<s>
In	O
the	O
field	O
of	O
artificial	B-Application
intelligence	I-Application
,	O
the	O
designation	O
neuro-fuzzy	B-General_Concept
refers	O
to	O
combinations	O
of	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
and	O
fuzzy	O
logic	O
.	O
</s>
<s>
Neuro-fuzzy	B-General_Concept
hybridization	O
results	O
in	O
a	O
hybrid	B-General_Concept
intelligent	I-General_Concept
system	I-General_Concept
that	O
combines	O
the	O
human-like	O
reasoning	O
style	O
of	O
fuzzy	B-General_Concept
systems	I-General_Concept
with	O
the	O
learning	O
and	O
connectionist	O
structure	O
of	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Neuro-fuzzy	B-General_Concept
hybridization	O
is	O
widely	O
termed	O
as	O
fuzzy	B-General_Concept
neural	I-General_Concept
network	I-General_Concept
(	O
FNN	O
)	O
or	O
neuro-fuzzy	B-General_Concept
system	O
(	O
NFS	O
)	O
in	O
the	O
literature	O
.	O
</s>
<s>
Neuro-fuzzy	B-General_Concept
system	O
(	O
the	O
more	O
popular	O
term	O
is	O
used	O
henceforth	O
)	O
incorporates	O
the	O
human-like	O
reasoning	O
style	O
of	O
fuzzy	B-General_Concept
systems	I-General_Concept
through	O
the	O
use	O
of	O
fuzzy	O
sets	O
and	O
a	O
linguistic	O
model	O
consisting	O
of	O
a	O
set	O
of	O
IF-THEN	O
fuzzy	B-General_Concept
rules	I-General_Concept
.	O
</s>
<s>
The	O
main	O
strength	O
of	O
neuro-fuzzy	B-General_Concept
systems	O
is	O
that	O
they	O
are	O
universal	B-Algorithm
approximators	I-Algorithm
with	O
the	O
ability	O
to	O
solicit	O
interpretable	O
IF-THEN	O
rules	O
.	O
</s>
<s>
The	O
strength	O
of	O
neuro-fuzzy	B-General_Concept
systems	O
involves	O
two	O
contradictory	O
requirements	O
in	O
fuzzy	O
modeling	O
:	O
interpretability	O
versus	O
accuracy	O
.	O
</s>
<s>
The	O
neuro-fuzzy	B-General_Concept
in	O
fuzzy	O
modeling	O
research	O
field	O
is	O
divided	O
into	O
two	O
areas	O
:	O
linguistic	O
fuzzy	O
modeling	O
that	O
is	O
focused	O
on	O
interpretability	O
,	O
mainly	O
the	O
Mamdani	O
model	O
;	O
and	O
precise	O
fuzzy	O
modeling	O
that	O
is	O
focused	O
on	O
accuracy	O
,	O
mainly	O
the	O
Takagi-Sugeno-Kang	O
(	O
TSK	O
)	O
model	O
.	O
</s>
<s>
Although	O
generally	O
assumed	O
to	O
be	O
the	O
realization	O
of	O
a	O
fuzzy	B-General_Concept
system	I-General_Concept
through	O
connectionist	O
networks	O
,	O
this	O
term	O
is	O
also	O
used	O
to	O
describe	O
some	O
other	O
configurations	O
including	O
:	O
</s>
<s>
Deriving	O
fuzzy	B-General_Concept
rules	I-General_Concept
from	O
trained	O
RBF	B-Algorithm
networks	O
.	O
</s>
<s>
Fuzzy	O
logic	O
based	O
tuning	O
of	O
neural	B-Architecture
network	I-Architecture
training	O
parameters	O
.	O
</s>
<s>
Realising	O
fuzzy	B-General_Concept
membership	I-General_Concept
function	O
through	O
clustering	B-Algorithm
algorithms	I-Algorithm
in	O
unsupervised	B-General_Concept
learning	I-General_Concept
in	O
SOMs	O
and	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Representing	O
