►
From YouTube: HTM Chat TIme - Forum Q&A and discussion
Description
Broadcasted live on Twitch -- Watch live at https://www.twitch.tv/rhyolight_
A
Let's
see,
of
course,
is
it
actually
playing
it.
Just
I
can't
play
that.
Okay,
my
wife
is
listening
to
this,
but
I
have
a
plan
B
hold
on
I,
have
a
plan
B
music
for
programming?
That's
what
I'm
gonna
go
with
you
guys
any
programmers
out
there.
If
you
like
electronica-
and
you
like
something
to
just
set
your
mind
at
ease.
This
usually
does
it
for
me.
A
A
A
We're
gonna
do
forum,
Q&A,
we're
gonna,
do
building
HTM
systems,
there's
going
to
be
a
research
meeting,
I
think
and
then
we're
gonna
continue.
Building
ishtam
systems
because
I
I
haven't
even
asked
if
there's
really
gonna
be
a
research
meeting.
So
I'm,
not
even
sure,
is
this
thing
on
just
thing
on
you
guys
out
there.
This
chat
working.
Somebody
chat
me
up
something
say:
hello:
livestream
live
now,
okay,
thought
it
through
it
and
discord.
I
am
throwing
it
up
into
the
forum.
Oh
crap,
sorry,
I'm
sure
you
guys
were
really
enjoying
that.
A
That's
the
bad
thing
about
this.
You
can't
you
can't
continue
where
you
left
off.
So
this
is
still
music
for
programming.
It's
just
a
different
album
I
started
with
the
first
one.
This
is
pure
aided
by
data
cassette.
You
can
find
him
on
Twitter
he's
he
does
some.
He
acure
puts
together
some
of
these
playlists
and
other
people
put
together.
Others
anyhow.
A
Sir,
there
we
go,
that's
what
I'm
doing
a
casual
work
session.
Ask
me
anything.
We're
answering
form
questions
generally
doing
whatever,
then
we're
gonna
go
into
building
HTM
systems
as
soon
as
we
get
through
this
we're
gonna
start
building
HTM
systems,
so
there's
sort
of
gonna
be
phases
of
the
stream.
Today,
it's
Jim
chat,
building
a
stream
systems,
probably
research
meeting.
It
reminds
me
I'm,
gonna,
I'm
gonna
have
to
check
in
at
the
office
real
quick,
so
you're
gonna
hear
me.
A
I'm
gonna
turn
this
all
the
way
down
and
I'm
gonna
go
check
in
with
Jeff
and
make
sure
there's
actually
research
meeting
today.
So
that
means
from
this
on
so
in
case,
I
shows
something
I
don't
want
to
show
I,
don't
want
to
show
Jeff
or
show
you
were
his
words
voice
or
image.
Without
his
explicit
permission,
hey
Jeff
is
there
research.
A
A
A
To
the
forum,
all
this.
A
This
is
a
the
other
repeat,
we're
talking
about
the
repeating
inputs
problem,
a
lot
lately
and
that's
great,
because
it's
been
hard
to
like
explain
and
now
a
lot
more
people
understand
what
the
problem
is.
I,
think
and
why
the
repeating
inputs
is
an
issue
if
you're,
if
you
don't
have
something,
that's
explicitly
closing
off
sequences.
Now
there
can
be
an
issue
or,
like
you,
have
to
have
something
to
direct
your
attention
to
something
Alice,
a
basically
I
say:
stop
building
this
sequence,
and
and
now
the
input
is
building
this
sequence.
A
A
If
you
execute
a
movement
through
an
object
space,
and
so
that
movement
of
execution
is
like
a
coordination
between
the
the
object
space,
that's
represented
in
your
cortex
and
then
like
the
movement
space,
which
I
think
is
represented
both
in
your
cortex
and
most
likely
in
other
parts
of
your
brain
too,
but
I
think
there's
a
representation
of
both
in
the
cortex.
Although
the
the
cortex
doesn't
control
the
motor
necessarily,
you
know,
the
the
lizard
brain
has
ultimate
control
over
those
things.
It
has
input
to
it.
A
A
A
Htm'
time
epic
encoder,
with
the
current
time
field,
set
staying
the
same
with
a
new
car.
I
think
this
is
so
exactly
what
we're
talking
about
right.
I
mean
like
framing
the
sequence.
