<s>
Hierarchical	B-Algorithm
temporal	I-Algorithm
memory	I-Algorithm
(	O
HTM	O
)	O
is	O
a	O
biologically	O
constrained	O
machine	B-Application
intelligence	I-Application
technology	O
developed	O
by	O
Numenta	O
.	O
</s>
<s>
Originally	O
described	O
in	O
the	O
2004	O
book	O
On	O
Intelligence	O
by	O
Jeff	O
Hawkins	O
with	O
Sandra	O
Blakeslee	O
,	O
HTM	O
is	O
primarily	O
used	O
today	O
for	O
anomaly	B-Algorithm
detection	I-Algorithm
in	O
streaming	O
data	O
.	O
</s>
<s>
Unlike	O
most	O
other	O
machine	O
learning	O
methods	O
,	O
HTM	O
constantly	O
learns	O
(	O
in	O
an	O
unsupervised	B-General_Concept
process	O
)	O
time-based	O
patterns	O
in	O
unlabeled	O
data	O
.	O
</s>
<s>
When	O
applied	O
to	O
computers	O
,	O
HTM	O
is	O
well	O
suited	O
for	O
prediction	O
,	O
anomaly	B-Algorithm
detection	I-Algorithm
,	O
classification	O
,	O
and	O
ultimately	O
sensorimotor	O
applications	O
.	O
</s>
<s>
A	O
typical	O
HTM	O
network	O
is	O
a	O
tree-shaped	O
hierarchy	O
of	O
levels	O
(	O
not	O
to	O
be	O
confused	O
with	O
the	O
"	O
layers	O
"	O
of	O
the	O
neocortex	O
,	O
as	O
described	O
below	O
)	O
.	O
</s>
<s>
It	O
relies	O
on	O
a	O
data	B-General_Concept
structure	I-General_Concept
called	O
sparse	O
distributed	O
representations	O
(	O
that	O
is	O
,	O
a	O
data	B-General_Concept
structure	I-General_Concept
whose	O
elements	O
are	O
binary	O
,	O
1	O
or	O
0	O
,	O
and	O
whose	O
number	O
of	O
1	O
bits	O
is	O
small	O
compared	O
to	O
the	O
number	O
of	O
0	O
bits	O
)	O
to	O
represent	O
the	O
brain	O
activity	O
and	O
a	O
more	O
biologically-realistic	O
neuron	O
model	O
(	O
often	O
also	O
referred	O
to	O
as	O
cell	O
,	O
in	O
the	O
context	O
of	O
HTM	O
)	O
.	O
</s>
<s>
Cortical	O
learning	O
algorithms	O
are	O
currently	O
being	O
offered	O
as	O
commercial	O
SaaS	B-Architecture
by	O
Numenta	O
(	O
such	O
as	O
Grok	O
)	O
.	O
</s>
<s>
To	O
which	O
Jeff	O
's	O
response	O
was	O
"	O
There	O
are	O
two	O
categories	O
for	O
the	O
answer	O
:	O
one	O
is	O
to	O
look	O
at	O
neuroscience	O
,	O
and	O
the	O
other	O
is	O
methods	O
for	O
machine	B-Application
intelligence	I-Application
.	O
</s>
<s>
This	O
theory	O
proposes	O
that	O
cortical	B-General_Concept
columns	I-General_Concept
at	O
every	O
level	O
of	O
the	O
hierarchy	O
can	O
learn	O
complete	O
models	O
of	O
objects	O
over	O
time	O
and	O
that	O
features	O
are	O
learned	O
at	O
specific	O
locations	O
on	O
the	O
objects	O
.	O
</s>
<s>
A	O
single	O
HTM	O
node	O
may	O
represent	O
a	O
group	O
of	O
cortical	B-General_Concept
columns	I-General_Concept
within	O
a	O
certain	O
region	O
.	O
</s>
<s>
HTM	O
nodes	O
attempt	O
to	O
model	O
a	O
portion	O
of	O
cortical	B-General_Concept
columns	I-General_Concept
(	O
80	O
to	O
100	O
neurons	O
)	O
with	O
approximately	O
20	O
HTM	O
"	O
cells	O
"	O
per	O
column	O
.	O
</s>
<s>
Integrating	O
memory	O
component	O
with	O
neural	B-Architecture
networks	I-Architecture
has	O
a	O
long	O
history	O
dating	O
back	O
to	O
early	O
research	O
in	O
distributed	O
representations	O
and	O
self-organizing	B-Algorithm
maps	I-Algorithm
.	O
</s>
<s>
For	O
example	O
,	O
in	O
sparse	B-Architecture
distributed	I-Architecture
