►
From YouTube: HTM Hackers' Hangout - Feb 2, 2018
Description
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.
If you have something specific you’d like to discuss, or if you just want to learn more about the HTM Community, please join HTM Forum at https://discourse.numenta.org. We have active discussions about HTM theory, research, implementations, and applications.
More info on all these topics at http://numenta.org
This month, we’ll talk about:
Grid cells
community fork
python 3
siraj raval
research update (?)
A
Hey
everyone
welcome
to
HTM
hackers,
hangout,
it's
February,
2nd,
it's
the
first
hackers
hang
out
at
2018.
Sorry
I
skipped
last
month,
but
I
totally
forgot
about
it.
Honestly,
it
was
busy
so
we're
going
to
get
a
lot
upon
line
here.
So
let
me
run
through
a
few
things:
I
want
to
go
through
and
then
we'll
open
up
discussion.
A
I
want
to
give
the
guys
who's
working
on
unity
for
like
a
chance
to
talk
to
each
other
to
at
the
end
of
this,
and
that's
actually
the
first
topic
that
I'm
gonna
steal
the
screen
and
talk
about
this
community
forum
of
you.
Don't
know
what
a
community
for
is
even
about
a
forum
created
the
topic
about
it
somewhere.
Other
is
what
is
it
important,
but
it's
basically
a
community
driven
for
the
codebase,
a
boopa
can
take
it.
A
What
you
like
with
it,
without
any
restrictions
from
us,
we
need
to
review
all
this
stuff
really
carefully.
I
knew
this
was
eventually
gonna
happen,
so
I'm
glad
that
it's
happening
everybody's
working
on
it
there's
some
initiative,
specifically
behind
the
community.
For
that
that
started,
it
I
think
the
important
ones
are
like
Python
3,
that
people
want
Visual,
Studio
and
c-sharp
bindings
for
something
to
wanted.
So
these
are
all
things
that
you
mentor.
The
company
doesn't
need
or
want.
A
So
we
put
a
bunch
of
resources
into
at
the
moment,
but
the
community
of
certainly
welcome
to
do
that
and
there's
potential
that
this
one
could
be
pulled
back
into
other
public
days
of
the
future,
but
we
decided
we
would
want
to
see
that
speech,
so
it
could
help
us
in
the
future
as
well.
I'm
totally
happy
that
I
want
to
help
you
guys.
However
I
can
sway.
My
community
helped
some
of
the
people
that
are
involved
in
this
are
right
now
online
Christian,
here
on
the
left
and
David
Eenie.
A
Guys
talk
a
little
bit
at
the
end
of
this
and
also
mark
Omaha's
of
longtime
community
member,
a
committer
new
pic
he's
gotten
involved
in
it,
which
is
great
okay
over
on
in
our
HTML
diecut
hub
organization.
This
is
where
at
least
two
of
them
two
of
these
Forks
are.
These
are
the
ones
that
mark
created,
good,
you
could
see
could
be.
You
could
buy.
A
And
discussions
going
on
he's
not
happy
to
the
names
that
he
end
up
with,
but
I
also
think
Christian.
You
had
some
points
to
a
fork
at
least,
and
someone
else
might
have
so
I
think
there's
still
some
work
that
needs
to
get
sort
of
incorporated
into
this
and
it
hasn't
the
chips
haven't
all
fallen.
So
I
know
that
there's
still
some
work
to
do
there,
but
oh
I,
know
I'll.
A
Ask
you
guys
about
that
at
the
end
of
this
and
I,
like
you,
so
I
wanted
to
mention
the
pork
I
think
that's
pretty
cool
I
got
there's
a
whole
community
fork
topic
on
on
HTM
for
on
here
there
underneath
it.
You
can
see
here.
So
it's
I
created
that
out
and
the
next
thing
I
wanted
to
talk
about
was
Suraj
from
all
who.
B
A
Holy
cow,
look
at
my
analytics.
It
was
a
huge
bomb
I'm
just
going
to
show
you.
It
really
helped
that
there's
the
suraj
effect
right
there.
So
and
even
right
now
we're
at
like
two
and
a
half
times
my
normal
YouTube
views.
So
it's
having
a
lasting
effect
on
YouTube
and
on
their
web
traffic,
which
is
been
great.
