►
From YouTube: HTM Hackers' Hangout - Dec 7, 2018
Description
Discussion at https://discourse.numenta.org/t/htm-hackers-hangout-dec-7/5031.
HTM Hackers’ Hangout is a live monthly Google Hangout held for our online community. Anyone is free to join in the discussion either by connecting directly to the hangout or commenting on the YouTube video during the live stream.
More info on all these topics at http://numenta.org.
A
A
So
there
is
a
link
to
you
know
the
forum
thread
where
I,
usually
post
discussions
and
the
things
we're
going
to
talk
about
here.
So
I
was
going
to
talk
about
a
few
things
that
I've
been
working
on
and
then
just
open
it
up
to
committee.
Questions
I
was
hoping
somebody
from
researcher
poppet
and
we'd
have
new
intern
and
I
wanted
to
introduce,
but
I
haven't
see
him
yet,
but
when
he
comes
online,
if
and
when
I'll
introduce
them,
and
so
so
there
are
a
couple
things
that
I've
been
working
on.
A
As
you
know,
I'm
not
one
of
the
neuroscientists
on
the
research
team,
I'm,
the
community
manager
and
I
create
a
lot
of
educational
material,
so
I'm
working
on
educational
material
to
back
up
the
two
papers
that
we
just
released
the
frameworks
paper
and
what
we
call
the
columns
plus
paper
so
I'm
working
again
on
the
frameworks
paper.
There's
another
HTM
video
I'm
trying
to
come
up
with
this
visualization
of
the
thousand
brains.
A
Theory
that
really
sort
of
shows
the
difference
between
the
classic
hierarchy
idea
and
the
thousand
brains,
hierarchy
idea
and
some
visualization
that
really
emphasizes
the
lateral
connections
and
the
importance
of
the
lateral
connections.
I'm
gonna.
Do
it.
I've
been
working
on
this
for
a
long
time
and
as
a
part
of
that
I
really
don't
understand.
Sensory
projections
in
visual
and
somatic
cortex,
so
in
the
link
down
below
you'll,
see
that
I
dove
into
this
idea
of
quilting
in
the
cortex.
A
The
term
quilting
is
not
a
neuroscience
term
that
something
that
I
or
mark,
or
one
of
us
came
up
with,
and
we
were
talking
about
this.
But
the
interesting
thing
is
that
you
know
there's
there's
all
of
these
projections
of
from
the
sensory
field
on
to
cortical
substrate,
you
can
see
it
in
v1,
v2,
v3,
there's
a
projection
of
the
field
of
view
and
an
s1
and
s2
there's
a
projection
of
the
entire.
A
You
know
body
space
on
to
cortex,
but,
as
you
go
up
the
heart,
you
know
they
get
weird
and
they
get
all
mixed
together
and
that
doesn't
that
there's
they're
not
like
total,
continuous
and
then
there's
these
other
types
of
projections
that
are
more
having
to
do
with
how
different
features
that
are
being
extracted
from
senses
are
projected
onto
the
same
topological
substrate.
This
is
like
the
tiling
or
not
tiling.
They
call
like
the
striping
you
know
of
like
ocular
dominance,
binocular
dominance
or
between
different
types
of
cells
and
somatic
input.
Anyway.
A
These
are
there's
a
lot
of
different
things
going
on
here.
So
I
really
was
an
educational
experience
for
me
because
I
didn't
know
what
was
what
so
that
whole
thread
is
basically
meet
at
the
path
of
self-discovery
for
me,
trying
to
figure
out
what's
going
on
here,
and
so,
if
you
want
to
follow
along
with
that,
I
didn't
find
anything
that
like
lit
up
any
light
bulbs
and
the
research
team
or
anything
but
I
I
got
things
better
in
my
head.
That
which
was
the
whole
point.
A
A
There's
that
and
the
other
thing
that
I've
been
working
on
recently
is
working
with
Mark
Brown
or
bit
King
on
the
forum,
because
I
really
mark
rounds
been
talking
about
his
idea
of
hexagonal
grids
and
how
they
map
to
many
columns
for
a
long
time
now
and
I've,
never
quite
understood
it.
So
I
I
was
like
I'm
gonna
dive
in
and
understand
this,
so
mark
and
I
went
back
and
forth
for
a
long
time
and
he's
been
really
patient
with
me
really
helpful
for
me
to
understand
this
idea.
