►
From YouTube: ONNX Edge WG meeting 20190515
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
B
B
C
B
B
C
B
B
B
A
B
Scope
and
really
profile
attributes
and
what
that
means.
So
we
can
use
that
later
on
to
define
actual
profiles.
Right,
yes
and
then
you
said
you
know
sometime
and
we
agreed
that
the
next
meeting
to
me
would
be
the
good
time
to
have.
You
know
that
draft
available
and
it
can
be,
we
can
review
it
the
for
this
meaning.
You
know
last
time,
I
said
an
offer
suggest
like.
Maybe
we
can
use
this
meaning
then
to
brainstorm
a
little
bit
content
of
this
document
and
what
yeah
covered?
What
really
should
be
covered?
B
If
that
is
not
clear,
so
that
is
basically
the
the
agenda
which
I
put
up
on
a
guitar
yesterday,
which
says
which
I
made
a
working
title
I,
don't
know
what
is
the
title
of
the
document,
but
you
know
I'd
scope
and
profile
definition,
and-
and
that's
that's
basically
that
so
you
know
we.
We
should
perhaps
consider
the
content
and
information
that
you
guys
captured
in
in
your
in
your
document
that
was
put
in
discussion
materials.
B
C
B
Okay,
so
I
mean
I,
don't
think
there
was.
There
was
expectation
actually
to
share
an
ending
in
this
meaning
from
you.
This
was
mainly
put
in
place
as
a
discussion
meeting
about
the
the
the
actual
or
brainstorming
me
meeting
about
what
actually
should-
or
maybe
you
know,
go
into
this
document
if
you're
talking
about
it,
maybe
you
had
some
ideas
or
you
would
like
to
share
or
ask
or
discuss
something
in
that
context,
so
to
help
you
put
this
document
together,
right,
yeah,
yeah,
and
so
that
is
basically
what
we
have
for
today.
B
Discussion,
discussion
on
this
topic
and
so
I'm
sharing
the
screen
and
the
the
one
that
that
I'm,
referring
to
that
you
guys
put
in,
will
be
put
in
discussion
materials.
Is
this
one
right,
and
so
there
is
there
is
this
document
is,
is
has
some
overlap
with
I
think
which,
with
what
we
want
to
put
in
this
next
document,
which
is
I
called,
roughly
speaking,
edge,
scope
and
profile
definition,
and
by
by
going
into
the
you
know
our
statement
of
work
by
the
way?
B
Is
that
so
I
just
put
I'm
thinking
like
I'm,
not
sure
if
this
is
good
name,
but
this
is
the
folder
very
put
documents
that
we
record
of
work
on
and
together,
and
we
really
kind
of
approve
when
we
may
not
make
them
play
no
record.
So
this
artifacts
folder
is
then
to
be
used
for
that
for
documents
that
we
as
a
group,
as
the
and
I
say
I
say
not
write,
output,
output
of
this
working
group,
Isaac
yeah.
A
B
Statement
yeah
like
something
we
produced
right,
yes,
and
then
you
know
not
to
say
that
it's
not
possible
to
change,
but
it
is
just
something
that
is
in
the
mature
state.
Yes,
yes,
and-
and
so
this
this
is
basically
transcribed.
Transcriptional
are
documents
reviewed
so
far,
and
here
in
this
we
said
that
these
two
dot
these
two
documents
here
we
could
combine
to
one
and.
C
A
B
Yes
right
that
right
here,
yes,
so
this
is
really
where
what
would
be
kind
of
exactly
I
agree.
This
is
the
kind
of
maybe
good
starting
point
as
discussion
for
for
the
for
what
this
document
should
capture
in
85,
you
have
three
back
of
the
of
the
edge
profile
and
and
then
you
know,
decide.
Okay,
these
are
four
or
five
attributes.
We
be
it's
like
dimensions.
B
We
would
like
to
use
as
to
categorize
edge
profiles
or
educator
is,
and
then
we
use
that
as
a
there's
a
you
know,
key
definition
space
when
we
decide
on
the
actual
profiles
like
like.
Is
it
like
you
know,
deeply
embedded
profile
is
like
infrastructure
edge
profile?
Is
it
like
phone
phone
device
profile?
I
mean
things
like
that
I'm
just
putting
things
up,
I
mean
I'm,
not
sure
what
we
are
gonna
come
up
for
it,
but
basically,
yes,
I,
agree.
I,
think
this
is
the
good
good
way
to
start
looking
to
it
guys.
B
Yeah
we
can
give
it
you
see
the
screen
on
your
phone.
I
can
see
it
hold
on
yeah.
Okay,
yes,
so
I'm
sure
did
you.
You
know
yeah,
so
yeah.
You
don't
mention
this.
This
fine,
basically
from
our
last
workshop
as
a
basic
starting
point
to
define
to
work
on
this
as
a
content
for
this
document.
We
are
discussing
right,
so
I
think
I
agree
that
this
get
exactly
what
we
would
like
to
do.
So
maybe
we
should
brainstorm.
Are
these
good
I
mean
we
just
put
bullet
points?
B
I,
don't
know
it
was
something
that
I
don't
know.
Is
it
like
too
many?
Is
it?
Is
it
just
enough
these
these
a
lot
here?
We
have
five
five
dimensions.
