►
From YouTube: HTM Hackers' Hangout - July 1, 2016
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
The
thing
is:
there's
a
research
meeting
that
if
its
offices
run
in
the
box
on
the
record
of
the
year
right
away
and
it's
the
Friday
before
independence
day
weekend
in
the
United
States,
so
a
lot
of
people
I
think,
take
the
day
off.
But
I
don't
care
I'm
going
to
do.
Htm
actors
hang
out
no
matter
what,
because
I
know
I
know
you
really
really
want
it.
So
I
do
have
a
couple
things
like
I'd
like
to
talk
about.
A
First
of
all,
sort
of
momenta
priorities:
I
guess,
keep
you
updated
on
what
what
we're
thinking
about
we're
doing
a
bit
of
a
priority
shift
for
our
engineering
team
to
move
away
from
building
sample
products
and
stuff
to
research.
So
you
might
have
noticed
that
we
just
released
a
product
called
HDM
studio
this
week.
I
think
it
looks
really
slick,
it's
a
desktop
application
for
OS
X
or
for
windows.
So
you
can
run
new
pic
and
it
runs
new
pic
inside
of
this
little
icon-
environment,
isolated
from
windows
or
OSX.
A
So
it's
a
kind
of
a
beefy
download
because
it
brings
its
own
environment
with
it,
but
it
was
the
only
way
we
could
reliably
deploy
this
little
desktop
app.
So
you
run
this
desktop
app
you!
You
can
learn
more
about
this
at
nemikham,
I!
Think
there's
like
a
it's!
It's
one
of
the
big
banners
on
the
front,
so
it's
HTM
studio
release
anyway,
you
downloaded
there,
you
can
run
you
can
run
it
your
data
straight
through
it.
A
All
you
have
do,
is
figure
data
in
a
comma
delimited
format
and
make
sure
that
its
temporal
and
it
will
find
anomalies
their
data
and
highlight
them
right
there
on
the
screen.
It's
really
pretty
slick.
So
if
you're
interested
in
trying
HTM,
then
you're
not
a
programmer,
you
have
some
data
lying
around
that
you
think
might
be
applicable.
A
A
We
want
to
do
in
the
meantime
before
we
focus
on
two
percent
on
research
stuff,
and
this
is
the
research
that
Jeff
has
talked
about
on
its
jemput
poor
arm
and
on
YouTube
research
meeting,
we're
going
to
start
focusing
on
that,
but
before
we're
quite
ready
to
do
that,
there's
some
cleanup
task.
We
wanted
to
do
a
new
pic
specifically
the
most
important
thing.
A
Scott-
and
I
both
agree
on
this-
is
to
get
the
serialization
stuff
done
so
it'll
be
be
kept
at
proto
all
over
stuff
and
that
will
increase
the
performance
of
serializing
and
deserialising
new
pic
models
to
disk
by
like
10
x
or
something
crazy.
So
it
will
speed
up
everything
automatically
well,
if
you're
searing
models,
that
means
it'll
makes
HTM
engine
a
lot
faster,
for
example,
just
by
upgrading,
so
we're
going
to
get
that
done
and
there's
other
some
other
things.
A
It
probably
will
still
be
playing
the
same
role
right
now:
I'm
still
kind
of
5050
trying
to
do
educational
stuff
versus
project
management,
new
pic
of
the
core,
so
I'm
going
to
still
be
doing
that,
but
I
am
going
to
have
a
nice
window
into
the
research
work
that
image
is
doing
and
that
should
all
be
happening
in
our
research
repository
that
we
usually
do
research
in
and
the
reason
you
have
a
see
any
of
that
happening
yeah.
That
is
because
we
haven't
done
it
yet.
A
So
all
of
the
research
that
we've
done
so
far
is
just
like
reading
and
thinking
and
we're
not
quite
to
the
point
where
we've
got.
We
have
the
pseudocode
down
or
writings
test
dummy
implementations
of
anything
when
that
does
happen,
you're
you're
going
to
see
that
in
the
research
repository.
So,
if
you're,
watching
that
it's
called
new
picked
up
research,
you'll
see
it
happen,
there
I
was
hoping
Aaron
right
might
be
on.
If
you're
watching
Aaron
no
big
deal
Aaron
went
to
do
they
flew
me.
A
Was
it
jun
27th
last
week
or
this
week
this
week,
I
couldn't
make
it
unfortunately
and
I
could
I'm
for
I,
don't
think
anybody
had
been
to
made
it
to
the
beat
up
honestly
wisdom.
