►
From YouTube: HTM Hackers' Hangout - Oct 6, 2017
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 week's content will be:
- Strange Loop / TWIML.AI
- Meetup Nov 3rd https://www.meetup.com/numenta/events/243501858/
- upcoming interviews
- guest blog post from Ali (sunguralikaan) coming soon
- SMI / Highbrow
Links:
https://www.thestrangeloop.com/2017/the-biological-path-towards-strong-ai.html
https://twitter.com/Numenta/status/913835384446046209
https://www.youtube.com/watch?v=-h-cz7yY-G8
https://twimlai.com/twiml-talk-10-francisco-webber-statistics-vs-semantics-natural-language-processing/
https://www.meetup.com/numenta/events/243501858/
https://numenta.com/blog/2017/10/04/guest-blog-post/
A
It's
got
a
notification
apparently
R'lyeh.
They
just
got
a
notification
of
it.
Welcome
to
the
HTM
hackers,
hangout
I'm
Matt
Taylor
from
Numenta
and
is
October
6th
I'm
gonna
mute.
You
Daniel
and
I
want
to
welcome
our
new
subscribers
from
a
strange
loop.
I
noticed
when
I
did
a
presentation
on
strange
loop
last
week
and
I
got
a
bunch
of
new
subscribers
on
our
YouTube
channel.
So
welcome
thanks
for
joining
us.
We
do
this.
A
Htm
hackers
hangout
live
once
a
month
and
it's
a
Google
hangout
held
for
our
online
community
anybody's
free
to
join
in
the
discussion
by
connecting
directly
to
the
Hangout
like
these
people
down
here
or
you
can
just
connect
to
the
video
or
watch
the
video
live
on
YouTube
and
comment.
There
I
can
answer
those
questions
check
out
more
information
in
the
video
description.
I
put
a
bunch
of
links
about
everything,
we're
going
to
talk
about
down
there.
A
If
you
have
something
specific
that
you'd
like
to
discuss
or
we
just
want
to
learn
more
about
the
HTM
community,
please
join
our
forum,
HTM
forum.
It's
a
discourse
that
no
matter
that
word
so
without
further
ado.
I'm
gonna
go
to
my
agenda,
so
we
can
push
through
that
and
I'm
going
to
share
some
ice
cream.
So
I
do
want
to
tell
you
a
bit
about
how
this
last
conference
went,
which
I'm
very
happy
with
so
screen.
One,
that's
not
the
right
one!
A
There
was
a
good
bright,
our
media,
the
page
here
there
does
picture
the
crowd,
so
my
room
was
big,
probably
the
second
biggest
room
there
beside
from
the
actual
Opera
House
room,
and
there
was
a
four
hundred
people
that
could
sit
there
and
there
were
standees
only
by
the
time
it
was
it
started.
So
it
was
a
good
audience.
So
I
was
really
happy
about
that.
I
also
had
a
chance
to
do
an
interview
with
these
this
podcast.
These
guys
called
this
weekend.
A
Machine
learning
and
AI
they
interviewed
Francisco
Webber
last
year,
they've
got
a
bunch
of
different
interviews
of
people
in
the
AI
community
on
their
website.
They
put
them
on
YouTube
and
they
do
podcasts
so
I'm
going
to
be
up
there
because
I
took
it.
I
did
a
video,
it's
strange
loop
with
them,
so
that's
great
and
the
next
big
thing
is
we're
having
a
meet-up
on
November
3rd.
A
So
this
is
on
the
same
day
that
I'm
talking
at
the
open
data
science
conference
in
San
Francisco
at
the
airport
there
in
that
area,
so
we're
going
to
have
a
meet-up
on
this
on
that
same
night.
So
after
that
event,
there'll
be
three
tops.
We
have
a
community
talk
from
Jason
toi
who's,
a
member
of
our
community
and
he's
going
to
talk
us
through
something
he
created
to
help.
