<s>
In	O
statistical	B-General_Concept
learning	I-General_Concept
theory	I-General_Concept
,	O
a	O
learnable	B-General_Concept
function	I-General_Concept
class	I-General_Concept
is	O
a	O
set	O
of	O
functions	O
for	O
which	O
an	O
algorithm	O
can	O
be	O
devised	O
to	O
asymptotically	O
minimize	O
the	O
expected	O
risk	O
,	O
uniformly	O
over	O
all	O
probability	O
distributions	O
.	O
</s>
<s>
One	O
usual	O
algorithm	O
to	O
find	O
such	O
a	O
sequence	O
is	O
through	O
empirical	B-General_Concept
risk	I-General_Concept
minimization	I-General_Concept
.	O
</s>
<s>
It	O
is	O
worth	O
noting	O
that	O
at	O
least	O
for	O
supervised	O
classification	O
and	O
regression	O
problems	O
,	O
if	O
a	O
function	O
class	O
is	O
learnable	O
,	O
then	O
the	O
empirical	B-General_Concept
risk	I-General_Concept
minimization	I-General_Concept
automatically	O
satisfies	O
(	O
)	O
.	O
</s>
<s>
This	O
is	O
the	O
well-known	O
overfitting	B-Error_Name
problem	O
in	O
statistics	O
and	O
machine	O
learning	O
literature	O
.	O
</s>
<s>
The	O
way	O
to	O
choose	O
an	O
optimal	O
in	O
finite	O
sample	O
settings	O
is	O
usually	O
through	O
cross-validation	B-Application
.	O
</s>
<s>
Part	O
in	O
(	O
)	O
is	O
closely	O
linked	O
to	O
empirical	B-General_Concept
process	I-General_Concept
theory	I-General_Concept
in	O
statistics	O
,	O
where	O
the	O
empirical	O
risk	O
are	O
known	O
as	O
empirical	B-General_Concept
processes	I-General_Concept
.	O
</s>
<s>
Interplay	O
between	O
and	O
in	O
statistics	O
literature	O
is	O
often	O
known	O
as	O
the	O
bias-variance	B-General_Concept
tradeoff	I-General_Concept
.	O
</s>