The
funny
thing
is
I
did
a
little
bit
of
research
like
I
said:
I've
tried
to
get
him
to
do
videos
for
a
while,
but
he
never
spotted
today
back
in
October
yeah
October
last
year.
A
He
did
this
video
about
the
future
of
deep
learning,
research
and
apparently
Suraj
has
said
negative
things
about
unitary
TM
in
the
past,
and
some
of
these
videos,
I,
never
saw
saw
it,
but
I've
seen
people
commenting
about
that
on
his
videos
and
I'm,
not
gonna.
Go
watch
like
hours
of
his
videos
to
just
go
find
out
what
he
said,
but
in
this
talk
that
he
talking
he's
talking
about
the
future
of
deep
learning.
This
is,
after
him
came
out
and
said
that
propagation
may
be
not
the
right
way
to
go.
A
So
he
did
this
video
and
talked
about
all
the
potential
other
things
that
that
machine
learning
community
might
work
towards,
and
he
didn't
mention
each
team.
And
if
you
look
through
the
comments
which
I
did
there's
several
comments,
I
don't
know
5
7,
something
like
that
of
people
saying
what
about
Numenta?
What
about
HTM?
You
know
you're
forgetting
about
this
big
thing.
Then
he
responds
and
admits.
I
think
I
should
learn
more
about
that
and
people
asked
him.
Why
didn't
you
or
why?
Why?
Why
are
you
making
a
joke
out
of
HTM?
A
So
I
guess
at
some
point
he
did,
but
he
has
changed
his
mind
and
in
this
video
called
momentum
explained,
he
was
really
very
positive
about
the
meta,
which
is
good.
You
should
watch
the
video,
it's
not
really
the
great.
It's
not
really
accurate.
I
mean
he
says
some
things
that
are
just
wrong.
He
totally
he
doesn't
understand
the
technology
enough
I.
A
A
You
can
think
of
a
grid
cell
module
as
an
array
of
neurons
and
each
neuron
has
some
projection
to
a
space
and
it
fires
at
certain
locations
in
the
space.
And
so,
if
you
have
lots
of
these
grid
cell
modules
sort
of
projected
onto
the
same
space,
you
can
get
an
accurate
idea
of
location,
a
representation
of
location,
but
that's
fake,
so
and
like
in
an
and
str
format,
just
by
concatenating.
These
been
cell
modules,
cells
and
they're
they're
active
cells
together.
A
B
A
And
our
pasta
nation
that
they
also
exist
in
the
neocortex
and
because
it's
really
crucial
to
our
current
theory
that
we
have
these
grids
film-like
properties
in
the
neocortex.
The
helpless
of
this
object,
location
and
object.
Recognition,
which
brings
me
to
the
next
topic
before
we
start
talking
about
community
fork
and
open
up
the
conversation
to
anybody
else
is
I
would
like
Marcus
that
you
up
for
it
to.
B
A
Update
on,
what's
going
on
in
research,
yeah
sure,
well,
yeah
I
mean
you
kind
of
brought
up
a
topic
that
everyone
here
is
talking
about
a
lot
and
that's
grid
cells.
I
guess
at
the
core
were
sort
of
treating
grid
cells
as
the
brains
coordinate
system,
but
the
way
that
bringing
handles
locations
the
way
it
relates
locations
to
each
other.
So
at
the
core
of
what
we
need
is
locations
and
grid
cells
just
seemed
to
be
a
biologist
way
of
doing
that.
A
We
don't
know
for
sure
if
cortex
has
like
literally
their
themselves
like
if
it's
the
same
exact
mechanism
or
is
something
more
analogous
to
it.
Maybe
it's
3d
rather
than
2d,
there's
just
a
lot
too.
There
is
a
there
are
a
lot
of
unknowns
there,
but
in
general
we're
kind
of
using
it
as
our
stand-in
for
how
the
cortex
probably
does
location,
because
it
probably
does
something
like
that.
A
So
we're
focusing
on
the
few
different
aspects
or
be
different
a
few
to
a
few
different
direction.
It's
like
all
right.
It's
like
stake
in
the
ground
grid
cells.
What
are
some
interesting
questions
here?
One
would
be
one
on
one
question
would
be
suppose:
one
part
of
your
brain
is
tracking
the
location
relative
to
one
thing
relative
to
urine,
pipe
and
suppose.
