-
Notifications
You must be signed in to change notification settings - Fork 46
ConceptNeuronModel
Related: ConceptSynapseModel
The Brain Simulator supports any number of models and new ones can be added easily.
- Source: Brain Simulator II _ The Guide - Charles Simon, * Chapter: Chapter 2
For convenience. This neuron model fires at a consistent rate every 'n' Neuron Engine cycles.
The “Always”-neuron-model can be useful for some digital circuit prototypes where continuous firing is needed. In practice, it is the same as the Random model with a standard deviation of zero.
Some biological neurons appear to fire bursts rather than individual spikes. Within the Brain Simulator, the “Burst” model performs a similar function.
There are two parameters:
- the number of spikes in one burst
- the rate of spikes (in cycles).
- 1 = the burst fires at the maximum neuron firing rate
In other respects, the Burst neuron model acts as an IF-neuron.
(a "Short-Cut Model")
The eye receives light and emits spikes down the optic nerve corresponding to (among other things) the color detected at any point in the visual field. The eye has separate sensors to detect different colors (red, green, blue, and gray intensity) and signals from these neurons seem to remain separate through the optic nerve.
The computer stores the RGB (or aRGB) triples in single memory words and decodes them as needed—a process that is not biologically plausible.
For convenience, there is a “Color” neuron model. It does not spike but simply stores an integer value that could represent a color. As an added convenience, the color of the neuron display in the user interface is governed by the RGB value, so if you have an array of Color neurons, you can see the color image in the user interface. Also, the internal value is displayed or modified in hexadecimal as the “Charge” of any Color neuron.
The content of a Color neuron is only useful to Modules because it generates no spikes. To extract a color from the Color neuron, you need a few lines of code which will mask off the portion of the color signal which is of interest. These components of color can then be handled by more biologically plausible neurons. The conversion from a single Color neuron to the spiking rates of four IF neurons (representing red, green blue, and gray intensity) is demonstrated by the code in the ModuleColorComponent Module.
(a "Short-Cut Model")
Like Color, there are times that a high-precision floating-point number is useful. Biological neurons have a limited range of possible, discernible values, because high noise levels (and leakage) limit the number of discrete values a neuron might represent. Also, representing a signal in a neuron’s spiking rate is limited by the speed at which a signal must be represented. Like the Color neuron, neurons with the FloatValue model do not spike and must be accessed via a software module.
(IF)
"A neuron is evaluated to determine if the internal value exceeds a threshold. If it does, then the synapses list is processed and the weight of each synapse is added to the internal value of the corresponding target neuron."
(LIF)
Leakage (ConceptNeuronLeakage)
Between incoming spikes from synapses, the 'charge' (weight) of the neuron “leaks” or decays away a little.
If the incoming spikes stop, the internal charge will gradually return to 0.0.
Leakage Rate
The fraction of charge which is 'subtracted' in each SoftwareNeuronEngineCycle.
You can set the leakage rate for individual neurons.
Can be used as a “high-pass” filter in that if the incoming spike frequency is high enough, the neuron will spike; but if it is lowered, the neuron will never spike.
/If you set the leakage rate to 0, then the LIF-neuron acts as an IF-neuron (as there will be no leakage). /Or if the incoming weight is at least the same as the threshold value of the neuron.
As the random character of neurons is not precisely known, the Brain Simulator has made a first step by incorporating a “Random” neuron model.
Neurons with this model act as IF-neurons except that in the absence of any stimulation, the neuron will fire at a random interval with a given mean and standard deviation.
The overall frequency of random firing is governed by the 'mean'. If the standard deviation is set to 0, the neuron will fire at a constant rate given by the mean. If the standard deviation is negative, the neuron will be disabled. A random neuron with a long mean will act generally like an IF neuron except that it will occasionally emit a random spike.
The Random neuron model is used internally to represent always-firing neurons.
"Setting the standard deviation to 0 will cause the neuron to fire at a constant rate."