A
I
mean
it
goes
back
to
its
96
book,
the
Cerebral
code
by
William
Calvin,
and
how
some
of
the
ideas
in
that
about
emergent
hexagonal
grids
based
on
interactions
between
pyramidal
neurons,
might
be
at
play
with
an
HTM
or
might
overlap
with
HTM
and
specifically
I
mean
there's,
there's
some
interesting
things
in
this
in
this
thread
that
mark
and
I
expose.
So
we
started
chatting
back
and
forth
and
private
messages,
and
eventually
we
got
to
the
point
of
let's
just
make
this
public,
so
we
cleaned
it
up
and
made
it
public.
A
So
it's
Oh
me
and
him
going
back
and
forth
like
how
do
hit
these
ideas
of
hexagonal
grids
and
HTM
like
play
together.
So
there's
a
ball
of
discussion
about
that.
The
most
interesting
thing,
I
think
is
like
that.
Mark
explained
to
me
that
that
I
liked
was,
if
you
have
an
SDR,
if
you
like,
a
mini
column
activation
and
some
big
cortical
substrate
ignore
the
cortical
columns.
Just
say
you
have
a
big
population
of
many
columns
and
cells
within
them
and
you
have
a
mini
column
activation,
that's
driven
by
feed-forward
sensory
input.
A
If
you
look
at
all
of
those
mini
columns,
that's
an
SDR
and
based
on
the
geography
and
the
continuous
tracker
model,
and
you
know
the
idea
that
a
hex
grid
can
emerge
based
upon
sensory
input
like
you
can
look
at
that
SDR
and
you
can
go
like
search
around
for
these
hex
grids.
So
there's
there's
ways
that
you
can
find
these
hex
grids
and
help
them
emerge
and
help
them
learn
interesting
idea
to
me
I
to
Jeff
and
Marcus
about
it
both
but
they're.
A
There
they're
really
focused
on
on
applying
the
ideas
about
the
grittiness
and
the
grid
cells
stuff
at
much
lower
levels
within
layers
of
columns
and
the
interactions
between
columns.
So
they're
not
interested
looking
at
that
right
now,
although
I
do
think
it's
an
interesting
idea
and
tons
of
credit
to
Mark
Brown
for
keep
continuing
to
push
on
that
because
of
I
love,
interesting
ideas.
So
thanks
to
him
for
that
I,
don't
see
anybody
from
the
research
team
online
right
now
so
I
guess
we
won't
have
research
update,
but
we
do
have
a
new
intern.
A
A
You
know
setups
EKGs
types
of
ups
and
make
it
really
easy
to
hack
into
him,
and-
and
so
it's
interesting
stuff,
but
I
I
do
not
know
how
his
gym
would
work
with
that
type
of
thing
mark
says:
marks
on
showered
now
he
can't
join,
but
he
says
he's
working
up
opposed
to
show
more
stuff
on
HTM
and
X
crits
in
the
visual
thing.
That's
too
too
detailed
to
bring
up
here.
A
Joshua
is
asking
again
about
BCI
stuff:
does
the
signal
from
nupoc
layers
mimic
what
the
brain
is
actually
doing
or
is
it
abstracted
it's
abstracted
in
some
ways?
For
example,
we
don't
model
inter
neurons
directly.
We
do
not
create
like
instances
of
inter
neurons,
that
are
associated
with
many
columns.
We
model
the
effects
of
inter
neurons.
We
try.
A
We
think
we
understand
what
the
inter
neurons
are
doing
and
to
form
many
columns,
and
so
we
we
model
the
the
effects
of
them
the
same
thing
and
in
all
of
these
grid
cells,
stuff,
there's
and
they
don't
really
know.
There's
a
lot.
There's
still
debate
about
this,
how
our
grid
cells
formed
it's
very
likely
that,
inter
neurons,
are
involved
in
that
it
seems.
A
There's
some
reports
that
say
no,
it's
not
possible
for
too
small,
but
anyway
the
grids
are
too
small,
not
the
inter
neurons
but
I
think
the
jury's
sort
of
still
out
on
it,
but
but
there's
probably
maybe
inter
neurons,
that
work
within
grid
cells,
but
we
don't
have
to
model.
You
know
if
you
understand
what
the
inter
neurons
are
doing
and
out
and
some
systems
I,
don't
think
you
have
to
model
them
directly.