A
Ai
systems
deal
with
3d
objects,
which
is
very
relevant
to
some
of
the
stuff
we're
working
on,
and
then
our
own
Marcus
Lewis
who's
connected
here
is
going
to
talk
about
location
relative
to
the
object
environment.
So
that's
going
to
be
a
lot
about
the
new
object
locations
and
and
how
we're
thinking
about
that
stuff
and
then
Jeff
will
actually
be
the
the
main
speaker
that
night
and
he'll
give
us
an
update
on
GUI
or
some
columns.
That's
going
to
be
mostly
just
going
over.
A
What's
in
the
paper,
especially
good
for
anyone
new
coming
into
this,
you
know
the
new
layers
and
columns
Theory
that'd
be
a
good
talk
for
that.
The
other
two
are
kind
of
more
deep
I
guess
more
it
more
detailed,
but
Jeff's
would
be
a
great
talk
for
anybody
to
come
in
and
listen
to
so.
I
want
to
point
that
out
that
is
November
3rd
at
7
p.m.
and
it's
the
plaza
san
francisco
airport.
A
Also
we
had
a
nice
guest
blog
post,
a
few
guys
are
in
our
community.
Hopefully
you
saw
this
amazing
demo
that
one
of
our
community
members,
ours,
have
ours,
I'm,
saying
his
name,
but
his
name's
Ali,
and
he
put
together
this
really
cool
thesis
with
this
big
3d
world
and
made
an
autonomous
agent
using
HTM
and
did
some
really
really
cool
visualization.
So
we
asked
him
if
he
would
do
a
blog
post
and
talk
about
how
how
this
worked
and
and
includes
some
of
the
the
visualizations
from
his
demo.
A
He
did
so
thanks
to
him
for
doing
that.
That
was
very
cool.
That's
I
think
all
that
I
wanted
to
talk
about
the
other
thing
out
on
my
agenda.
I,
don't
really
have
much
to
talk
about.
I
am
I,
am
working
on
some.
This
like
real,
like
live
demo
of
a
hand
touching
something
and
the
neural
activations
in
HTM
and
showing
like
how
it
can
resolve
an
object.
I'm
working
on
this
with
Louis
one
of
our
engineers
and
I'm,
going
to
try
to
incorporate
that
into
whatever
I
next
HTM
schools.
A
I.
Don't
have
anything
to
show
about
that
yet,
but
I
put
it
on
the
agenda
just
to
kind
of
pride
me
and
to
continue
working
on
it.
Okay,
you
know
that
was
pretty
good
for
five
minutes.
I
got
through
my
part
of
the
agenda,
really
quick
so
like
to
pass
it
on
Marcus.
Do
you
want
to
say
anything
about
research
at
all
sure,
as
usual,
this
is
pretty
informal.
I
didn't
really
plan
much
so
yeah
a
few
things
are
going
on
in
parallel.
A
That
is
that's
all
coming
along
I'm
still
working
on
the
whole
task
of
inferring,
the
allocentric
location
and
that's
gone
quite
this
gun
quite
well.
Les
I
did
one
of
these
little
research
updates.
I
I
claimed
that
in
a
couple
weeks,
I'm
going
to
poster
right
up
showing
a
showing
a
big
idea
and
I've
finally
finished
that
write-up
is
now
a
few
months
later,
and
so
that
will
go
on
github
within
the
next
few
days.
A
I
would
say,
and
we're
really
aiming
for
like
getting
that
into
a
paper
and
such
so
so
I'd
say
one
of
our
big
research
focuses
is
calculating
the
allocentric
location,
so
I'm
excited
to
read
Daniels
paper,
because
it
was
timely
that
he
pretty
much
proposed
a
solution
to
that
as
well.
So
I'm
excited
to
read
that
paper.
I
would
say
the
other
big
task
of
the
big.
The
big
question
we're
considering
is.
A
There
are
different
terms
you
could
apply
to
this,
but
the
general
problem
of
if
you're
learning
objects
and
like
these
modalities,
specific
cortical
columns
like
one
that's
attending
to
your
finger
and
another.
