<s>
Adaptive	B-Algorithm
resonance	I-Algorithm
theory	I-Algorithm
(	O
ART	O
)	O
is	O
a	O
theory	O
developed	O
by	O
Stephen	O
Grossberg	O
and	O
Gail	O
Carpenter	O
on	O
aspects	O
of	O
how	O
the	O
brain	O
processes	O
information	O
.	O
</s>
<s>
It	O
describes	O
a	O
number	O
of	O
neural	B-Architecture
network	I-Architecture
models	I-Architecture
which	O
use	O
supervised	B-General_Concept
and	O
unsupervised	B-General_Concept
learning	I-General_Concept
methods	O
,	O
and	O
address	O
problems	O
such	O
as	O
pattern	O
recognition	O
and	O
prediction	O
.	O
</s>
<s>
The	O
model	O
postulates	O
that	O
'	O
top-down	O
'	O
expectations	O
take	O
the	O
form	O
of	O
a	O
memory	O
template	O
or	O
prototype	B-Application
that	O
is	O
then	O
compared	O
with	O
the	O
actual	O
features	O
of	O
an	O
object	O
as	O
detected	O
by	O
the	O
senses	O
.	O
</s>
<s>
the	O
problem	O
of	O
acquiring	O
new	O
knowledge	O
without	O
disrupting	O
existing	O
knowledge	O
that	O
is	O
also	O
called	O
incremental	B-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
The	O
basic	O
ART	O
system	O
is	O
an	O
unsupervised	B-General_Concept
learning	I-General_Concept
model	O
.	O
</s>
<s>
There	O
are	O
two	O
basic	O
methods	O
of	O
training	O
ART-based	O
neural	B-Architecture
networks	I-Architecture
:	O
slow	O
and	O
fast	O
.	O
</s>
<s>
ART	O
3	O
builds	O
on	O
ART-2	O
by	O
simulating	O
rudimentary	O
neurotransmitter	O
regulation	O
of	O
synaptic	B-Application
activity	I-Application
by	O
incorporating	O
simulated	O
sodium	O
(	O
Na+	O
)	O
and	O
calcium	O
(	O
Ca2+	O
)	O
ion	O
concentrations	O
into	O
the	O
system	O
's	O
equations	O
,	O
which	O
results	O
in	O
a	O
more	O
physiologically	O
realistic	O
means	O
of	O
partially	O
inhibiting	O
categories	O
that	O
trigger	O
mismatch	O
resets	O
.	O
</s>
<s>
ARTMAP	O
also	O
known	O
as	O
Predictive	O
ART	O
,	O
combines	O
two	O
slightly	O
modified	O
ART-1	O
or	O
ART-2	O
units	O
into	O
a	O
supervised	B-General_Concept
learning	I-General_Concept
structure	O
where	O
the	O
first	O
unit	O
takes	O
the	O
input	O
data	O
and	O
the	O
second	O
unit	O
takes	O
the	O
correct	O
output	O
data	O
,	O
then	O
used	O
to	O
make	O
the	O
minimum	O
possible	O
adjustment	O
of	O
the	O
vigilance	O
parameter	O
in	O
the	O
first	O
unit	O
in	O
order	O
to	O
make	O
the	O
correct	O
classification	B-General_Concept
.	O
</s>
<s>
An	O
optional	O
(	O
and	O
very	O
useful	O
)	O
feature	O
of	O
fuzzy	O
ART	O
is	O
complement	O
coding	O
,	O
a	O
means	O
of	O
incorporating	O
the	O
absence	O
of	O
features	O
into	O
pattern	B-General_Concept
classifications	I-General_Concept
,	O
which	O
goes	O
a	O
long	O
way	O
towards	O
preventing	O
inefficient	O
and	O
unnecessary	O
category	O
proliferation	O
.	O
</s>
<s>
Simplified	O
Fuzzy	O
ARTMAP	O
(	O
SFAM	O
)	O
constitutes	O
a	O
strongly	O
simplified	O
variant	O
of	O
fuzzy	O
ARTMAP	O
dedicated	O
to	O
classification	B-General_Concept
tasks	O
.	O
</s>
<s>
They	O
support	O
several	O
learning	O
paradigms	O
,	O
including	O
unsupervised	B-General_Concept
learning	I-General_Concept
,	O
supervised	B-General_Concept
learning	I-General_Concept
and	O
reinforcement	O
learning	O
.	O
</s>
<s>
There	O
are	O
several	O
derived	O
neural	B-Architecture
networks	I-Architecture
which	O
extend	O
TopoART	O
to	O
further	O
learning	O
paradigms	O
.	O
</s>
<s>
LAPART	O
The	O
Laterally	O
Primed	O
Adaptive	B-Algorithm
Resonance	I-Algorithm
Theory	I-Algorithm
(	O
LAPART	O
)	O
neural	B-Architecture
networks	I-Architecture
couple	O
two	O
Fuzzy	O
ART	O
algorithms	O
to	O
create	O
a	O
mechanism	O
for	O
making	O
predictions	O
based	O
on	O
learned	O
associations	O
.	O
</s>
<s>
Additionally	O
,	O
it	O
can	O
perform	O
logical	O
inference	O
and	O
supervised	B-General_Concept
learning	I-General_Concept
similar	O
to	O
fuzzy	O
ARTMAP	O
.	O
</s>