A
How
do
you
think
H
dam
can
be
modified
to
work
with
back,
cooperate,
propagation
algorithm,
so
back
propagation
is
sort
of
a
way
to
attempt
to
do
like
credit
and
assignment
and
then
so.
Let
me
point
you
to
a
few
things:
I,
don't
know.
I
don't
know,
but
like
our
position
here
in
the
meta
is
a
background,
but
a
back
props
not
going
on
in
the
brain,
and
we
don't
need
it
for
the
HTM
model.
There's
other
ways.
It's
brain
stirring
credit
assignment.
I've
talked
recently
to
blake
richards
about
this
is
a
theoretical
neuroscience.
A
A
look
at
our
podcast
use.
Last
podcast
that
we
did.
We
talked
a
bit
about
this
back
propagation
and
how
it
relates
to
the
brain,
and
where
does
the
brain
could
do
back
propagation?
So
he
has
some
ideas
about
that
sort
of
credit
assignment
being
done
through
apical
dendrites,
which
is
totally
in
line
with
the
HTM
world.
So
I
think
if
there
is
credit
assignment
being
doing
it's,
probably
it's
probably
being
done
in
that
way.
You
might
want
to
read
Mark
Brown
ping
ping
yuan.
This
probably
but
mark.
A
You
might
want
to
read
that
three
visual
streams
paper,
because
that
is
along
the
same
logic,
sort
of
shows
a
profit
proposition
for
how
backpropagation
could
might
be
done
in
the
digital
cortex.
With
these
three
streams
and
the
thalamus
is
blackboard,
so
you
might
want
to
look
into
that,
but
the
mental
position
is
backdrops
are
not
going
on
in
the
brain.
We're
not
trying
to
model
it.
A
B
B
A
Actually,
looking
at
this
right
now,
because
this
is
it's
in
the
columns
paper-
not
the
columns
plus
paper
at
the
original
columns
paper
and
in
the
it
cells,
it
says
it
says
exactly
how
to
do
it.
What
but
what
they
do
is
they
take
some
portion
of
those
distal
inputs
and
they
they
get
them
from
another
column.
A
So,
like
their
output,
is
now
their
input,
and
so
they
first
will
come
up
with
what
what
many
columns
are
active
based
on
the
feed-forward
input
and
then
they'll
use
the
other
input
as
a
bias,
but
the
but
they'll
they'll
be
predicting
basically
in
the
same
space,
it's
in
the
columns
paper.
I
can
I'm
actually
reading
it
right
now,
so
I
don't
have
it
like
fully
formed
in
my
head.
So
I
can't
talk
it
back
to
you.
But
if
you
look
for
our
columns,
why
does
the
cortex
have
layers
and
columns
I?
A
Someone
else
is
online.
Is
this
where
you
hello
bill
nice
to
see
you
Phil,
you
have
anything,
you
know
pipe
up,
otherwise,
I
don't
see
any
other
QA
coming
from
chat
and
there's
like
18
people
watching
so
hello.
Everyone
thanks
for
watching
I,
think
we're
wrapping
up
unless
we
have
any
more
questions
coming
through
LLL
wait
for
a
few
seconds,
Oh.
A
Phil
says
he
doesn't
have
a
microphone
no
problem
I
enjoy
having
them
it's
a
way
for
us
to
update.
You
know
guys.
What's
going
on
so
it's
exciting
times,
you
know
there's
a
lot
of
stuff
going
on
right
now
in
AI
and
I,
don't
know
if
you
guys
heard
the
my
last
podcast
interview
with
Blake
Richards
I
thought
it
was
really
good.
I
said
a
couple
things
that
were
technically
incorrect.
They
were
just
misspoken,
but
he
Blake's
said
all
these
things.
A
That
I'm,
like
yes,
I,
totally
totally
agree
with
all
these
things
that
he
said
it's
so
great
to
hear
someone
else
working
in
the
field
of
theoretical
neuroscience,
coming
from
the
field
of
neuroscience,
that's
so
well
aligned
with
the
with
sort
of
the
picture
we're
putting
together
of
the
it's
very
refreshing,
so
I'd
recommend
taking
a
listen
to
our
last
podcast
episode.
If
your,
if
you
haven't
been
there
part
of
our
community,
otherwise
everybody
have
a
wonderful
holiday
season.
I
will
not
be
seeing
you
guys
until
January.