| The Ensemble Generator
The ensemble generator is for constructing .mat files that describe
the various neural ensembles need for the definition
and simulation of a model.
File Information
Population Name: This is a user defined name that determines
the first characters of the file name.
File Name: This is the name file that stores the ensemble data.
This name consists of the population name, "_N##' where ## is the
number of neurons in the population, and 'D##' where ## is the dimensionality
of the vector space represented by the population. This is the name
(without the .mat extension) that should be used in the definition
of the system being simulated.
Directory: This is the directory where the most recently saved
.mat file resides. The default is to put it in the 'NeuronData' directory
on the path (which was defined when installing NESim). This can be any
directory (e.g. if you want all files in the current directory, type
'.').
Note: Everytime the 'Generate Population' button is pressed,
the file is saved with the new population data.
Population Parameters
Number of Neurons: This is the number of neurons in the population.
Threshold Range: This determines the range, [-Radius,
+Radius], over which the neuron thresholds are (evenly) distributed.
The threshold is the value of the represented variable after which a
neuron begins to fire. This range is inverted for the 'off' neurons.
Thus defining the range as [-0.25, 0.75] for the 'on' neurons means
that the thresholds for the 'off' neurons will lie between [-0.75, 0.25].
Saturation Range: This is the range (in KHz) over which neurons
in the population may 'saturate'. More precisely, this is the range
over which neurons may be firing when the value of the encoded variable
reaches the (+,-) Radius.
Random Seed: This seeds the random number generator in Matlab.
A value of -1 gives a random run. A positive value sets the seed to
that value and allows for the identical regeneration of neural ensembles
under noise and with 'random' distributions.
On's Only: Choosing this box ensures that only the 'on' neurons
are in the population in the scalar case. This box is irrelevant for
higher dimensions.
Dimension: This is the dimensionality of the vector being represented
by the population (D=1 is a scalar).
Radius: This is the radius of the sphere over which the preferred
direction vectors are distributed. Choosing 1 normalizes the responses
to the unit sphere.
Precision: This defines the step size along the dimension being
encoded used for evaluating the relevant integrals. A small step size
results in more accurate decoders, but is computationally more demanding.
Weighting: This parameter determines the relative weighting
of the accuracy of the representation around zero. A weight of 0 means
all values of the encoded variable are equally important. Higher weights
mean that the accuracy of the representation near zero is proportially
more important (i.e. must be more accurate).
Neuron Parameters
Model Type: This is the type of single neuron model used in
the population. Currently, rectified linear (straight line responses
rectified to 0 before reaching threshold) and leaky-integrate-and-fire
(LIF) neurons are supported. New neuron types can be added by defining
the appropriate .m files (see LIFsoma.m for an example). Future releases
will include more neurons.
Tau Refractory: This is the absolute refractory time (in seconds)
of the LIF neurons (i.e. the time after a spike during which no other
spike may be emitted).
Tau RC: This is the RC time constant of the model LIF neurons.
This is equal to the product of the capacitance and resistance in the
circuit defining the LIF model.
Noise: This is the standard deviation of the noise used when
determining the decoders for the population. The standard deviation
provided is assumed relative to a normalized maximum firing rate. This
parameters determines the expected noise during operation in the simulation
and should generally match or be greater than the noise parameter in
the simulation panel.
Plots
Population: This plot shows the response functions of all the
neurons in the population. For populations where D>1, all response
functions are shown along the preferred direction vector only.
Linearity: This plot shows the linearity of the representation.
For D=1: The red line indicates perfect linearity, the blue line indicates
how linear this population is, and the green line indicates the error
for each point along the encoded variable. For D=2: The error surface
is ploted using vectors whose length indicates the degree of error.
For D>2: The error at 1000 randomly selected points around the hypersphere
are shown.
Decoders: This plot shows the magnitude of the optimal decoders
found for this population.
Generating Ensembles
Generate Population: Pushing this button generates the ensemble
defined by the parameters described above.
|
