Individual Neurons - Computational Explorations in Cognitive Neuroscience - page 67

result of this accommodation, as we saw before, the unit
does not then get activated immediately by the second
input.

function, and can help to make sense of their underlying
biological properties.

Now, you can play with the hyst parameters and
see what they do to the unit's response properties.
In summary, you can see that there is considerable
potential for complex dynamics to emerge from the in-
teractions of these different channels. The fact that ac-
tual neurons have many such channels with even more
complex dynamics suggests that our basic point neu-
ron model is probably a lot simpler than the real thing.
Due to pragmatic constraints, we typically ignore much
of this complexity. As the simulations later in the
text demonstrate, such simplifications do not make that
much of a difference in the core aspects of behavior that
we are modeling.

Biology of the Neuron

The neuron is an electrophysiological entity that can be
understood using the principles of electricity and diffu-
sion. Inputs come into the neuron primarily through
channels located in synapses , allowing charged atoms
or ions into and out of the neuron. Different ions have
different concentrations on the inside and outside of
a neuron, and this leads to the generation of electri-
cal current when channels open and allow ions to flow
along their concentration gradients (because of diffu-
sion ) into or out of the cell. Thus, neurons become ex-
cited when positively charged sodium ions (Na + )en-
ter the cell through synaptic channels in their receiv-
ing areas called dendrites . These synaptic channels are
opened by the neurotransmitter known as glutamate ,
which is released by the sending or presynaptic neu-
ron. Different inputs can provide different amounts of
activation depending on both how much neurotransmit-
ter is released by the sender and how many channels on
the postsynaptic (receiving) neuron open as a result.
These different synaptic efficacies for different inputs,
which we refer to as weights , are critical for determin-
ing what it is that a neuron detects. Neurons become
inhibited when negatively charged chloride ions (Cl ￿ )
enter the neuron through channels that are opened by
the neurotransmitter GABA , which is released by in-
hibitory interneurons . They also have a basic neg-
ative current caused by positive ions (potassium, K + )
leaving the neuron via leak channels that are always
open. A simple equation describes the way in which
the overall electrical voltage of the cell, known as its
membrane potential , integrates all these currents into
one real-valued number.
The membrane potential that results from integrating
neural inputs in turn determines whether a given neuron
will produce an action potential or spike , which causes
neurotransmitter to be released at the ends of a neuron's
sending projection or axon , which then forms synapses
onto other neuron's dendrites (see above). The action
potential is thresholded , meaning that it only occurs
when the membrane potential (at the start of the axon,

To stop now, quit by selecting Object/Quit in the
PDP++Root window.

2.10 Summary

Neuron as Detector

The biological and functional properties of a neuron are
consistent with it being a detector , which constantly
evaluates the information available to it looking for con-
ditions that match those that it has become specialized
to detect. Whereas standard serial computers are rela-
tively general purpose computational devices, a neuron
is relatively specialized or dedicated to detecting some
particular set of things. We emphasize that it is typi-
cally difficult to describe exactly what a neuron detects
with simple verbal terms (e.g., unlike “smoke” for the
smoke detector), and that there is an important dynamic
component to its behavior. Neurons exist in huge num-
bers and operate in parallel with one another, whereas
a standard computer operates in serial performing one
operation at a time. There are good reasons to think
that neurons perform a relatively simple computation.
A mathematical description of a neuron as a detector
can be given using the framework of Bayesian statisti-
cal hypothesis testing, which produces the same form
of mathematical activation function as the point neu-
ron activation function. In sum, the detector model of
the neuron provides a good intuitive model of how they