Another
part
of
your
brain
is
tracking
it
location
relative
to
something
else.
How
can
those
two
populations
of
cells
work
together?
How
can
you
determine
one
location
from
another
location?
A
Another
way
to
describe
that
is
a
coordinate
transform.
How
do
how
would
the
brain
do
coordinate
transforms
with
grid
sites?
That's
that's
one
thing:
I
didn't
focus
again,
a
fair
amount.
Another
thing
I've
been
focusing
on
did
a
few
of
us
have
Billy
is
it
is
related?
It's
how
we've
talked
a
lot
about
how
the
brain
would
learn,
objects
as
sets
of
feature
location
pairs,
sets
of
features
at
locations,
but
we
haven't
up
until
more
recently,
we
hadn't
really
focused
on
how
would
the
brain
compose
those
objects?
However,
they
create
composite
objects.
A
How
would
it
take
those
objects
and
make
more
out
of
them,
and
we
think
that
that
has
a
lot
of
inner
section
with
coordinate
transplants,
its
it
is
relating,
if
you
think
of,
if
you
think,
of
an
object
as
being
learned
as
like
a
CAD
file
as
like
a
CAD
diagram,
and
you
have
another
object,
you
have
another
yeah,
we'll
call
it
an
object,
also
learned
as
a
CAD
file
conceptually.
These
are
just
sets
of
coordinates.
You
have
one
in
this,
and
one
object
in
these
coordinates.
A
One
object
in
these
coordinates
suppose
you
want
to
learn
some
higher
object
that
is
composed
of
the
two
of
them.
This
can
be
done
through
basically,
a
coordinate
transform,
which
we
it
can
be.
It
can
be
represented
by
by
some
circuit
that
can,
if
a
circuit
can
say,
here's
my
location
in
one
of
those
files,
one
of
those
diagrams.
Therefore,
here's
my
location
and
the
other
by
the
other
diagram,
then
it
has
linked
them
together
and
that's
that's
another
that
it.
A
This
is
like
a
mechanism
for
imposing
objects
with
grid
cells
that
we
find
pretty
exciting.
So
we've
been
also
focusing
on
that.
Otherwise,
like
like
specific
events,
Geoff
is
he
just
left
the
office
he's
about
to
fight
to
Genelia
and
give
a
talk
there
I
think
on
Monday
I'm,
not
sure
if
that's
being
recorded
or
not,
but
it's
so
he's
actively
updating
his
MIT
talk.
A
You
like
making
little
tweaks,
but
it's
so
I
think
is
somewhat
similar
to
what
he
is
going
to
be
some
more
similar
to
really
good
at
IT
and
otherwise
we're
preparing
for
cosine,
which
is
a
conference
that's
going
to
be
in
Denver
this
year
it's
a
computational
neuroscience
conference,
I'm
preparing
a
poster
on
computing
allocentric
locations.
So
it's
going
to
be
very
good
and
silly,
and
it's
going
to
be
basically
the
content.
Oh
I,
didn't
I
didn't
talk
at
our
meetup
in
the
fall
because
that
November
November
I
did
a
talk
there.
A
It's
going
to
basically
need
the
content
of
that
talk,
but
just
updated
a
little
bit
of
our
latest
thinking
and
subitize
doing
a
poster
on
str
properties,
so
it'll
be
cool
to
see
that
I,
don't
think
you've
presented
that
other
post
at
our
conference
before
and
and
Jeff
is
a
co-presenter
on
both.
So
we're
actively
I,
don't
know
we're
actively
putting
our
thoughts
together
in
ways
that
we
can
present
to
people.
For
example,
Jeff
writing
a
talk
in
let's
write
in
posters,
so
it's
very
much
a
preparing
to
communicate
stage.
A
Well
then,
it's
like
it's
like
our
theory.
Work
is
almost
like
a
guilty
pleasure
at
this
point
or
is
like
coming
up
with
new
coming
up
with
new
stuff
right
now.
It's
like
you
don't
have
time
to
come
up
with
new
stuff.
We
have
to
write
a
freaky
poster,
but
then
here
we
are
like
solving
two
things:
it's
okay!
It's
a
good
problem
to
have
I!
Guess
anyway,
that's
the
best
quick
snap,
stop,
there's
a
lot
of
stuff
dealing
with
good
cells.