That's
focused
on
a
particular
patch
of
your
retina
and
another
part.
That's
a
particular
patch
of
your
cochlea,
however,
that
all
works.
How
is
it
that
you
can
learn
an
object
in
one
cortical
column
and
then
infer
it
and
another,
the
general
task
of
just?
How
does
that
work?
A
A
Is
an
example
we
use
sometimes
and-
and
you
can
also
set
up
the
problem
other
ways
you
can
set
it
up
as
if
you
have
a,
if
you
have
a
array
of
sensors,
that
each
have
partial
understanding
each
have
partial
info
for
an
object
like
no
one
sensor
has
sensed
every
location
on
the
object,
but
all
of
them
put
together
have
sense
most
of
the
locations.
So
how
do
all?
How
do
the
sensors
work
together?
A
Does
it
require
them
all
to
learn
everything
somehow,
like
maybe
well
you're
sensing
an
object
while
you're
moving
your
hand
over
an
object?
Maybe
every
sensor
is
learning
everything
they
or
or
maybe
each
is
just
learning
little
bit
and
somehow
they
still
work
together,
and
there
are
different
ways
of
approaching
that
problem
and
I
would
say
that
that
would
be
that's
like
the
I'd
say.
A
A
But
just
if
you
want
like
a
picture
of
what,
what
are
we
doing
with
all
this
all
this
time?
That's
that's
another
thing,
so
yeah
diving
into
these
questions
and
just
having
these
conversations
with
people
cool.
You
know
that
I.
It
seems
like
from
my
perspective
anyway.
I'm
not
doing
research
actively,
but
I
try
and
keep
up.
A
And
after
this,
after
the
layers
and
columns
paper,
there
was
kind
of
a
focus
on
sensory
motor
integration,
but
the
the
idea
that
that
it
exposed
is
that
sort
of
that
modular
circuit
unit
that
that's
potentially
constructible
from
layers
and
columns
with
in
different
ways,
I
think
there's
a
lot
of
different
ways.
You
could
take
that
you
know
and
you're
thinking
how
I'd
try
it
when
you're
thinking
about
how
things
work
so
hey,
let's,
let's
open
it
up,
I
guess,
if
any,
let's
we
were
talking
to
Daniel
right
earlier.
A
B
Sort
of
there's
sort
of
objects
in
their
abstract
form,
just
sort
of
like
a
phone,
regardless
of
its
orientation
or
location
objects,
their
location
in
space,
their
orientation
in
space
relative
to
themselves,
sort
of,
and
then
the
states
or
orientation
of
the
body
relative
to
itself,
and
it
turns
out
that's
actually
very
similar
to
how
you
think
of
motor
commands
motor
commencer,
really
just
the
change
in
that
state.
And
so
essentially
the
theory
says
that
each
egocentric
or
Alison
fusion
is
modeling.
Two
of
these
things,
and
actually
those
two
help
each
other
out
a
lot.
B
But
there's
also
a
link
between
the
two
sister
regions
themselves
between
me
tonight,
because
I'm,
agnostic
and
so
they're,
actually
talking
to
this,
at
least
in
the
case
of
prints
they're
talking
to
each
other
at
the
level
of
what
I
call
the
CT
module
CT
module
is
a
set
of
layers,
usually
put
a
feed-forward
Park.
It
consists
of
layers,
six
AIDS,
five
D
beta
by
B
alpha
and
then
five
a
which
has
two
parts
that
are
really
distinguished
company.
B
So
basically,
these
two
modules
within
the
region
are
actually
communicating
each
other
in
the
same
way
that,
for
instance,
four
and
six,
they
aren't
communicating
with
each
other
in
a
similar
way,
and
essentially
it's
passing
back
this
information
about
orientation
and
the
orientations
of
body
and
its
current
state.