It
was
a
bad
timing
and
I
didn't
have
much
notice
or
able
to
try
to
make
it
up.
So
sorry,
we
missed
that
flank
meet
up
an
errand
if
you're
out
there.
A
Let
us
know
how
it
went
post
on
HD
inform
how
it
was
received.
What
are
the
think
people
thinking
about
html5
out
about
that?
Okay
on
to
new
news
hour,
nap
competition?
Remember
nav!
Is
the
nimitta
anomaly
benchmark
where
we
were
comparing
anomaly:
detection
capabilities
of
a
bunch
of
temporal
streaming
unsupervised
learning
systems,
including
new
pic
lstm,
etsy,
skyline,
twitter.
A
That
competition,
which
we
have
had
open
for
a
while
now,
which
was
going
to
be
closed
today,
we've
extended
it
I
think
until
july.
Twenty
second
go
did
it
admit
to
that
org,
slash,
AB
and
maybe
to
find
out
about
that.
We
have
a
few
entries,
but
there's
still
big
opportunity
to
win
up
to
five
thousand
dollars
there.
A
A
A
So,
what's
Matt
been
working
on,
I
have
been
just
neck
deep
in
spatial
pulling
research.
So
if
you're
following
each
Jim
school,
you
know
I
owe
everyone
spatial
pooling,
HTM
full
episodes
and
turns
out
at
this
facial
pulling
algorithm.
While
I
can
say
I
understand
it
at
this
point,
I
can
tell
it's
doing.
I
know
why
it
works.
I
know
how
it
works,
but
as
far
as
visualizations,
it's
very
difficult
to
come
up
with
an
intuitive
way
to
show
how
and
why
it
works,
especially
in
the
format
of
a
video
capacity
received.
A
So
to
get
to
get
what
I
want,
which
is
to
have
people
understand
how
to
speech
well
like
networkers
I'm,
something
like
there's
a
really
big
chance
that
the
HTM
going
to
release
next
Friday
won't
actually
be
about
spatial
pulling
because
I
want
another
couple
weeks
to
work
on
honestly.
This
is
hard
and
I
don't
want
to
put
out
a
subpar
episode
just
because
I
didn't
have
enough
time
to
research,
the
material.
A
So
so
there
may
not
be.
There
probably
will
not
be
a
special
polling
HDM
school
next
week,
but
there'll
be
something
and
then
two
weeks
after
that,
I'll
make
a
tentative
promise
that
you'll
have
episode
10
of
spatial
pooling,
there's
going
to
definitely
be
more
than
one
episode,
I'm
taking
three
or
four,
probably
just
on
spatial.
Fooling
there's
some
important
things
that
deserve
their
own
episodes
like
how
does,
in
addition
work.
How
does
boosting
work?
What
about
topography
I
haven't
even
turned
on
topography?
A
Everything
I've
done
so
far
has
just
been
with
one-dimensional
spatial
polling,
but
there's
this
whole
other
aspect
of
spatial
pulling
with
to
Agra
fee
that
might
actually
make
it
easier,
visualize
and
understand
our
works.
But
honestly,
we
hardly
ever
have
typography
turned
on
I'm,
seeing
topography,
I
mean
topology
right
anyway,
topology
yeah,
so
search
for
place
in
your
brain
history,
all
those
things
so
the
work
on
spatial
pulling
for
each
team
school
and
I
am
still
behind
on
ticket
review
if
it's
still
got
hundreds
and
hundreds
of
new
pic
tickets
to
review,
update,
triage,
etc.
A
A
Community
updates,
just
looking
at
HTM
forum,
see
if
there's
anything
interesting,
we
should
I
should
bring
up
talk
about
nab
a
lot
of
great
shatter
on
HTM
form
thanks
everybody
for
for
migrating
over
I'm
still
getting
some
good
help
instructions,
questions
and
how?
How
do
I
get
this
working?
Why
isn't
this
working
and
I'll
try
to
get
to
those
as
soon
as
I
can.
A
Right
so
I
think
that's
all
I
have
said
since
nobody
else's
this
slide
and
online.
This
was
a
very
fast
it
champ
hackers
hang
out,
but
thanks
for
watching
either
way
and
I
appreciate
it
and
I'll
do
this
in
another
month.
So
do
one
of
these
a
month
a
lot
of
four
more
people
from
the
community
able
to
join
in
and
take
care
and
I'll
see
you
on
HTM
forum
have
wonderful
Independence
Day
weekend,
if
you're
in
the
united
states,
otherwise
have
a
wonderful
global
weekend.