Neuron
From BEAM Robotics Wiki
[edit] Under Construction
Please be patient
Neurons are analog, or more accurately, Mixed Signal processing elements that are the fundamental building blocks of a neural network, be it a biological nervous system, or an artificial construct of sme kind. Neurons can be embodied in the form of a biological cell, an electronic circuit, or as a series of instructions in a computer program.
When a sufficient number neurons are assembled together, and the right connections are made between them (as in the brain), neurons can perform either or both information processing or motor control functions.
[edit] Logic Gates vs. Neurons
It has been said that a neuron is to a brain as a logic gate is to a digital computer. And as a simple abstract comparison that sounds good. However, if taken too far, the camparison could minimize actual capabilities of the individual neuron.
A neuron is in fact a much more sophisticated mechanism than a logic gate is, and thus carry out far more complex functions.
NOTE: The following comparisons are made between common CMOS logic gates and biological neurons. Althought Logic gate]]s that have 3-state outputs are not mentioned in this comparison, they would not demonstrate a significant increase in processing capability over logic gates that do not have 3-state outputs.
- - - - - - -
- Logic gates are strictly binary digital devices. That is to say that they are designed and intended to respond to, and produce, only one or the other of two possible input states (high or low, 1 or 0, on or off).
- Neurons are Analog or Mixed Signal devices. Depending on the exact type of neuron, they can respond to, and produce eithergraded potentials or pulses called Action Potentials.
- - - - - - -
- Logic gates typically have a fixed threshold (relative to supply voltage)
- Neuron Thresholds can be made to vary (as needed) within a given range
- - - - - - -
- Logic gate fan-out (the number of possible connections to other logic gates) is comparatively small, typically below 100.
- Neuron fan-out (the number of possible connections to other neurons) is comparatively very large, numberin in the thousands.
- - - - - - -
- Connections between logic gates (once made) typically remain fixed.
- Connections between neurons can be added or removed, strengthened or weakened, with learning.
- - - - - - -
- Logic gates function very fast (in the nano-second range).
- Neuron function is comparatively quite slow (in the milli-second range)
- - - - - - -
- Logic gates are synchronous devices in that the outputs changes only in respond to changes in concurrent input signals affect signal.
- Neurons are more or less asynchronous mechanisms as the incoming signals do not need to be concurrent. They can (within limits) arrive at different moments in time, and still be summed together to effect the output.
- - - - - - -
- Signals on logic gate inputs are seen as having an absoulute value of "0" (low) or "1" (high).
- Signals received by a Neurons can vary in value. The neuron sees them as having an initial value or "weight}, which is determined in part by where along the dendrite (the input portion of the neuron) a signal is received, and how long ago it arrived. In other words, a signal's original weight wi]]ll decrease over time. Therefore a newly received signal will carry more [[weight (hold a greater value) than will an equivalent signal received at the same location, but at an ealier time.
- - - - - - -