B
A
Would
say
word:
that's
that's,
who
we're
trying
to
impress
then,
and
yes
so
I
mean
I
would
say,
that's
probably
true
as
well.
Yes,
it's
definitely
true,
but
it's
also
what
we're
aiming
for,
but
Jeff
and
Siebert
side
did
this
sort
of
East
Coast
tour,
two
different
neuroscience
groups,
what
the
end
of
last
year-
and
it
was
just
received
really
well-
and
there
was
a
money
of
came
to
his
last
talk
came
from
you
know.
A
A
Okay,
so
I've
got
his
Matt
Pilar,
so
here
that
hilar
david
keene
and
christian-
and
these
are
all
three
people
that
are
that
are
involved
in
the
community,
for
I
want
to
give
you
three
people
a
chance
to
talk
amongst
yourselves
right
now
about
anything
that
you
would
like
about
the
community
fork.
If
you
guys
want
to
do
yeah
everything
you
want
to
discuss
any
takers.
G
Right
now
it
feels
a
little
rough,
so
just
getting
started
so
I
would
like
to
you
know
we
mentioned
with
who
is
Bresnik
is
that
is
he
here
or
no.
A
That
I
introduced
him
at
the
beginning
of
the
Hangout.
He
is
a
longtime
community
member
like
really
long
time.
He
was
here
in
2013
when
we
went
open
source,
he
was
the
first
person
to
build
Newton
after
18
days
of
everyone
else.
Fair
and
so
he's
got
a
lot
of
experience
with
this
code
base
and
the
C++
and
with
neuroscience
in.
F
A
G
Oh
well,
I
wasn't
sure
if
he
was
working
with
a
mentor.
Not
that
explains
that
so
I
was
Kyle
can
towards
him
for
directions,
since
you
recommended
that
still
well,
but
I'm,
just
kind
of
wondering
where
we
won't
want
to
start
right
now,
he's
pulling
in
the
Python
three
bindings,
I
think
Christian.
You
know
I
started
that,
but.
A
Let
me
ask
you
guys,
because
I
I'm
not
keeping
track
of
all
the
stuff.
That's
going
on
in
the
community
for
discussion.
You
guys
need
me,
you
know
a
peon
in
the
in
the
post
and
all
I.
Don't
read
it,
but
first
thing
you
got
to
get
builds
running
and
I,
don't
know
if
you
have
builds
running
yet,
but
that
was
to
be
the
first
thing.
I
would
work
on
automated
builds,
continuous
integration
of.
G
G
B
G
A
A
No,
not
not
that
that's
a
bad
thing,
but
he's
got
a
lot
of
things
that
he
would
do
to
get
the
core
to
improve
it,
that
the
meant
is
that
we
just
don't
have
time
to
properly
peer
review
and
test
and
all
that
stuff,
so
I
think
we're
going
to
need
to
before
with
the
core
to
do
the
stuff
that
you
guys
want
to
do,
especially
if
you're
gonna
add
language
bindings.
It's
to
have
two
beautiful
unless
you
put
those
language
bindings
and
some
other
new
repository
or
something.
A
Is
an
option
if
you
I,
think
you're
gonna
need
you're
gonna
want
to
make
changes
to
make
the
core
at
some
point
that
anamensis
not
gonna
want
I'm,
not
gonna
have
time
to
properly
that's
one
of
the
reasons
I
had
tried
to
encourage
mark,
especially
to
work
on
a
fork
at
this
point,
because
he
has
so
many
things
he
wants
to
do.
Okay,.
B
B
I,
am
more
the
newest
developments
of
order,
but
the
main
initiative
is
basically
formed
for
me
to
get
everything
going
on
Windows
using
the
visual
studio
and
be
along
the
way,
maybe
also
improved
into
code,
especially
for
new
picker
and
hopefully
move
the
whole
new
pic
code
to
Python
3.
So
what
I
have
done
is
basically
I
have
basically,
we
implemented
the
whole
Python
interface
from
for
you,
pakora,
using
an
open
source
library
called
PI
point
11,
which
is
just
similar
to
booze,
and
it's
but
the
good
thing
about
it's.
B
It's,
it's
very
modern
C++
code,
and
it's
it's
a
header,
only
library,
so
you
don't
really
have
to
compile
anything.
And
it's
it's
it's,
so
you
can
almost
think
about
it.
