<s>
Competitive	B-Algorithm
learning	I-Algorithm
is	O
a	O
form	O
of	O
unsupervised	B-General_Concept
learning	I-General_Concept
in	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
,	O
in	O
which	O
nodes	O
compete	O
for	O
the	O
right	O
to	O
respond	O
to	O
a	O
subset	O
of	O
the	O
input	O
data	O
.	O
</s>
<s>
A	O
variant	O
of	O
Hebbian	O
learning	O
,	O
competitive	B-Algorithm
learning	I-Algorithm
works	O
by	O
increasing	O
the	O
specialization	O
of	O
each	O
node	O
in	O
the	O
network	O
.	O
</s>
<s>
It	O
is	O
well	O
suited	O
to	O
finding	O
clusters	B-Algorithm
within	O
data	O
.	O
</s>
<s>
Models	O
and	O
algorithms	O
based	O
on	O
the	O
principle	O
of	O
competitive	B-Algorithm
learning	I-Algorithm
include	O
vector	B-Algorithm
quantization	I-Algorithm
and	O
self-organizing	B-Algorithm
maps	I-Algorithm
(	O
Kohonen	B-Algorithm
maps	I-Algorithm
)	O
.	O
</s>
<s>
There	O
are	O
three	O
basic	O
elements	O
to	O
a	O
competitive	B-Algorithm
learning	I-Algorithm
rule	O
:	O
</s>
<s>
The	O
neuron	O
that	O
wins	O
the	O
competition	O
is	O
called	O
a	O
"	B-Algorithm
winner-take-all	I-Algorithm
"	I-Algorithm
neuron	I-Algorithm
.	O
</s>
<s>
Competitive	B-Algorithm
Learning	I-Algorithm
is	O
usually	O
implemented	O
with	O
Neural	B-Architecture
Networks	I-Architecture
that	O
contain	O
a	O
hidden	O
layer	O
which	O
is	O
commonly	O
known	O
as	O
“	O
competitive	O
layer	O
”	O
.	O
</s>
<s>
Here	O
is	O
a	O
simple	O
competitive	B-Algorithm
learning	I-Algorithm
algorithm	O
to	O
find	O
three	O
clusters	B-Algorithm
within	O
some	O
input	O
data	O
.	O
</s>
<s>
Thus	O
,	O
as	O
more	O
data	O
are	O
received	O
,	O
each	O
node	O
converges	O
on	O
the	O
centre	O
of	O
the	O
cluster	O
that	O
it	O
has	O
come	O
to	O
represent	O
and	O
activates	O
more	O
strongly	O
for	O
inputs	O
in	O
this	O
cluster	O
and	O
more	O
weakly	O
for	O
inputs	O
in	O
other	O
clusters	B-Algorithm
.	O
</s>