And
so
you
get
this
sort
of
communication
between
these
two
sister
regions
and
what
happens
is
the
location
of
an
object,
sort
of
helps
the
motor
commands
which
help
the
orientation
of
an
object
which
helps
the
object,
model
and
sort
of
ghost
food?
B
And
it
also
goes
back
in
a
way
and
so
and
actually
there's
sort
of
even
more
detail
between
the
losses
that
I
won't
get
into,
but
essentially
turns
out
just
the
mere
act
of
computing
and
how
about
some
big
location?
It's
actually
only
one
part
of
the
loop
is
actually
all
these
other
things
that
are
happening.
That
you
know
be
done
to
the
computation
of
the
altitude
location
signal,
but
are
actually
just
as
important
and
actually
it.
For
instance,
you
can
go
on
mode
of
commands,
there's
actually
the
exact
same
process
going
on.
B
You
know
in
terms
of
generating
outstanding
location
signal,
as
there
are
inferring
what
the
state
of
the
body
is,
and
actually
only
that
I
guess
on
this-
that,
via
the
really
sort
of
scary
part
of
this
theory,
that
really
non-intuitive
part
is
that
you
technically
don't
even
need
a
copy
of
motorman's
actually
to
compute.
The
allison
implication
signal
to
compute
the
current
state
of
the
body
to
do
all
the
things
that
we
would
think
you
would
need
a
motor.
B
Those
motor
movements
and
those
are
sort
of
future
motor
movements
that
can
be
used
for
the
computation
of
the
object,
and
so
it's
a
sort
of
recursive
thing
and
in
the
paper
I
I
explained
in
detail
of
the
connections
that
I've
found
rious
hypothesized
to
a
degree
through
my
research
that
I've
done,
you
know
sent
papers,
and
so
essentially
it
actually
says
that
we
actually
might
have
all
the
machinery
you
need
actually
make
two
of
these
sister
regions
at
least
something
forward
or
portions
of
them.
You
actually
have,
for
instance,
a
inference.
B
Algorithm
meant
that
just
recently
publish
a
paper
talking
about
I
even
have
sensory
motor
inference
and
temporal
inference
in
the
same
set
of
cells
and
I.
Think
that's
exactly
right
that
layer,
four
is
actually
doing
both
temporal
and
sensory,
motor
and
so
and
along
6a
as
well,
but
also
it
turns
out,
you
need
the
same.
You
know
Larry
I,
guess
a
temporal
pool
in
there
is
looking
my
deposits
it's
this
thing.
B
That
is
actually,
if
you
think
about
it,
the
difference
between
an
inference
layer
and
the
cooling
layer
or
a
temporal
foyer
is
actually
nearly
the
cells
which
they
can
then
de
temporally
organize
cells
with
a
connective
so,
for
instance,
in
an
inference
layer,
specifically
with
the
set
of
temporal
inference
it's
connecting
to
the
t,
minus
1
cells.
You
know
the
cells
from
the
previous
time
step
now
what
I've
had
homicides?
B
That's
that's
all
said
and
done
the
boiling
layer,
that's
all
mainly
hypothesized
Senate
done
I've,
also
sort
of
thought
that
those
are
actually
the
only
two
things
you
need
to
put
into
code
essentially
to
actually
have
in
the
cortex
to
be
able
to
make
a
full
cortex,
and
so
it
there's
only
two
types.
You
know
that
that's
a
hypothesis
and
so
I've
been
actually
trying
to
create
a
cooling
layer
of
my
own,
along
with
an
inference
layer
which,
of
course
the
Coda's
are
to
establish
for
that
hook.
B
I've
been
trying
to
make
a
point
layer
and
code
myself
and
it's
coming
along
for
an
attempted
to
pseudocode,
for
it
I
think
dementia
actually
has
a
point
layer
already
made.
I,
looked
at
the
curler
I,
don't
know
if
it's
exactly
biologically
so
I
want
to
be
accurate.