It's
like
it's
for
some
reason
say
it
is
working
and
it's
it's
almost
magical.
I,
don't
know
how
how
many
meta
template
programming
they
had
to
to
go
through
to
make
anything
working,
but
it's
actually
pretty
cool
and
it
comes
with
Snoopy
bindings
and
some
metrics
library
called
eigen
is
also
implemented
there
and
a
test
bindings
for
all
day
at
the
STL.
B
I
don't
know
how
much
you
guys
know
about
c++
code,
but
it
has
all
the
bindings
for
like
STL,
vektor,
STL
set
and
Q
and
all
these
data
structures.
So
he,
the
good
thing,
is
you
don't
have
to
have
to
convert
between
a
Python
list
to
a
standard,
STL
vector,
for
instance,
everything
all
of
that
is
basically
working
up
out
of
the
box.
B
What
I
have
also
done
is
I
have
started.
My
gradient
in
you
pick
quote
to
Python
3,
so
there's
a
script
out
there
called
2
to
3,
which
is
provided
by
the
Python
paper,
that
this
eh
translate
Python
2.0
2003
code
and
it's
almost
working,
but
there
are
still
some
holes
to
fill
it
which
I
have
done
and
if
you
like,
you
can
go
to
my
repository
and
you
can
see
all
the
the
tests.
The
new
pick
test
code,
panic,
Python,
fetoscope,
already
working
and
it
looks
pretty
promising.
There
are
a
few
issues.
B
B
A
Let's,
let's
bring
it
up
real
quick,
I'll
I'll,
just
show
it
it's
prettier.
This
was
in.
B
B
B
Him
hit
him
up
again
and
if
you
told
my
there
was
one
more
issue
that
I
was
that
I
was
tackling
days.
It's
it
has
to
do
with
stills.
This
is
this
is
parameter
type
cache
and
I
have
no
clue
how
you
guys
do
that
in
Swift.
I
cannot
find
any
source
code
in
this
you
pick
war
or
in
the
swift
code.
That
then
makes
it
work.
Somehow
I
have
a
feeling
we're
supposed
to
be
happening,
but
I
just
don't
see
where
and
and
you
Pig.
This
is
done.
B
A
B
So
if
you,
if
you
click,
there's
a
sentence,
do
the
region
constructor
tries
to
create
an
attribute
called
M
type
cache,
and
then
it
goes
into
new
pakora,
but
I
have
no
idea.
How
swift
does
it?
Oh
I
see
see.
There
is
some
logic
that
I
simply
don't
don't
that
I
cannot
find
in
any
of
the
source
code
and
I.
Don't
I,
don't
think
it's
anything
Python
3.
It
must
be
somewhere
in
new
mentor
code
and
I
simply
do
not
know
how
to
how
to
do
it.
B
But
I
tell
them
that
I
mean
I,
can
tell
you
guys.
I
mean
I,
have
one
fool
like
50/50
testing
scripts
from
new
Pig
and
the
imodium
mostly
running
foo,
so
I
think
that's
that's
theirs.
That
is
pretty
encouraging
the
only
drawback
or
suppose.
The
drawback
is
that
I
have
completely
removed
anything
kept
and
pulled
over
eyes.
I
know
that
I
know,
that's
it.
That's
that's
something
that
is
did
us
that
did
it.
It
did
something
did
it
is.
Did
you
guys
use
a
lot
I
did
that
you
like
to
use,
but
it
it's.
B
I
mean
aside
from
this
continuous
integration
that
there
is
something
that
I'm
not
really
good
at.
To
be
honest,
I
think
we
have
been
I
mean
you
have
been
making
some
good
progress,
and
now
it's
basically
just
to
tie
it
into
the
community
fork.
I
have
everything
on
github
in
my
own
repository
and
I'm
working
this
mark
now
to
basically
put
it
into
the
community
for
content.
Hopefully
you
get
some
more
success.
G
Quick
question
in
the
community
fork
arm:
I
have
been
wondering
is
there,
since
it
is
a
fork?
Is
there
a
degree
to
which
what
we're
doing
is
too
much
or
something
that
you
know
a
point
in
which
it
might
not
be
beneficial
to
fork?
I.