fuzzification	O
,	O
fuzzy	O
inference	O
and	O
defuzzification	B-General_Concept
through	O
multi-layers	O
feed-forward	O
connectionist	O
networks	O
.	O
</s>
<s>
It	O
must	O
be	O
pointed	O
out	O
that	O
interpretability	O
of	O
the	O
Mamdani-type	O
neuro-fuzzy	B-General_Concept
systems	O
can	O
be	O
lost	O
.	O
</s>
<s>
To	O
improve	O
the	O
interpretability	O
of	O
neuro-fuzzy	B-General_Concept
systems	O
,	O
certain	O
measures	O
must	O
be	O
taken	O
,	O
wherein	O
important	O
aspects	O
of	O
interpretability	O
of	O
neuro-fuzzy	B-General_Concept
systems	O
are	O
also	O
discussed	O
.	O
</s>
<s>
A	O
recent	O
research	O
line	O
addresses	O
the	O
data	B-Algorithm
stream	I-Algorithm
mining	I-Algorithm
case	O
,	O
where	O
neuro-fuzzy	B-General_Concept
systems	O
are	O
sequentially	O
updated	O
with	O
new	O
incoming	O
samples	O
on	O
demand	O
and	O
on-the-fly	O
.	O
</s>
<s>
Thereby	O
,	O
system	O
updates	O
not	O
only	O
include	O
a	O
recursive	O
adaptation	O
of	O
model	O
parameters	O
,	O
but	O
also	O
a	O
dynamic	O
evolution	O
and	O
pruning	O
of	O
model	O
components	O
(	O
neurons	O
,	O
rules	O
)	O
,	O
in	O
order	O
to	O
handle	O
concept	B-Algorithm
drift	I-Algorithm
and	O
dynamically	O
changing	O
system	O
behavior	O
adequately	O
and	O
to	O
keep	O
the	O
systems/models	O
"	O
up-to-date	O
"	O
anytime	O
.	O
</s>
<s>
Comprehensive	O
surveys	O
of	O
various	O
evolving	O
neuro-fuzzy	B-General_Concept
systems	O
approaches	O
can	O
be	O
found	O
in	O
and	O
.	O
</s>
<s>
Pseudo	O
outer	O
product-based	O
fuzzy	B-General_Concept
neural	I-General_Concept
networks	I-General_Concept
(	O
POPFNN	O
)	O
are	O
a	O
family	O
of	O
neuro-fuzzy	B-General_Concept
systems	O
that	O
are	O
based	O
on	O
the	O
linguistic	O
fuzzy	O
model	O
.	O
</s>
<s>
The	O
"	O
POPFNN	O
"	O
architecture	O
is	O
a	O
five-layer	O
neural	B-Architecture
network	I-Architecture
where	O
the	O
layers	O
from	O
1	O
to	O
5	O
are	O
called	O
:	O
input	O
linguistic	O
layer	O
,	O
condition	O
layer	O
,	O
rule	O
layer	O
,	O
consequent	O
layer	O
,	O
output	O
linguistic	O
layer	O
.	O
</s>
<s>
The	O
fuzzification	O
of	O
the	O
inputs	O
and	O
the	O
defuzzification	B-General_Concept
of	O
the	O
outputs	O
are	O
respectively	O
performed	O
by	O
the	O
input	O
linguistic	O
and	O
output	O
linguistic	O
layers	O
while	O
the	O
fuzzy	O
inference	O
is	O
collectively	O
performed	O
by	O
the	O
rule	O
,	O
condition	O
and	O
consequence	O
layers	O
.	O
</s>
<s>
Various	O
fuzzy	B-General_Concept
membership	I-General_Concept
generation	O
algorithms	O
can	O
be	O
used	O
:	O
Learning	O
Vector	O
Quantization	O
(	O
LVQ	O
)	O
,	O
Fuzzy	O
Kohonen	B-Algorithm
Partitioning	O
(	O
FKP	O
)	O
or	O
Discrete	O
Incremental	O
Clustering	B-Algorithm
(	O
DIC	O
)	O
.	O
</s>
<s>
Generally	O
,	O
the	O
POP	O
algorithm	O
and	O
its	O
variant	O
LazyPOP	O
are	O
used	O
to	O
identify	O
the	O
fuzzy	B-General_Concept
rules	I-General_Concept
.	O
</s>
