►
From YouTube: Deeper into the Neuroscience (Part 2)
Description
We go even deeper into the neuroscience behind new theories of sensorimotor inference in the cortex and HTM.
You should have some background in HTM before understanding these videos. See http://numenta.org/htm-school/ for more videos explaining basic HTM theory.
Music: "Holy Roller" by YACHT
(used with permission from Free Music Archive)
A
Now,
let's
go
back
you
a
little
bit
more
detail:
we
can
yeah.
So
let's
start
with
your
slide,
this
is
basically
just
a
PowerPoint
picture.
What
I
was
just
drawing
on
the
whiteboard
here,
and
so
you
here
you
see,
we
have
this
piece
poet
input
this
after
I've
been
puttin
century.
Please
go
into
later
for
that's
what
terminates
its
classic
neuroscience
late
for
this
in
the
green
arrows
letter
for
projection
layer,
two
three
neat
art
and
they
are.
A
A
This
is
nothing
special
about
this
at
all,
also
layer,
two
three
projection
along
distances:
we'll
talk
about
that
moment
and
rejects
back
to
Leo
I'm,
not
showing
here
the
temple
memory
layer
again,
number
I
said
earlier:
I'm
just
going
to
talk
about
the
new
model,
we're
not
going
to
show
the
temple
ground
so
we're
just
showing
you
later
for,
and
we
see
you
later,
three
we're
kind
of
ignoring
lower
leg,
sweep
3b
3b,
okay.
So
that's
the
basic
model
is
a
two
layer
in
for
the
infant
layer.
It
is
an
output
layer.
A
This
is
well
established,
neuroscience.
It
turns
out
just
to
get
between
a
little
more
complicated
I'm,
a
bargain
at
slated
this
same
motif.
If
you
will
appears
twice
in
every
region,
the
same
input
that
goes
to
lay
or
goes
to
layer,
six,
a
it's
one
of
the
same
types
of
representations,
spaceship
or
many
columns-
tell
us
stuff
that
projection
to
layer.
Five
as
a
five
is
an
output
layer
and,
and
they
five
project
long
distances
and
they
project
back.
So
this
this
two
layers,
secret,
we've
come
up
with.
A
A
Some
big
clues,
but
we
don't
know
right.
We
got
some
hypotheses.
We
have
some
a
lot
of
evidence
to
sift
through
it's
not
like.
Oh
my
gosh,
we
have
no.
I
yeah
it's
just
like
well,
this
possible
possible.
So
if
you
ask
you
to
lend
look
of
just
sort
of
slide
here,
you
see
that
the
states
repeated
twice
now
with
the
new
thing
which
we
were
just
talking
about.
This
is
orange
line.
The.
A
B
A
Layer
6a
and
that's
our
go
thinning
down
and
then
later
a
becomes
the
input
to
layer
four,
so
the
Alex
actual
outside
you
can
put
the
layer
for
is
coming
to
my
success.
Oh
my
goodness.
What's
going
on
and
we're
sir,
we
haven't
doing
that.
Talk
about
that
I'm,
giving
you
circuitry
and
plumbing
yeah
okay.
These
are
these
are
physical
facts,
okay,
and
what
we've
said
is
that
the
distal
inputs
on
the
input
layer
are
going
to
be
allocentric,
location
right
and.
A
Fella
yeah,
you
have
to
be
careful.
We
understand
a
lot
that
something
to
done.
I
well.
I
wouldn't
normally
want
the
circuitry
understand
this
thule
circuit
really.
Well,
we
don't
mention
it
occupies
twice.
We
don't
stands
back
all
the
transformations
on,
so
I'm
just
giving
you
the
facts
and
then
we're
going
to
come
back
come
on.
A
A
A
There
and
now
two
separate
regions.
Imagine
we
think
about
a
hierarchy
of
you
know
the
region,
region
region.
The
hierarchy
was
actually
two
parallel
hierarchies
region,
region,
region
and
one's
called.
This
is
first
discovered
a
vision,
so
envision
is
the
ventral
and
dorsal
stream,
but
the
term
the
term
one
where
is
sort
of
likes
about
it.
So
this
entry
in
these
two
parallel
vision,
streams
and
there's
two
parallel
auditory
streams
is
to
parallel
somatosensory
a
touchscreen.
A
What
and
where?
Okay
and
let's
talk
about
vision,
first,
okay,
because
indigenous
wrists
first
discovered
they
should
want
to
look
at
this-
is
two
separate
visual
systems
going
on.
What's
the
difference
between
well,
the
reason
they
say,
one
is
called
what
is
because
that
seems
to
represent
objects
in
the
world,
the
what
is
out
there
and
the.