Imagine
the
only
reason
to
folk
at
this
point
is
to
pull
down
from
core
and
potentially
a
Numenta
to
pull
into
the
shoes,
but
if
we
get
to
the
point
where
it's
so
different,
but
if
it's
pulled
in
it
breaks
everything,
imagine
that's
not
desirable.
A
Fork
so
there's
the
stood
as
a
keyword
there
there's
nothing
that
we
can
do
to
restrict
you
from
doing
anything
one.
We
could
provide
guidance
in
my
opinion,
I'm
gonna,
be
honest
with
I,
don't
see,
I
think
it's
there's
a
low
probability
that
noventa
that's
going
to
pull
from
the
community
fork
in
the
future.
A
It's
it's
still
sort
of
like
a
research
platform,
although
we
call
you
know
you
pick
one
point
now
when
you
say
you
can
use
it
in
production,
I
think
it's
pretty
stable
at
this
point,
I
think
something
else
is
what's
what's
going
to
evolve
into
the
big
HTM
project
that
people
really
start
using
I've
always
felt
like
that.
I
think
that's
what
the
way
the
technology
works.
A
So
my
advice
to
you
is:
don't
worry
about
it.
Do
what
you
do,
what
you
think
works
best
for
you,
guys,
the
people
that
are
going
to
be
using
it
and
don't
feel
too
concerned
about
whether
we'll
be
able
to
take
advantage
of
it,
because
we've
got
our
own
priorities
and
and
they
may
not
align
and
if
they,
if
they
do
line
in
the
future,
we'll
try
and
pick
your
work
up.
But
if
it's
hard,
that's
our
own
fault.
So.
G
I
guess
so
I
mean
I
would
like
to
just
use
new
pick,
and
you
know
whatever
language
I'd
like
at
the
time
or
an
HTM
solution
arm
right
now,
as
since
it's
in
research
is
count
hard
to
pick
the
one
that's
leading
since
there
is
no
leader
arm,
which
kind
of
brought
me
back
to
new
pick.
And
you
know
if
there's
an
HTM
solution.
I
would
like
to
use
the
official
one
if
there
isn't
one
Newton.
G
G
Frank
is
an
HTM
solution.
I
know
there's
the
HTM
Java.
So
if
you
don't
cringe
a
bow,
but
you
know
it's
it's
kind
of
I'd
rather
not
have
to
learn
out
entirely
new
language
in
order
to
use
a
new
picked
and
I
can
but
I
mean
I
like
to
just
pick
it
up
and
play
with
it
and
I
think
they
might
bring
in
more
people
into
the
community.
If
you
know
we'd
offer
flavours
that
they're,
you
know
appeal
of
them.
G
G
A
G
G
B
G
Well,
I'll
look
into
that.
I
was
looking,
you
know
unless
I
play
I
mean
bill,
different
applications
use
it
use
it
using
it
pick.
Not
just
you
know,
use
an
application
that
uses
new
pick,
but.
A
D
I'm
doing
with
the
C
sharp
interface
pretty
big
project.
My
my
goal
is
to
not
actually
change
anything
that
or
exist
in
or
would
rather
build
a
set
of
wrappers
for
all
of
the
classes
that
are
that
are
part
of
the
interface
and
those
wrappers
would
be
written
in
C++,
C,
L,
R,
Co,
I,
suppose
that
they
can
then
be
called
c-sharp.
D
D
It's
we're
a
long
ways
from
having
show
anybody
its
roughed
out
now.
A
lot
of
the
the
type
classes
in
type
of
Factory
I
have
some
of
those
set
up
and
I'm
starting
to
work
on
the
test
routine.
So
as
I
build
tests
to
check
out
each
of
these
all
sighs,
hopefully
they
won't
have
to
modify
any
of
the
existing
code,
the
tip
or
just
wrapper
them.
So.
D
Yeah
I
would
think
so.
I
should
give
you
all
the
capability.
There
is
one
problem
in
addition.
Well,
in
addition
doing
the
seats
sure
I
wouldn't
also
step
this
out,
such
as
C++
right
now,
there's
a
piece
of
pythons
required
to
wrapper
the
algorithms
so
that
they
could
be
plugged
in
and
there's
a
plug-in.
D
Add
it
to
Marxism,
but
is
it
pretty
much
a
defendant
at
this
point?
At
some
point,
I
would
have
pulled
out
the
latest
stuff
that
they're
doing
it
uses
my
basin
put
my
modules
in
the
end
and
then
send
them
in,
but
right
now,
I
just
sort
of
find
them.