Essentially
you
have
to
frame
the
sequence
yeah,
that's
the
same
thing
that
I'm
that
I
was
just
blathering
about
I,
think
so
I
think
we're
on
the
same
page
mark.
We
have
some
HDM
way
of
establishing
a
circular
repetitive
task
with
online
learning.
This
should
be
an
important
research
line
as
as
common
tasks
in
the
brain
that
central
pattern
generator
comes
to
mind.
A
A
Excuse
me:
okay,
they
are
a
source
of
tightly
coupled
patterns
of
neural
activity
that
drive
rhythmic
motions
like
walking,
breathing
or
chewing.
The
ability
to
function
without
input
from
higher
brain
areas
still
requires
modulatory
inputs,
the
ability
to
function
without
input
from
higher
brain
areas,
I'm
assuming
higher
brain
areas
as
cortex
still
requires
modulatory
inputs.
So
that's
and
their
outputs
are
not
fixed.
It
doesn't
explain
where
the
modular
inputs
are
well,
some
modular
inputs,
okay,.
A
A
B
A
A
So
I
think
we
have
to
think
a
lot
more
about
how
we're
gonna
deal
with
movement
like
the
interaction
of
self
with
reality
like
I,
think,
that's
I,
don't
know,
I,
don't
know
the
answer
to
this
question
pooling,
and
this
is
what
I
get
when
I
plot
all
the
previous
things
with
PI
implementation
of
my
stare
across
far
as
you
suggested.
This
might
be
a
related
issue.
Okay,
in
summary,
with
PI
implementation
prediction,
confidence
behaves
as
expected,
ie
generally
having
a
hundred
percent
confidence
in
predicting
a
very
effective
signal.
A
A
A
Taking
the
time
and
effort,
listen
I've
been
a
part
of
open
source
communities
and
it
takes
a
lot
of
commitment
to
like
do
something
like
this,
like
what
Gulen
has
done
here,
just
to
get
the
software
running
and
start
playing
around
with
it
and
like
this
guy's
put
in
a
lot
of
time.
Thank
you
for
doing
I'm.
Just
saying
thanks.
Thank
you.
A
I'm
not
sure
what
to
make
of
this
honestly,
the
the
SDR
classifier
from
an
HTM
standpoint
is
not
interesting.
I
hate
saying
that,
but
it's
just
it's
not
something.
I
know
it's
not
something
that
new
mento
will
put
effort
into
figuring
out
because
it's
outside
of
the
realm
of
HTM
theory.
It's
a
it's
a
typical
classifier.
So
if
somebody
wants
to
go
in
and
look
and
figure
this
out,
what's
going
on
in
there
you're
welcome
to
do
it.
A
A
B
A
A
A
So
he's
just
gonna
deploy
it,
he
already
deployed
it.
It's
words:
demystifying
over
parameterization
and
deep
learning,
Shawn
usually
posts
his
deep
learning
stuff
he's
watching
this
video
about
overpour
amortization
of
neural
networks.
I'm
not
gonna,
watch
this
right
now,
I
just
want
to
see
who
it
is:
Mon,
DS,
Seoul
tunnel,
I
can't
Brown
soon.
A
A
A
B
A
We're
gonna
move
on
to
the
BH
TMS,
soon
I
think
or
as
they
say,
although
though
it
would
wouldn't
it
require
knowing
the
sequences
or
their
links
ahead
of
time
like
yeah
like
how
does
it
know,
to
treat
ABCDE
and
exbc?
Why,
as
separate
sequences
in
a
real
streaming
scenario,
we
won't
inherently
know
and
one
sequence
ends
and
another
begins
less
there's
a
mechanism
that
determines
that
yeah,
so
I
think
in
a
real
streaming
scenario,
you
sort
of
get
your
sequences
are
tied
to
movements.
A
That's
my
this
was
my
Twitter
and
I,
just
instinctively
open
Twitter.
Sometimes,
oh
there
you
are
mark
okay,
you
could
do
it
right
and
bring
in
the
subcortical
drivers
of
these
repetitive
actions.
I
would
be
I
would
be
a
fan
of
it,
but
it
does
seem
like
a
long
way
from
where
we're
at
at
the
moment,
adding
a
small
driver
to
abstract
the
function
for
now
seems
like
a
reasonable
compromise.
Fair
enough.