memory	I-Architecture
(	O
SDM	B-Architecture
)	O
,	O
the	O
patterns	O
encoded	O
by	O
neural	B-Architecture
networks	I-Architecture
are	O
used	O
as	O
memory	O
addresses	O
for	O
content-addressable	B-Data_Structure
memory	I-Data_Structure
,	O
with	O
"	O
neurons	O
"	O
essentially	O
serving	O
as	O
address	O
encoders	O
and	O
decoders	O
.	O
</s>
<s>
Similar	O
to	O
SDM	B-Architecture
developed	O
by	O
NASA	O
in	O
the	O
80s	O
and	O
vector	O
space	O
models	O
used	O
in	O
Latent	O
semantic	O
analysis	O
,	O
HTM	O
uses	O
sparse	O
distributed	O
representations	O
.	O
</s>
<s>
The	O
semantic	B-General_Concept
folding	I-General_Concept
theory	O
builds	O
on	O
these	O
SDR	O
properties	O
to	O
propose	O
a	O
new	O
model	O
for	O
language	O
semantics	O
,	O
where	O
words	O
are	O
encoded	O
into	O
word-SDRs	O
and	O
the	O
similarity	O
between	O
terms	O
,	O
sentences	O
,	O
and	O
texts	O
can	O
be	O
calculated	O
with	O
simple	O
distance	O
measures	O
.	O
</s>
<s>
Likened	O
to	O
a	O
Bayesian	O
network	O
,	O
an	O
HTM	O
comprises	O
a	O
collection	O
of	O
nodes	O
that	O
are	O
arranged	O
in	O
a	O
tree-shaped	O
hierarchy	O
.	O
</s>
<s>
A	O
Bayesian	O
belief	O
revision	O
algorithm	O
is	O
used	O
to	O
propagate	O
feed-forward	O
and	O
feedback	O
beliefs	O
from	O
child	O
to	O
parent	B-Application
nodes	I-Application
and	O
vice	O
versa	O
.	O
</s>
<s>
Like	O
any	O
system	O
that	O
models	O
details	O
of	O
the	O
neocortex	O
,	O
HTM	O
can	O
be	O
viewed	O
as	O
an	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
.	O
</s>
<s>
The	O
tree-shaped	O
hierarchy	O
commonly	O
used	O
in	O
HTMs	O
resembles	O
the	O
usual	O
topology	O
of	O
traditional	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
HTMs	O
attempt	O
to	O
model	O
cortical	B-General_Concept
columns	I-General_Concept
(	O
80	O
to	O
100	O
neurons	O
)	O
and	O
their	O
interactions	O
with	O
fewer	O
HTM	O
"	O
neurons	O
"	O
.	O
</s>
<s>
For	O
example	O
,	O
feedback	O
from	O
higher	O
levels	O
and	O
motor	O
control	O
is	O
not	O
attempted	O
because	O
it	O
is	O
not	O
yet	O
understood	O
how	O
to	O
incorporate	O
them	O
and	O
binary	O
instead	O
of	O
variable	O
synapses	B-Application
are	O
used	O
because	O
they	O
were	O
determined	O
to	O
be	O
sufficient	O
in	O
the	O
current	O
HTM	O
capabilities	O
.	O
</s>
<s>
LAMINART	O
and	O
similar	O
neural	B-Architecture
networks	I-Architecture
researched	O
by	O
Stephen	O
Grossberg	O
attempt	O
to	O
model	O
both	O
the	O
infrastructure	O
of	O
the	O
cortex	O
and	O
the	O
behavior	O
of	O
neurons	O
in	O
a	O
temporal	O
framework	O
to	O
explain	O
neurophysiological	O
and	O
psychophysical	O
data	O
.	O
</s>
<s>
Similarities	O
of	O
HTM	O
to	O
various	O
AI	B-Application
ideas	O
are	O
described	O
in	O
the	O
December	O
2005	O
issue	O
of	O
the	B-Application
Artificial	I-Application
Intelligence	I-Application
journal	O
.	O
</s>
<s>
Neocognitron	B-Algorithm
,	O
a	O
hierarchical	O
multilayered	O
neural	B-Architecture
network	I-Architecture
proposed	O
by	O
Professor	O
Kunihiko	O
Fukushima	O
in	O
1987	O
,	O
is	O
one	O
of	O
the	O
first	O
Deep	B-Algorithm
Learning	I-Algorithm
Neural	B-Architecture
Networks	I-Architecture
models	O
.	O
</s>
