<s>
In	O
machine	O
learning	O
,	O
lazy	B-General_Concept
learning	I-General_Concept
is	O
a	O
learning	O
method	O
in	O
which	O
generalization	O
of	O
the	O
training	O
data	O
is	O
,	O
in	O
theory	O
,	O
delayed	O
until	O
a	O
query	O
is	O
made	O
to	O
the	O
system	O
,	O
as	O
opposed	O
to	O
eager	B-General_Concept
learning	I-General_Concept
,	O
where	O
the	O
system	O
tries	O
to	O
generalize	O
the	O
training	O
data	O
before	O
receiving	O
queries	O
.	O
</s>
<s>
The	O
primary	O
motivation	O
for	O
employing	O
lazy	B-General_Concept
learning	I-General_Concept
,	O
as	O
in	O
the	O
K-nearest	B-General_Concept
neighbors	I-General_Concept
algorithm	I-General_Concept
,	O
used	O
by	O
online	O
recommendation	B-Application
systems	I-Application
(	O
"	O
people	O
who	O
viewed/purchased/listened	O
to	O
this	O
movie/item/tune	O
also	O
...	O
"	O
)	O
is	O
that	O
the	O
data	O
set	O
is	O
continuously	O
updated	O
with	O
new	O
entries	O
(	O
e.g.	O
,	O
new	O
items	O
for	O
sale	O
at	O
Amazon	O
,	O
new	O
movies	O
to	O
view	O
at	O
Netflix	O
,	O
new	O
clips	O
at	O
YouTube	O
,	O
new	O
music	O
at	O
Spotify	O
or	O
Pandora	O
)	O
.	O
</s>
<s>
The	O
main	O
advantage	O
gained	O
in	O
employing	O
a	O
lazy	B-General_Concept
learning	I-General_Concept
method	O
is	O
that	O
the	O
target	O
function	O
will	O
be	O
approximated	O
locally	O
,	O
such	O
as	O
in	O
the	O
k-nearest	B-General_Concept
neighbor	I-General_Concept
algorithm	I-General_Concept
.	O
</s>
<s>
Because	O
the	O
target	O
function	O
is	O
approximated	O
locally	O
for	O
each	O
query	O
to	O
the	O
system	O
,	O
lazy	B-General_Concept
learning	I-General_Concept
systems	O
can	O
simultaneously	O
solve	O
multiple	O
problems	O
and	O
deal	O
successfully	O
with	O
changes	O
in	O
the	O
problem	O
domain	O
.	O
</s>
<s>
This	O
can	O
be	O
demonstrated	O
in	O
the	O
case	O
of	O
the	O
k-NN	B-General_Concept
technique	O
,	O
which	O
is	O
instance-based	O
and	O
function	O
is	O
only	O
estimated	O
locally	O
.	O
</s>
<s>
Theoretical	O
disadvantages	O
with	O
lazy	B-General_Concept
learning	I-General_Concept
include	O
:	O
</s>
<s>
In	O
practice	O
,	O
as	O
stated	O
earlier	O
,	O
lazy	B-General_Concept
learning	I-General_Concept
is	O
applied	O
to	O
situations	O
where	O
any	O
learning	O
performed	O
in	O
advance	O
soon	O
becomes	O
obsolete	O
because	O
of	O
changes	O
in	O
the	O
data	O
.	O
</s>
<s>
Also	O
,	O
for	O
the	O
problems	O
for	O
which	O
lazy	B-General_Concept
learning	I-General_Concept
is	O
optimal	O
,	O
"	O
noisy	O
"	O
data	O
does	O
not	O
really	O
occur	O
-	O
the	O
purchaser	O
of	O
a	O
book	O
has	O
either	O
bought	O
another	O
book	O
or	O
has	O
n't	O
.	O
</s>
<s>
Lazy	B-General_Concept
learning	I-General_Concept
methods	O
are	O
usually	O
slower	O
to	O
evaluate	O
.	O
</s>
<s>
K-nearest	B-General_Concept
neighbors	I-General_Concept
,	O
which	O
is	O
a	O
special	O
case	O
of	O
instance-based	O
learning	O
.	O
</s>
<s>
Lazy	O
naive	B-General_Concept
Bayes	I-General_Concept
rules	O
,	O
which	O
are	O
extensively	O
used	O
in	O
commercial	O
spam	O
detection	O
software	O
.	O
</s>
