the output of neuron i turns into 0. So, a pulse output is produced. It is
clear that the pulse generator is responsible for the modeling of the
refractory period.
3. The modified MCPCNN model
3.1. Notions and definitions
The neural network architectur e of the modified model can be
regarded as a graph G ¼ðV; EÞ,whereV is the set of nodes (neur ons)
and E is the set of edges (connections between neurons). An e xample
graph is shown in Fig. 2. There ar e many single-pair correspondences
between the nodes and edges in graph G, which means the connec-
tions between neurons in the network architecture. The numbers of
neurons and connections are supposed to be finite. In order to present
the modifiedmodelmoreclearly,somenotationsanddefinitions are
giveninthissubsection.
In the network architecture, R
i
denotes the set of neurons which
can be reached by neuron n
i
directly. In other words, 8j A R
i
,therehas
been an edge fr om i to j. The connection weight between n
i
and n
j
is
denoted by w
ij
.Ifn
j
is not in the neighbor set R
i
,thenw
ij
¼1.Notice
that the weights between neurons maybe are not symmetric, i.e.,
w
ij
a w
ji
in most cases. For example, as shown in Fig. 2, the neighbors
for the each neuron can be denoted as follows:
R
1
¼f2; 4g; R
2
¼f1; 3; 5g; R
3
¼f2; 6g; R
4
¼f1; 5; 7g;
R
5
¼f2; 4; 6; 8g R
6
¼f3; 5; 9g; R
7
¼f4; 8g; R
8
¼f5; 7; 9g;
R
9
¼f6; 8g
The connection weights from n
2
to each other neuron are
w
2j
¼
4:1; j ¼ 1;
0:5; j ¼ 3;
2:3; j ¼ 5;
1; j ¼f4; 6; 7; 8; 9g:
8
>
>
>
>
<
>
>
>
>
:
ð7Þ
In our model, each neuron has a single output Y(t).
Definition 1. A neuron is said to fire at time T Z 0, if ( εZ 0, such
that
YðtÞ¼
0; T εr t o T;
1; t ¼ T;
0; T o t r T þε:
8
>
<
>
:
ð8Þ
We denote this time as t
ðkÞ
i
to represent the kth fire time of n
i
.
The set of all firing times of n
i
is denoted by
ϝ
i
¼ft
ðf Þ
i
: f ¼1; 2; 3; …g¼ftjU
i
ðtÞ¼θ
i
ðtÞg ð9Þ
Definition 2. We use f
ik
to denote the kth fire of n
i
. Consider the
fact that a fire may be triggered directly by more than one neuron
simultaneously, we use P
ðkÞ
i
to denote the set of neurons that
trigger the kth fire of n
i
. P
ðkÞ
if
denotes the corresponding set of fires
that trigger the kth fire of n
i
.Iff
jk
ðjA R
i
Þ directly triggers the mth
fire of n
i
, we call n
j
the parent neuron of f
im
, and f
jk
the parent fire
of f
im
. In this case, n
j
A P
ðmÞ
i
and f
jk
A P
ðmÞ
if
.Forn
i
,anyfire f
jk
ðjA R
i
Þ
could potentially become the parent fire of n
i
. We use p
c
i
A R
i
to
denote the neuron that will directly determine the next fire of n
i
.
All neurons will not fire until one of its neighbors has fired,
except for the source neuron. For any given n
j
,ifj A R
i
and the kth
fire time t
ðkÞ
i
of n
i
is determined directly by t
j
f
which is the most
recent fire time of n
j
, we say the kth fire of n
i
is on the stimulation
of n
j
. Then we call j the parent of the kth fire of n
i
. In this case, n
j
is
denoted as P
ðkÞ
i
and t
j
f
is denoted as t
f
P
ðkÞ
i
. Sometimes the parent of n
i
will be changed by the fires of other neuron in R
i
. For example, if
neuron k is the parent of n
i
at time t
k
f
, and n
i
is going to fire at time
(τ) in the future. In this case, p
c
i
¼k. If a neighbor node j of n
i
fires,
the firing of n
j
could stimulate n
i
to fire before time τ, then j will
replace k and become the new parent of n
i
, i.e., p
c
i
¼j.
3.2. Neuron structure model
The proposed model is topologically organized with only local
lateral connections among neurons. The source neuron fires first,
then the firing event spreads out through the lateral connections
among the neurons, just like the propagation of a wave. Each
neuron records its parent represented the neighbor which caused
it to fire. It proves that the generated wave in the network spreads
outward with travel times proportional to the connection weight
between neurons.
The modified MCPCNN model is shown in Fig. 3. In the model,
each n
i
has one output Y
i
:
Y
i
ðtÞ¼StepðU
i
ðtÞθ
i
ðtÞÞ¼
1; U
i
ðtÞZ θ
i
ðtÞ;
0 otherwise:
ð10Þ
For i¼1,2,…,N, U
i
(t)andθ
i
ðtÞ are the internal activity and thresh old
function, respectively. t denot es the time and N ¼jVj is the total
number of neurons. The threshold function of the n
i
can be expressed
Fig. 1. The network structure of PCNN.
Fig. 2. An example graph.
G. Liu et al. / Neurocomputing 149 (2015) 1162–117 6116 4