<s>
In	O
statistics	O
and	O
machine	O
learning	O
,	O
double	B-General_Concept
descent	I-General_Concept
is	O
the	O
phenomenon	O
where	O
a	O
statistical	O
model	O
with	O
a	O
small	O
number	O
of	O
parameters	O
and	O
a	O
model	O
with	O
an	O
extremely	O
large	O
number	O
of	O
parameters	O
have	O
a	O
small	O
error	O
,	O
but	O
a	O
model	O
whose	O
number	O
of	O
parameters	O
is	O
about	O
the	O
same	O
as	O
the	O
number	O
of	O
data	B-Application
points	I-Application
used	O
to	O
train	O
the	O
model	O
will	O
have	O
a	O
large	O
error	O
.	O
</s>
<s>
It	O
was	O
discovered	O
in	O
2019	O
when	O
researchers	O
were	O
trying	O
to	O
reconcile	O
the	O
bias-variance	B-General_Concept
tradeoff	I-General_Concept
in	O
classical	O
statistics	O
,	O
which	O
states	O
that	O
having	O
too	O
many	O
parameters	O
will	O
yield	O
an	O
extremely	O
large	O
error	O
,	O
with	O
the	O
2010s	O
empirical	O
observation	O
of	O
machine	O
learning	O
practitioners	O
that	O
the	O
larger	O
models	O
are	O
,	O
the	O
better	O
they	O
work	O
.	O
</s>