A
To
sort
of
be
something
they
see
say
so
represents:
where
is
right,
so
it
turns
out,
one
is
an
egocentric
and
ones
an
allosteric
representation.
Okay.
So
let's
take
the
allocentric
representation,
which
would
be
classic
v1
and
v2,
and
before
is
what
this
is,
the
one.
If
I
have
I'm
looking
at
an
object
and
I
take
and
I
say:
okay
I'm
going
to
damage
the
left,
half
way,
I'm
going
to
damage
the
two
or
something
like
that.
So
what
do
I
know
as
a
person?
I
would
say:
oh
I
see
there's
something
there.
B
A
If
you
just
have
no
idea
what
it
is,
it's
like
it's
something
I
can't
tell
you
what
it
is
that
announced
that,
can
you
use
it
as
your
country?
Well,
that's!
Well,
it's
not
to
get
there
because
I.
If
my
image
is
my
vision
right
now,
okay,
but
I
can
say
well,
I,
don't
know
what
there
but
I
can't
reach
it
and
I
grab
it
and
you
can
do
it
really
well,
so
it's
like
I
have
to
provide
it
if
you
damage
the
other
side,
the
where
pathway
but
then
I.
A
Look
as
though
there's
a
coffee
cup
I
know
there's
a
pen
and
this
whatever
does
that,
but
I
can't
reach
it.
I
said
I,
don't
I,
don't
know
how
to
get
to
it.
I'm
looking
at
it
I
see
it.
I
know
what
it
is,
but
I
can't
actually
get
to
it
an
active
on
again
because
the
egocentric
representations,
blah
I,
don't
know
it
is
relative
to
me,
but
I
can
know
what
it
is
and
the
other
times
I
know
way
to
drop
me,
but
I
can't
see
what
it
is
right.
B
A
A
classic
v1
or
as
one
of
these
are
the
primary
sensory,
cortex
or
vision
in
touch
and
we're
saying
count
of
that
work,
and,
if
we're
talking
about
that,
that's
a
what
pathway
right
I
need
to
get
an
a
licentious
location.
Now,
where
is
the
allostatic
location
coming
from
right
all
right?
Let's
go
back
and
look
at
this
slide
here.
Okay,
this
is
some
more
Anatomy.
That's
basically
known
little
sketchy
untouchable
literature,
but
there
are
because
we
have
these
two
parallel
pathways.
A
What
we
think
is
going
on
is
any
signals
exist.
We
think
that
there's
a
communication
between
the
what
and
the
where
it's
basically
providing
this
information
between
these
two
hundred,
which
a
movement
that
hierarchy
right.
So
that's
our
somatic
sensory
there's
a
motor,
cortex
m1
and
if
it's
more
complicated
this,
but
the
sages
m1s
wants
or
the
motor
cortex
in
the
somatosensory
cortex
there
are
there
like
where
and
what
and
they
connect
to
each
other
now.
A
Why
would
you
want
to
do
this
because,
when,
if
I'm
about
to
move
relative
to
something
let's
say,
I,
have
my
coffee
cup
I'm
going
to
move
my
hand
and
my
finger
relative
to
it?
Well,
I'm
just
going
to
move
my
hand
right
here,
I'm
just
going
to
move
my
finger
I'm
going
to
move
my
finger
down.
I
want
to
be
able
to
predict
we're
going
to
feel
yeah
so.
A
A
B
A
That
the
cortex
infuses
it
everywhere
yeah
it's
everywhere-
is
these
conversions
between
reference
strings
going
on
all
the
time.
Senses-
allergens,
it's
everywhere
now.
So
in
this
joint,
maybe
it's
worth
pointing
out!
I
show!
This
is
more
of
a
conceptual
diagram
of
these
took
these
basic
circuitry
except
the
Alpha
bond
was
saying
hey
when
the
body
moves
in
generates
behavior
m1.
That
has
to
generate
a
signal
which
gets
across,
though,
which
of
these
distant
sensory
cortex
and
becomes
an
ala,
centric
location
right.
A
A
My
body
and
once
I
do
all
that
I
can
make
a
prediction
right,
but
this
has
to
happen
because
you
make
pictures
all
the
time
so
there's
a
process
here
and
how
this
get
converted
from
the
behavior
into
an
ALICE
nitrification.
We
have
some
theories
about
that.
We
have
two
models
that
we're
not
exactly
something
exactly
where
it's
occurring
and
all
the
different
layers.
So
we're
I'm
not
going
to
go
deeper
to
that.
Okay,
but
we
have
to
occur
yeah.
The
other
thing
is
they
have
to
go
backwards.
A
A
B
A
Tend
to
depend
orientation
or
education
right,
so
I
can
say:
I
love
a
movement
I
want
to
make
relative
to
the
pet
enough.
