<s>
In	O
machine	O
learning	O
and	O
computer	B-Application
vision	I-Application
,	O
M-theory	B-General_Concept
is	O
a	O
learning	O
framework	O
inspired	O
by	O
feed-forward	O
processing	O
in	O
the	O
ventral	O
stream	O
of	O
visual	O
cortex	O
and	O
originally	O
developed	O
for	O
recognition	O
and	O
classification	O
of	O
objects	O
in	O
visual	O
scenes	O
.	O
</s>
<s>
M-theory	B-General_Concept
was	O
later	O
applied	O
to	O
other	O
areas	O
,	O
such	O
as	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
On	O
certain	O
image	O
recognition	O
tasks	O
,	O
algorithms	O
based	O
on	O
a	O
specific	O
instantiation	O
of	O
M-theory	B-General_Concept
,	O
HMAX	O
,	O
achieved	O
human-level	O
performance	O
.	O
</s>
<s>
The	O
core	O
principle	O
of	O
M-theory	B-General_Concept
is	O
extracting	O
representations	O
invariant	O
under	O
various	O
transformations	O
of	O
images	O
(	O
translation	O
,	O
scale	O
,	O
2D	O
and	O
3D	O
rotation	O
and	O
others	O
)	O
.	O
</s>
<s>
In	O
contrast	O
with	O
other	O
approaches	O
using	O
invariant	O
representations	O
,	O
in	O
M-theory	B-General_Concept
they	O
are	O
not	O
hardcoded	O
into	O
the	O
algorithms	O
,	O
but	O
learned	O
.	O
</s>
<s>
M-theory	B-General_Concept
also	O
shares	O
some	O
principles	O
with	O
compressed	O
sensing	O
.	O
</s>
<s>
It	O
results	O
in	O
much	O
simpler	O
classification	O
problem	O
and	O
,	O
consequently	O
,	O
in	O
great	O
reduction	O
of	O
sample	B-General_Concept
complexity	I-General_Concept
of	O
the	O
model	O
.	O
</s>
<s>
Invariant	O
representations	O
has	O
been	O
incorporated	O
into	O
several	O
learning	O
architectures	O
,	O
such	O
as	O
neocognitrons	B-Algorithm
.	O
</s>
<s>
M-theory	B-General_Concept
provides	O
a	O
framework	O
of	O
how	O
such	O
transformations	O
can	O
be	O
learned	O
.	O
</s>
<s>
Another	O
core	O
idea	O
of	O
M-theory	B-General_Concept
is	O
close	O
in	O
spirit	O
to	O
ideas	O
from	O
the	O
field	O
of	O
compressed	O
sensing	O
.	O
</s>
<s>
In	O
the	O
introductory	O
section	O
,	O
it	O
was	O
claimed	O
that	O
M-theory	B-General_Concept
allows	O
to	O
learn	O
invariant	O
representations	O
.	O
</s>
<s>
As	O
a	O
result	O
,	O
sample	B-General_Concept
complexity	I-General_Concept
of	O
learning	O
compositional	O
concepts	O
may	O
be	O
greatly	O
reduced	O
.	O
</s>
<s>
M-theory	B-General_Concept
is	O
based	O
on	O
a	O
quantitative	O
theory	O
of	O
the	O
ventral	O
stream	O
of	O
visual	O
cortex	O
.	O
</s>
<s>
Prior	O
to	O
the	O
use	O
of	O
visual	O
neuroscience	O
in	O
computer	B-Application
vision	I-Application
has	O
been	O
limited	O
to	O
early	O
vision	O
for	O
deriving	O
stereo	O
algorithms	O
(	O
e.g.	O
,	O
)	O
and	O
to	O
justify	O
the	O
use	O
of	O
DoG	O
(	O
derivative-of-Gaussian	O
)	O
filters	O
and	O
more	O
recently	O
of	O
Gabor	O
filters	O
.	O
</s>
<s>
While	O
mainstream	O
computer	B-Application
vision	I-Application
has	O
always	O
been	O
inspired	O
and	O
challenged	O
by	O
human	O
vision	O
,	O
it	O
seems	O
to	O
have	O
never	O
advanced	O
past	O
the	O
very	O
first	O
stages	O
of	O
processing	O
in	O
the	O
simple	O
cells	O
in	O
V1	O
and	O
V2	O
.	O
</s>
<s>
M-theory	B-General_Concept
learning	O
framework	O
employs	O
a	O
novel	O
hypothesis	O
about	O
the	O
main	O
computational	O
function	O
of	O
the	O
ventral	O
stream	O
:	O
the	O
representation	O
of	O
new	O
objects/images	O
in	O
terms	O
of	O
a	O
signature	O
,	O
which	O
is	O
invariant	O
to	O
transformations	O
learned	O
during	O
visual	O
experience	O
.	O
</s>
<s>
M-theory	B-General_Concept
employs	O
this	O
idea	O
.	O
</s>
<s>
In	O
authors	O
applied	O
M-theory	B-General_Concept
to	O
unconstrained	O
face	O
recognition	O
in	O
natural	O
photographs	O
.	O
</s>
<s>
This	O
theory	O
can	O
also	O
be	O
extended	O
for	O
the	O
speech	B-Application
recognition	I-Application
domain	O
.	O
</s>