I
say
no
good,
but
there's
a
lot
of
it.
There's
a
lot
of
classes
that
she
does
and
each
one
of
those
has
to
be
ham
and
wrapper
yeah.
D
E
F
D
B
D
B
D
D
A
D
B
I've
I
think
with
what
David
is
talking
with
bonus
item
was
very
cute
amendment
out
of
its
region.
It
is
to
find
umpires
and
you
can
pull
it
into
you
pick
or
so.
There
is
something,
obviously,
that's
that
you
pick
or
does
not
provide.
So
you
have
like
certain
encoders
or
or
scalar
terms
of
scalars
or
some
scalar
and
kula,
for
instance
over
so
there
will
be
implemented
and
person
only
and
not
come
from
you
pick
or
but
I
think
that
just
to
be
a
dancer,
I.
A
Okay!
Well,
let
me
bring
up
one
more
thing
about
the
community.
If
you
guys
think
like
this
sort
of
powwow
is
useful
and
you
want
to
do
like
a
monthly
community
for
hangout
I'm
happy
to
sort
of
moderate
that
for
15-20
minutes
once
a
month
now.
So
you
know
if
you
want
to
do
that,
I
might
bring
up
on
the
power.
A
C
C
Marcus
when,
when
you
talk
about
you
know
these
grid
cells
as
a
model,
for
instance
modeling,
Allison,
type
locations
or
locations
in
a
room,
and
also
possibly
you
know,
sort
of
missus
mechanism
of
composing
objects.
You
talk
about
that
as
well
as
composing
locations.
How
is
all
this
different
from?
Let's
say
a
generic
pooling
function?
You
know
this
idea
of
pooling
in
neuroscience
where
you
take
a
bunch
of
inputs
and
you
try
to
make
one
sort
of
constant.
You
know
a
SGR.
C
You
know
set
of
neurons
active
over
all
the
inputs
that
are
coming
into
sort
of
abstract.
What's
happening,
I
was
thinking
about
you
know.
Rex
is
composing
two
objects
and
that's
more
or
less
the
the
operation.
That's
going
on
in
that
sort
of
from
my
understanding
of
what
a
pooling
layer
would
do.
Let's
say
we
would
take
two
objects
and
said
compose
them
as
as
one
a
super
object
or.
A
A
C
C
Would
it
be
similar
to
them,
for
instance
like
I?
This
is
what
I
call
it.
I
can
say
it
be,
for
instance,
like
a
temporal
memory
algorithm
or
a
sensorimotor
temporal
memory,
algorithm
that
that's
sort
of
you
know
like
an
inference
layer
that
could
possibly
do
that.
Mapping
is
that
sort
of
more
now
gets
it
to
the
mechanism
that
that
would
require
a
little
of
lovely
you're.
Fine
I
would.
A
Say
that
something
similar
to
a
temporal
memory,
London
tend
to
specialise.
To
has
to
be
special
in
that
sense,
like
two
paths
that
land
at
the
same
point
need
to
resolve
in
the
same
representation
which
you
wouldn't
get
this
at
all.
Neither
but
something
similar
can
Ken
I
can't
approach
this
problem
sort
of
like
what
what
that
can
do
is
it
can
say,
look
okay,
say:
I,
know
my
location
in
this
place
and
say:
I
know
my
location
in
this
place.
A
Don't
I
have
what
these
cases,
the
suppose
you
know
both
of
those
then
you're
correct
that
a
temporal
memory,
X
system
called
a
path
integration
system
could
respond
to
motor
commands
and
computing
any
location
when
a
temporal
memory
can't
be
is
supposed
to
know
a
location
in
one
of
these
faces,
and
you
don't
know
your
location.
Otherwise,
you
don't
you
don't
know
the
sector's.
What
do
you
do
not
like?
How
do
you
figure
out
here,
location
in
this
space
and
I
would
say
that
composing
objects?
A
One
way,
one
way
of
approaching
the
problem
of
composing
objects
is
solving.
That
problem
is
solving
the
question
of
how
you
link
a
bunch
of
these
faces
together,
such
that
given,
when
in
the
location,
you
can
figure
out
the
other
luggage.
