►
From YouTube: HTM Hackers' Hangout - Dec 1, 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 discussions will be about this much at least:
- theta oscillations, learning vs inference cycles
- nupic community fork for python 3 work?
- attitudes toward AI hype
Links:
- https://discourse.numenta.org/t/unsupervised-learning-and-storing-the-location-with-the-feature/3107/2
- https://discourse.numenta.org/t/should-we-make-nupic-python-3-compliant/1537/15 / https://discourse.numenta.org/t/what-is-a-community-fork/3112/2
- https://discourse.numenta.org/t/does-anybody-else-find-sophia-concerning/3109
A
To
HTM
hackers,
hangout,
that's
December,
first
on
that
Taylor
from
Numenta,
and
we
were
just
all
chatting
about
how
our
generation
is
going
to
be
the
last
to
die
and
that
everyone
else
will
be
immortal.
That's
a
completely
different
subject.
We
don't
need
to
talk
about
that
on
YouTube,
so
I
have
a
few
things.
I
want
to
talk
about
and
then
we'll
open
up
for
discussion.
Community
I
already
have
a
few
people
connected.
A
Let's
talk
about
data
oscillations
which
came
up
sort
of
unexpectedly
markets
are
glad
you're
here,
because
I
asked
Jeff
to
respond
to
something
that
was
about
allocation
and
he
responded
talk
about
the
plate,
oscillations
and
learning
versus
inference
cycles.
So
hey
Marcus
I'm
going
to
talk
about
that
sure.
B
That's
where
the
local
field
potential,
basically
measures.
So
if
you
measure
the
local
field,
potential
you'll
find
that
it
oscillates
it
makes
kind
of
a
wave.
It
goes
up
and
down
up
and
down
up
and
down
in
a
sort
of
systematic
way.
It's
not
just
random
and
you
can
think
of
each
of
these
oscillations.
Like
when
it
goes
up
down
and
then
up
down
and
repeats,
each
of
those
is
like
a
computational
epic,
it's
like
you,
you
might
call
the
time
step
it's.
B
It
is
one
full
time
step
of
like
of
like
some
cells
like
earlier
in
it
some
cells
spike
later.
But
you
have
this
oscillation
and
like
a
really
cool
thing,
is
that
a
given
cell
can
attend
to
I
mean
it
depends
on
the
cell
type,
but
many
many
types
of
cells
tend
to
fire
at
a
certain
part
of
the
oscillation,
like
some
of
them
will
tend
to
fire
earlier
in
it.
Some
of
them
will
tend
to
fire
later
in
it,
and
some
will
fire
in
multiple
times
I
mean
the
firing
rates
of
cells.
A
B
B
A
B
B
The
fact
that
there
are
these
multiple
kind
of
stages
of
the
time
step
is
thought-provoking
and
one
thing
that
has
been
studied.
I
wasn't
knowing
I
didn't
review
this
paper
myself,
so
I
can't
describe
it
in
too
much
detail.
But
what
has
been
found
is
that
in
the
theta
oscillation,
maybe
the
later
portion
of
it
I
guess
you
I
guess
it's
kind
of
arbitrary,
which
part
you
called
the
beginning
and
which
part
to
call
the
end.
B
I'm
gonna
do
this
the
M
part
it
there's
greater
plasticity
during
that
phase,
or
at
least
at
least
in
response
to
that
phase,
and
the
the
later
part
of
that
oscillation.
The
later
part
of
the
time
step
seems
to
drive
more
of
the
learning
that
occurs
and
and
what
make
what
this
cause
Jeff
to
think
of
was
so
in
the
temporal
memory.
B
B
A
B
One
more
time,
if
you're
learning,
multiple
sequel
with
shared
subsequences,
it
is
pretty
slow
at
learning,
because
it's
like
say
you
see
ABC
quite
frequently
in
the
middle
of
various
sequences.
So
later
you're
learning,
like
xx,
sequence,
you've
ever
learned
and
you
see
an
A.
You
burst
because
it's
your
first
time,
you've
seen
it
but
you've
seen
an
a
before
that
predicts,
B
and
then
you're
in
then
you
see
B
and
you're.
Like
oh
I
know
this.
This
is
ABC
and
the
problem
there
is
that
you
think
you
it's
like.
B
You
suddenly
treat
it
as
if
you
something
you've
seen
this
before.
But
what
would
be
nice
is
is
if
the
temporal
memory
could
sort
of
entertain
both
positive
possibilities
of
of
considering.
Maybe
this
is
something
I've
seen
before,
or
maybe
it's
something
novel
and
the
idea
of
having
separate
learning
and
inference
phases
is
that
the
maybe
the
third,
the
early
part
of
that
the
cycle
would
continue
on
with
the
predictions.
