Individual Neurons - Computational Explorations in Cognitive Neuroscience - page 25

in other words, when it has detected enough smoke to
send it over the threshold for genuine concern.
The neuron also has a thresholding mechanism that
keeps it quiet until it has detected something with suffi-
cient strength or confidence to be worth communicat-
ing to other neurons. At a biological level, it takes
metabolic resources for a neuron to communicate with
other neurons, so it makes sense that this is reserved
for important events. When the integrated input (as re-
flected in the membrane potential) goes over threshold,
the neuron is said to fire, completing the integrate-and-
fire model. The neural threshold is applied right at the
start of a long and branching finger or process extend-
ing from the cell body called the axon . This axon then
forms synapses on other neuron's dendrites, providing
them with the inputs described earlier and repeating the
great chain of neural processing.
Because of the good fit between the detector model
and the integrate-and-fire view of neural function, we
can use this detector model as a computational-level de-
scription of what a neuron is doing. However, for the
detector model of the neuron to be of real use, it must
be consistent with some larger understanding of how
networks of such detectors can perform useful compu-
tations and exhibit humanlike cognition. The details
of this larger picture will be spelled out in subsequent
chapters.
To foreshadow one of the main ideas, we will see that
learning can provide a means of getting a network of
neural detectors to do something useful. Learning in
neurons involves modifying the weights (synaptic ef-
ficacies) that provide the main parameters specifying
what a neuron detects. Thus, by shaping the weights,
learning shapes what neurons detect. There are pow-
erful ways of making sure that each neuron learns to
detect something that will end up being useful for the
larger task performed by the entire network. The over-
all result is that the network after learning contains a
number of detectors that are chained together in such a
way as to produce appropriate outputs given a set of in-
puts (and to do so, it is hoped, using internal detectors
that are related in some way to those used by humans in
the way they perform a task).

2.3 The Biology of the Neuron

Having described the overall function of a neuron and
the functions of its basic parts in terms of the detector
model, we will now see how these functions are im-
plemented in the underlying biological machinery of
the neuron. In this section, we will provide a general
overview of the biology of the neuron. In the next sec-
tion, we will go into more detail on the electrophysiol-
ogy of the neuron — how the electrical and physiolog-
ical properties of the neuron serve to integrate inputs,
and trigger the thresholded communication of outputs
to other neurons.
Figure 2.2 shows a picture of a biological neuron,
with the parts labeled as described in the previous sec-
tion. Perhaps the most important biological fact about
a neuron is that it is a single cell . Thus, it has a cell
body with a nucleus , and is filled with fluid and cellu-
lar organelles and surrounded by a cell membrane ,just
like any other cell in the body. However, the neuron is
unique in having very extended fingerlike processes of
dendrites and axons. As noted previously, most of the
input coming into a neuron enters in the dendrites, and
the axon, which originates at the cell body, sends the
output signal to other neurons.
To enable different neurons to communicate with
each other despite being encased in membranes, there
are little openings in the membrane called channels
(imagine a cat door). The basic mechanisms of informa-
tion processing (i.e., for integrating inputs, threshold-
ing, and communicating outputs) in a neuron are based
on the movement of charged atoms ( ions ) in and out of
these channels, and within the neuron itself. As we will
see in greater detail in the next section, we can under-
stand how and why these ions move according to basic
principles of electricity and diffusion (diffusion is the
natural tendency of particles to spread out in a liquid or
a gas, as is evident when you pour cream into coffee, for
example). By applying these principles and their asso-
ciated equations, we can develop a mathematical picture
of how a neuron responds to inputs from other neurons.
A key element in the electrical model of the neuron is
the difference in electrical charge (voltage) of the neu-
ron relative to its external environment. This electrical