulations. This is effected via container nodes from the black pile and the two grey connections: the triangular one for exclusive subsets and the zig-zag one for nonexclusive subsets. A container node houses cells whose properties are defined elsewhere in a normal node. For example, a container node may be used for all cells in a population which is then divided up into a few different cell types given by real nodes, and connected to the container by the thin triangular exclusive subset links. Then, whenever the number of cells in the top container is changed, this propagates down to all the subpopulations according to the relative fractions f in each. The number of cells is shown in white beside each population. Containers may contain containers, as long as somewhere the properties are defined by a real node.

The non-exclusive zigzag connector is used to select a random subset as some fraction f of anther population. This may be used for logical purposes, for example to select only a subset of cells to receive a certain class if inputs while their intrinsic properties are indistinguishable from the rest of the population. Or it may be used for convenience or efficiency to select only a subset of the whole population to be plotted. For example, with a large network defined hierarchically from a top container node, one could choose to keep detailed output from one in ten cells by connecting a new container to the top one with a non-exclusive link and marking this new container as the only recordable population.