B
While
the
later
part
of
the
cycle
would
would
consider
the
idea
that
this
is
novel
so
that,
in
a
few
time,
steps
when
you
see
something
new,
it
hasn't
lost
all
context
and
they
can
continue
learning
the
sequence
without
having
to
like
come
back
and
repeat
that
sequence
again
and
again,
because
it
assumed
it
was
something
familiar.
I
did
a
regna
I'm,
sorry
about
that.
No.
A
A
A
So
the
second
thing,
so
that's
sort
of
our
research
update,
so
they've
got
a
research
of
data.
Second
thing:
I
wanted
to
talk
about
was
Python
three
and
new
pick,
so
we
wanted
well.
The
community
is
wanted.
Python
three
ended
for
a
lot
of
time:
everybody's
moving
to
Python,
3,
I
thought
2
is
meant
to
fit
only
to
be
sunset,
and
so
it
makes
sense
to
move
it.
A
The
problem
is
that
in
our
current
state,
that's
that's
a
totally
supporting
research
entirely
100%,
aside
from
you,
know,
me
nominated
guy,
and
so
we're
not
going
to
take
our
research
engineers
record.
Take
too
much
of
their
time
to
do
anything.
Honestly,
that
was
increasing
or
improving
the
effectiveness
points
in
maintenance
mode
personally,
the
way
it
is
so
the
community,
if
they
make
PR,
that
does
spend
a
lot
of
changes
and
testing
to
get
something.
A
Python,
3
compliant
I,
believe,
and
even
if
it
is
easy,
we
all
need
to
have
a
pipeline
as
soon
as
work
needs
to
be
done
through
it
and
I.
Don't
want
the
community
to
be
held
up
that
the
communities
are
willing
to
do
this
work
I,
don't
want
any
false
expectation,
so
confirmed
not
going
to
have
a
lot
of
time
to
review
the
PRS
with
it,
including
very
quickly
to
get
back
on
three
in
place.
A
So
I've
been
saying:
there's
an
option
to
create
a
community
fork,
that's
not
divergent
from
nupoc,
but
doesn't
have
the
commitment
fictions.
So
that's
always
an
option.
I
want
you
guys
just
to
know
that
and
if,
because
we're
not
going
to
be
moving
very
fast
on
funding
that
stuff
we're
moving
fast
right
now
in
research
for
abs
in
the
future,
that
will
change
and
we'll
come
back
and
and
try
to
put
together
more
production
api's.
A
A
Last
thing
this
because
I
think
this
is
an
interesting
conversation
was
a
AGI
height.
So
I,
don't
know
if
you
guys
read
on
the
forum,
but
there's
something
about
who
brought
this
stuff.
I
can't
remember
some
a
newer
form
number
but
he's
been
very
active
recently
concern
over
height
about
a
GI
which
I
which
I
can
share
the
sentiments,
because
if
all
the
Terminator
talk
and
everything
concerns
about
over-regulation,
it's
you
know.
I
still
think
a
lot
of
that
is
are
in
the
future.
I,
don't
think,
there's
much
of
a
near
near-term
threat.
A
However,
the
misrepresentation,
I
think,
is
a
problem.
So
whenever
we
looked
at
other
places
and
see,
we
think
that's
misrepresent
ating
rudeness,
representing
the
state
of
AGI
or
the
state
of
AI
I
think
we
should
also
look
at
ourselves
and
make
sure
we're
not
misrepresenting
of
what
what
the
state
of
our
progress.
So
it's
an
opportunity
to
reflect
and
the
main
thing
is
a
science.
A
You
know,
even
if
it's
theoretical
science
at
science
so
show
the
proof,
let's
get
benchmarks
in
place,
to
figure
out
how
we
test
things
and
compare
them
and
see
how
we
can
do
everything
better
in
life
of
each
other.
So
whenever
we
get
something,
that's
hype,
I
think
the
best
thing
that
we
can
do
is
scientists
and
discoverers
and
engineers
to
say.
Show
me
the
proof
and
then
and
let's
test
the
thing
and
the
general
public
is
going
to
figure
that
out.
You
know
the
proofs
in
the
pudding
and
they
can't
return
results.
A
C
B
C
Happening
at
this
stage,
like
I
know,
it's
gonna
happen
eventually
right,
you
know,
I,
think
robots
are
gonna,
become
at
the
level
of
people
right
that
at
least
intelligently,
but
prefer
to
happen
now.
I
think
it's
sort
of
a
misrepresentation
of
what
artificial,
general
intelligence
could
be
considered.