The
code
that
I
saw,
but
this
this
way
of
thinking
about
it,
might
actually
make
the
problem
simpler,
because
it's
virtually
gonna
be
almost
identical
to
the
previous
code.
It's
just
that
you're
connecting
to
you
know,
I
go
to
the
inference
code.
B
It's
just
that
you're
connecting
to
currently
on
themselves,
which
really
makes
the
bigger
house
and
so
I've
been
working
on
that
and
hopes
of
actually
testing.
My
theory
testing
the
generation
of
the
alistair,
typical
patient
signal
theory
and
info
trying
to
make
the
two
sister
regions
and
also
code
in
the
connection
which
I
hypothesize
in
the
paper.
Now
I
guess
the
little
final
other
thing
is.
That
is
instead
of
the
weird
thing
I
came
across,
but
you
actually
don't
even
mean
say
that
again
you
don't
need
one,
oh
that
then
just
still
connections.
B
Okay,
so
you
cut
out
for
a
second.
So
thank
you.
So
in
your
actually
modeling
into
code,
just
like
a
temporal
inference,
layer
would
not
have
any
that's.
Just
connections
already
decided
before
it
would
learn
those
over
time
based
on
such
I.
Think
the
e
and-a
region
is
are
actually
communicating
each
other
just
to
leave
these.
They
have
to
learn
those
connections.
So
there's
essentially
going
to
be
a
potential
connectivity
between
certain
layers
here,
and
certainly
is
here,
but
yeah
exact
connections
of
weather
connecting
to
this
layer
within
this
layer
and
that's
also
learning.
B
What
is
necessary,
though,
is
the
able
connections
which
there
are
able,
or
my
research
I
found,
that
there
are
able
connections
between
these
two
regions,
that
sort
of
determine
the
general
flow,
the
real
proximal,
little
information,
I,
I,
say
proximal,
not
that
it's
actually
proximal,
and
you
know
it's
not
actually
seen
on
some
to
the
proximal
inputs,
but
it's
causing
a
action
potential,
or
at
least
in
the
paper.
I
hypothesize
that
essentially
distal
and
an
apical
inputs
and
currently
and
there's
evidence
for
this.
B
I
think
that
that
concurrency
of
those
two
types
of
inputs
which,
by
themselves
don't
actually
come
back
together,
they
actually
cause
an
action
potential
in
the
cell,
which
is
maybe
to
the
you
know,
a
same
potency
as
a
prompting
word
and
so
essentially
be
able
connected
a
very
important
as
a
sort
of
pruning,
the
activity
of
where
informations
being
corrected
in
terms
of
on
layer
by
layer
basis
and
that's
a
little.
You
know.
A
C
You
can
you
hear
me
now:
I
could
hear
you
before
Oh
No.
Okay,
now
can
you
hear
me?
I
can
still
hear
you,
okay,
okay,
that's
all
right
if
they're
still
lying
through
my
microphone
on
my
side,
I'm
working
on
two
projects
and
it's
early
too
early
to
talk
about
the
met,
but
I
have
been
working
on
them
for
a
long
time.
C
One
is
I
want
to
know
what
HTM
feels
like
I'm
beholden
to
the
psychology
side,
so
everything
I
work
with
function,
all
systems
and
I
want
to
work
towards
where
HTM
is
so
functional
system
evening
online
humans,
I
know
and
I'm
trying
to
reach
out
to
the
center.
So
what
I'm
trying
to
get
to
the
point
where
I
can
propose
a
model
to
you
guys
from
the
psychology
side
and
see
if
we
can
meet
in
the
middle
somehow.
C
The
second
thing
I'm
working
on
is
what
is
the
basal
ganglia,
companion,
theory
to
HTM
and
that's
coming
along
super
well
and
I
love
it.
But
it's
also
way
too
early
to
talk
about
it.
But
those
are
my
two
projects
someday.
They
will
be
done
and
then
I
will
be
super
excited
to
share
them
great.
Thanks
for
the
update.