You
can
picture
this
as
as
like,
let's
just
talk
about
like
an
experiment
with
a
raft
walking
into
a
room
and
it
sees
a
cute
card.
A
I
sees
a
big
white
card
in
front
of
it
and
then
see
is
like
it's
sense,
isn't
odor
near
it
and
with
that
information,
it's
like
it's
like
a
know.
This
room
I've,
seen
this
room
before
and
based
on
my
experience
in
this
room.
There's
cheese
or
there's
like
there's
a
bit
of
a
reward:
okay,
and
because
it
figured
out
this
location
in
this
room.
It
now
knows
its
location
relative
to
the
cheeks
without
having
seen
it
without
having
fun,
and
this
is
where
this
linking
of
spaces
come
to
time.
A
C
Let's
say
like
an
infant's
lair
that
would
go
to
a
pooling
layer
and,
basically
you
know
the
inference.
I
was
telling
you,
okay,
I'm
in
this
I.
Think
I'm
right
here
in
this
room,
I
saw
a
motor
movement.
So
you
know
it's
dynamically.
Changing
the
point
layer
would
then
be
constantly.
You
know
trying
to
say
you
know
what
what
room
do
I
think
I'm
in
you
know
in
the
alicent
reform.
What
object
am
I
honor?
C
You
know
whatever,
and
so
this
this
cooling
layer
would
sort
of
say
I'm
in
this
room
and
now
let's
say
you
figure
out
what
room
you're
in,
because
you
get
a
successful
prediction
of
you
know,
sort
of
egocentric
features
or
whatnot,
and
you
can
begin
to
make
it
a
sort
of
apical
feed
back
down
to
what
you
would
predict.
Let's
say
an
object
would
be
in
that
room
based
on
place
on
your
knowledge
of
the
room
and
your
pooled
reputation
and
then,
as
it
seems
like
it's
almost
a
two-step
process.
C
Like
that
word,
you
know
you
first
try
to
find
your
position
in
the
room.
Then
you
figure
out
how
it
relates
to
your
current
or
your
previous
model
of
that
room
objects
in
it
stuff
like
that,
and
then
you
then
feed
that
back
down
to
what
objects
might
be
in
that
rias
entually.
You
know
what
what
things
are
at
that
location
in
your
sort
of
pooled
environments,
and
it
seems
like
it's
there's
not
from
my
understanding
of
the
cortical
Anatomy,
there's,
not
necessarily
a
straight
direct
path
from
I'm
in
this
location
in
the
room.
C
A
Because,
like
you're
correct,
because
because
of
course
you
couldn't,
the
rat
couldn't
have
known
like
a
priori
that
there's
cheese
over
they,
some
I
didn't
really
mention,
and
all
that
where
a
memory
must
have
been
formed,
the
rat
lungs
didn't
somehow
Andrew
formed
his
memory.
That's
like
the
build
spaces
are
like
basic
and
that's
where
something
similar
to
okay.
C
You
would
probably
need
something
more
along
the
lines
of
Queen
layer
or
inference
layer.
You
know
the
sort
of
general
function
rather
than
a
more
specific.
You
know,
type
of
physical
sense
of
a
grid
cells,
and
so
this
is
sort
of
shift
between
you
know.
Pulling
and
grid
cells.
I
feel
like
that's
what
the
the
neocortex
did.
Essentially,
when
it
transition
from
the
and
Serano
cortex
maybe
evolved
to
the
neocortex,
is
it
sort
of
it
made
this
more
general
to
possibly
cope
with
these
abstract
objects?
C
To
not
necessarily
have
to
represent
everything
as
physical
stuff
would
be
able
to
abstract?
The
point
of
you
know
a
low
centric
locations
where
they're
not
actually
analysis.
Patients
are
just
the
idea
of
them
relative
relative
to
the
abstract
concept
or
whatever
yeah.
A
A
This
looks
like
crazy,
interesting:
it's
hexagons,
boys,
yeah,
okay,
anything
else
all
right.
This
has
been
a
really
great
discussion.
Thanks
everybody
for
joining
a
lot
of
participation
and,
like
I,
said
I'll
post
on
the
forum.
If
you
guys
want
to
do
a
hangout
to
catch
up
on
community
ports
work,
let
me
know
otherwise.
I'll
see
you
guys
on
a
shame
forum
and
have
a
wonderful
weekend
thanks.
Everyone.