A
citizen
I
think,
maybe
you
know
at
the
point
of
Sofia
I,
feel
like
it's
more
of
a
cool
party
trick
than
a
really.
You
know
an
actual
well.
C
D
I
think
AI.
Can
you
hear
me
yep,
okay,
I.
Think
AI
is
the
next
computer
chip.
It's
the
next
Internet.
It's
the
next
come
question
engine
like
we're,
creating
something
amazing
and
wonderful
that
everyone
will
love
once
it's
out,
but
indeed
there's
something
like
you
can
say
the
internet.
There
are
lame
things
about
the
internet.
That
would
go
away
if
we
got
rid
of
the
internet,
but
it's
not
worth
getting
rid
of
same
thing
with
microchips,
same
thing
with
combustion
engines,
so
I
I
think
once
everyone
sees
how
great
the
world
is.
A
Let's,
let's
hope
so
I
think
all
technology
brings
this
the
same,
similar
things
to
the
table,
good
things
and
bad
things
and
the
bad
people
are
gonna,
exploit
the
bad
things,
no
matter
what
so
hey,
there's
someone
just
popped
up:
hi
James!
You
can
see
you
don't
know
if
you
can
talk,
though
you're
just
taking
community
comments.
If
you
want
to
get
anything
there,
they
doing
a
bad
thing.
A
Okay,
let's
let's
wrap
it
up.
Oh
I
just
released
a
video
this
morning.
If
you
didn't
see
it,
it's
a
what
is
it
dendritic
spike
I
was
hoping
to
like
boil
down
like
that.
This
is
like
the
core
of
HTM.
Theory
is
like
the
definition
of
an
HTM
they're,
not
so,
and
I
honestly
never
found
a
way
to
communicate
that
very
well.
So,
like
two
minutes,
you
can
learn
the
chord
HJ.
If
you
watch
that
video.
D
That's
it
anybody
else.
Oh
wow
I,
just
one
more
thing
to
say:
I'm
working
on
my
project
about
six
hours
a
day.
My
goal
is
to
have
it
something
to
show
you
guys
that
at
Valentine's
Day,
because
it's
it's
so
it's
so
much.
It's
gonna
take
a
long
time.
Man,
that's
probably
too
optimistic,
but
my
goal
is
that
I
will
have
a
Valentine's
present
for
everyone
here
and
it
will
be
my
presentations
on
the
amazing
things.
I'm
working
on
I,
don't
just
say.
A
C
Let's
follow
anybody
else,
I
think
I
just
wanted
to
possibly
ask
Marcus
yeah
yeah
I
heard
you
know
the
theta
oscillations
you
were
talking
about.
I
just
got
thinking.
I
think
you
know
the
way,
I
think
of
learning
in
an
Iran,
it's
more
of
I.
Guess
it's
not
initially.
I
got
more
time
step
based
thing,
but
it's
actually
like
the
back
propagation
of
the
action
potential
in
the
neurons
sort
of
causes.
C
B
C
Yeah
yeah
I'm,
saying
I'm,
basically
like
considering
the
idea
that
you
know
they'd
oscillations
sort
of
say
that
the
learning
happens
at
the
end
of
the
time,
step
right
or
more
towards
the
end,
and
then
I
guess
contrasting
that
with
the
idea
that
in
a
neuron
you
know
there's
this
sort
of
back
propagating
action
potential
that
goes
into
the
dendrites
and
sort
of
maybe
affects
them
in
terms
of
growing.
Would
you
say
I
guess
one
is
learning
at
the
small
scale,
maybe
one's
learning
at
the
large
scale
or
what's
the
relation
between
those
two
ideas?
B
It's
my
room:
okay,
that's
her
I,
think
yeah
and
it
was.
It
was
weird
like
to
me
hearing
that
that
there's
more
learning
associated
with
a
certain
part
of
the
the
yeah
oscillation
was
like
it
kind
of
challenged,
my
idea
of
like
what
that
what
it
I
need
to
understand
more
about
that,
because
it
was
someone
else
so.
B
C
Think
I
also
one
other
little
comment,
but
we
can
talk
about
it
later.
Maybe
if
you
want,
but
it's
a
I,
think
I
I
recently
looked
at
the
presentation
during
the
the
meet
up
a
little
while
ago,
and
your
your
presentation
of
how
you
how
to
solve
the
allocentric
location
signal,
I,
I,
think
I,
know
the
gist
of
your
your
solution
and
actually
I
I
feel
like
we're
actually
getting
at
something
similar
between
us,
as
in
there
actually
they're
functionally
identical
I'll,
probably
go
over
later.