Biomedical Engineering Reference
In-Depth Information
describe sufficiently well a large part of the information characteristics of the
neuron, but it is too complex for the simulation of information processes in the
networks, which consist of a large quantity (many thousands) of neurons. Since our
target is the study of the properties of neuron assemblies, i.e., the large populations
of neurons, it is necessary for us to make the maximum permissible number of
simplifications, ensuring only that these simplifications would not lead to the loss of
the most interesting properties of neuron assemblies. The neuron model given
below, in our opinion, is sufficient to achieve this goal, and at the same time its
simplicity makes it possible to facilitate the fulfillment of computerized numerical
experiments and the development of the corresponding specialized hardware
(neurocomputer).
Let us examine the neuron model, which has one output and many inputs
(Fig.
5.1
). The signal at the output can have two values, 0 or 1.
If the output signal is equal to 1, then the neuron is excited. We shall distinguish
between several types of input: set (
S
), inhibitory (
R
), associative (
A
), training (
Tr
),
synchronizing (
C
), and threshold control (
Th
). Let us designate inputs and the
output of the neuron with uppercase letters, and signals on them with the
corresponding lowercase letters. The set (
S
), inhibitory (
R
), associative (
A
), and
synchronizing inputs (
C
) are binary-valued. We shall say that the signal is present at
the input if it is equal to “1,” and it is absent at the input if it is equal to “0.” The
inputs
Tr
and
Th
obtain gradual signals whose ranges are -1
<
tr
<
1and0
<
th
<
n
,
where
n
is a quantity of associative inputs.
The neuron works as follows. The synchronizing input obtains the signal, which
has alternately low (0) and high (1) levels. The signal at the neuron output (
q
)
can change only at the moment when the synchronizing signal passes from the low
level to the high. If at this moment, on at least one of the inhibitory input signals,
r
equals 1, then the signal at the output takes the 0 value (
q
= 0), regardless of what
signals enter other inputs. So, inhibitory inputs dominate over all other inputs.
If at the inputs
R
the signals are absent, and on at least one of the inputs
S
the
signal is present, then at the neuron output appears (
q
= 1). If the signals at the
inputs
S
and
R
are absent, then the value of the output signal is determined in
accordance with the expression:
8
<
if
P
n
1
;
a
i
w
i
>
th
;
i¼
1
q
¼
(5.1)
if
P
n
:
0
;
a
i
w
i
th
;
i
¼
1
Fig. 5.1. Neuron model
