Gaussian Infinite Impulse Response Projection Process

Description:

Connects 2-dimensional sheets to each other. Connection fields are gaussians, or the first or second order gaussian derivatives along each dimension, with individual or globally specified sigma values. No learning. this implementation of the gaussian is considerably faster than the weight list implementation, and performance is irrespective of the sigma value used, but this speed comes with some usage limitations.

State data:

sigma (DOUBLE) 2x1/1x1:

Either a single global sigma value for all dimensions, or a length 2 array with specific sigma values for each of the dimensions.

cacheNum (DOUBLE) 1x1:

A value that can be used to uniquely identify a projection's cache, required when using multiple learning projections with identical starting parameters.

offset (DOUBLE) 2x1 (default = [0 0]):

A list of 2 offsets given in Locs along each of the 2 dimensions. Shifts the projection pattern centre on the destination sheet.

type (STRING) ('first_derivative_vert'; 'first_derivative_horiz'; 'second_derivative_vert'; 'first_derivative_horiz'; 'gaussian') (default = 'gaussian'):

Defines a dimension to have a non gaussian form.

delay (DOUBLE) 1x1 (default = 0):

Number of simulation time seconds that data will be delayed before being passed on. If delay is not an exact number of simulation iterations a warning will be generated.

invert (UINT32) 1x1 (default = 0):

Invert the connection pattern. (does nothing).

decimate (DOUBLE) 1x1 (default = 0):

Amount of weights to randomly cull (0->1, 0 = none, 1 = all).

norm (STRING) ('none'; 'aff'; 'affx'; 'eff'; 'effx'; 'proj') (default = 'affx'):

The normalisation type. 'none' is no normalisation, 'aff' is normalisation over afferent connections, with 'affx' using the MAXIMUM normalisation value for all connections (to eliminate edge effects), 'eff' is normalisation over efferent connections, with 'effx' using the MAXIMUM normalisation value for all connections (to eliminate edge effects), 'proj' normalises over projections (efferents AND afferents).