We
should
do
this
all
the
time
right
beyond
another
yeah,
so
I
say
I'm,
a
woman
I
want
to
create
in
allocentric
ordinance
I
have
to
send
it
back
to
the
motor
cortex.
You
can
burn
it
into
an
egocentric
look.
So
as
you
can
sososo,
this
communication
has
to
go
both
ways
right.
A
B
A
A
A
I
know
that
the
more
you
think
about
this
going
to
come
to
realize
this,
that
everything
you
perceive
in
the
world
is
its
own
location
in
space,
and
therefore
everything
has
to
have
this
out
of
such
a
rotation
of
social
yeah.
It's
a
pretty
I
mean
in
some
sense
would
be
some
time
to
say
out
the
obvious
any
kind
of
in
August
you
can
produce
it,
but
on
the
other
hand,
it
is
a
detailed
model
that
this
is
occurring
in
all
century
regions.
A
The
cortex
is
novel,
and
so,
as
we
know,
nobody
even
thinks
aquatic
society
like
be
watching
doesn't
need
any
clothes
of
the
certificate,
because
when
we
can
now
show
that
even
a
primary
sensory
region,
epsilon
V
one
can
learn
the
entire
structure
of
3d
objects
and
most
people
think
it's
just
extracting
some
feature
tapping
on
someone
else,
though
it's
far
more
powerful
in
easily
anyway.
So
this
is
the
drawing
we've
got.
You
guys
need
a
one
there
where
we
in
a
what
hierarchy
and
what
are
hierarchies
and
ones
working
analysis.
Take
one
burgundy.
A
B
A
A
B
A
A
A
With
each
other,
in
the
first
layer
in
higher
the
juice
too
far,
parties
too
much
information
to
pin
bottle
that
yeah,
okay.
Well,
it's
no
different
than
saying
parts
of
the
retina
and
parts
of
my
skin
are
too
far
apart,
and
so
there
is
a
convergent.
As
you
go
up
the
hierarchy,
there's
a
convergence
that
occurs
cells.
A
A
The
coordinate
is,
it
know,
a
predicting
understand:
it's
natural
we're
getting
a
little
far
field
there.
I,
don't
believe
there
is
any
obvious
if
you
can't
advise
them
up
willy-nilly
sure
there
is
a
convergence
or
occurrence,
but
there
is
nothing
that
says:
okay,
I'm,
processing,
vision
and
auditory.
A
B
A
Everyone
else
up
right,
yeah,
there's
evidence
by
the
way
is
we
now
discover
them
into
something
called
border
border
ownership
which
scientists
have
found,
which
is
showing
exactly
what
I'm
saying
that
the
even
being
one
can
recognize
complex
objects
that
are
bigger
than
you
think
that
they
can
our
without
hierarchy
in
a
layer
within
a
region
a
raise
or
a
level
of
the
hierarchy
show.
So
the
way
to
think
about
this
is
above
you've
got.
A
Some
you've
got
some
large
layer
of
some
impressive
sheets
of
destroying
over
one
dimensional,
two
dimensional,
sir,
and
it's
getting
this
input
of
something
we
don't
really.
It
doesn't
really
care
with
input.
Coming
from
it's
going
to
able
it's
able
to
within
some
region
of
this
thing,
it's
going
to
be
able
to
model
3d
objects.
That
is
some
part
of
the
inference.
If.
A
Well,
you
guys
I,
can't
I
have
to
have
another
level
in
the
hierarchy
to
do
that
right.
So,
but
the
mechanism
is
occurring,
the
same
exact
questions
was
occurring
everywhere,
that
a
single
region
can
learn
the
three-dimensional
structure
of
an
object
given
somewhat
extensive
input
space.
It
doesn't
matter
that
vision
to
touch
whatever
it
is,
makes
no
difference,
and
but.
A
A
A
A
The
word
I
mean
just
reckon
worried
about
much
making
of
it.
So,
but
this
is
safe
to
say,
recognize
word
around:
let's,
let's
go
there.
There
is
no
there's
a
lot
of
place
in
businesses,
oh
the
meaning
of
something
they
don't
have
to
so
we're
going
to
do
it
far
field.
Here
we
haven't
even
talked
about
what
we'd
actually
accomplished
in
this
materia,
which
were
giving
some
background.
A
A
The
whole
model
is
relying
on
the
fact
that
I
have
a
hollow,
centric,
location
and
and
I
prefer
fishing
house
that
you
can
equal
centric
and
escrow,
not
true
the
west
where,
but
in
the
end,
we
have
a
two
layer
model
which
does
how
this
works,
and
we
should
talk
about
that
paper.
But
because
that's
the
message
that
thing
we
discussed.