Basically,
but
let
you
have
more
than
five,
you
have
yes,
so
the
point
being
this
is,
was
you
mean
meant
to
be,
as
example,
but
I
guess
we
had
here
we
have
to
decide
what
are
the
good
attributes
to
define
the
edge
profile
and
I
clearly
think
that
compute
capabilities
were
one
memory
footprint
and
memory
size
is
another
and
and-
and
so
you
know
open
it.
B
C
B
The
reason
I
think
we
put
this
was
data
locality
right
and
that
is
connected
with
the
security
and
privacy,
and
so
you
could,
if
you
think,
about
the
edge
as
a
as
a
do.
You
know
profile
what
what,
if
the
edge
profile,
one
category
average,
is
actually
the
infrastructure
edge.
By
that
what
I
mean
it's,
it's
a
it's
on
other,
it's
not
on
device
itself.
It's
on
the
you
know
network
edge,
and
that
means
that
data
needs
to
travel
from
device
to
the
that
net
for
catch.
So
it's
what
kind
of
fork
consider
it
fork.
C
B
B
It's
just
there
close
to
the
to
the
link
and
I
think
this
is
the
space
that
is
potentially,
which
is
one
of
the
possible
edge
profiles,
because
of
especially
we
did
with
advances
in
5g
and
everything
in
the
lower
agencies
and
stuff,
like
that.
I
think
the
the
latency
constraints
and
in
terms
of
real-time
processing
and
whatnot
are
going
to
be
lesser,
and
so
then
you
could
consider
that
it
to
be
one
edge
profile,
which
is
like
I
call
infrastructure.
B
B
A
Come
College
probably
now
all
of
these
talks
about
hacking,
cars
and
things
like
that
cause
definitely
need
to
be
very
much
secured
and
protected
home
externally
attempts,
and
even
in
its
crooked,
even
to
the
elections,
One
Direction
is
trying
to
steal
the
data
to
have
to
get
the
data
turns.
The
other
direction
is
trying
to
modify
the
data
to
make
the
curl
do
something
that
it
was
not
supposed
to
do
alright,
so
even
in
indicate
that
they
Megan
presented.
If
you
have
hybrid
scenario
of
processing,
some
of
the
processing
happens
in
the
car.
A
B
B
To
consider
I
feel
me
feel
I
would
say
if
the
device
is
deeply
embedded
device,
like
some
Sun
sensor,
that
it
does
all
the
processing
locally
right
yeah.
That,
then,
is
maybe
the
instead
of
calling
it
security
or
privacy.
Maybe
we
should
call
it
the
data
locality.
I,
don't
know
that
is.
That
is
if,
if
the
data
leaves
the
device
and
comes
back,
then
that
is
something
that
is
defining
one
of
the
defining
attributes
of
the
profile
and
I.
A
Think
the
point
here
is
how
maybe
to
peddle
exercise
why
we
think
it's
important:
how
does
that
impact
a
device
that
is
supposed
to
run
a
onyx
modems,
so
one
possible
aspect?
Is
it
modified
the
model
that
the
device
needs
one?
Because
in
a
hybrid
scenario,
you
are
in
the
different
model
compared
to
a
standard,
almost.
B
A
C
D
C
D
C
B
Yeah,
so
just
click
quick
refresh
for
Sarah,
because
he
missed
couple
last
meetings,
so
basically
we
did
approve
and
still
work.
We
are
now
discussing
the
next
delivery
from
this
working
group.
You
know
work
on
and
your
dong
is
gonna
put
proposal
on
that
in
that
topic,
it's
about
defining
the
edge
scope
and
edge
profile
attributes
and
what
defines
the
edge
profile
so
get
on
then
is
pratap.
B
B
You
know
as
something
if
you
kind
of
discuss
and
then
you
don't
go,
then
you
use
this
discussion,
as
maybe
input
for
his
document.
He's
gonna
propose
and
so
it's
actually
their
deaths.
The
main
thing:
do
you
mean
you
know,
thoughts
are.
Do
we
need
to
leave
the
security
privacy
nor
not
into
this
mix,
I'm,
not
sure
what
you
think
about
it
sorrow.
D
B
A
D
A
That
kind
of
why
we
believe
offers
a
very
important
part
of
the
reason
and
connecting
it
now
today,
security
thing,
so
security,
Oh,
more
important
privacy
is
dictating
different
kind
of
models
focused
on
the
device,
and
this
is
connected
water
to
what
you
said
hang
on.
In
one
case,
the
entire
processing
can
happen
on
the
device
because
of
issues
of
privacy
in
another
case,
another
go
find
other
type
of
device.
The
processing
happens,
half
of
it
on
the
device
half
of
it
on
the
edge,
and
that
dictates
a
different
type
of
model.
A
B
B
And
or
training
the
or
whatever,
maybe
there's
a
some
training
happening
use
case.
Maybe
a
mother
was
updated
from
device
now
periodically
or
you
know
so.
Basically
that
ask
that
that
that
aspect
is
about
their
being
on
device
leaving
device
or
not
in.
However,
whoever
was
to
interpret.
That
is
what
I
don't
maybe
care
it's
about.
The
locality.
B
And
you
say
these
are
the
profiles
that
we
define
as
reporting
for
the
edge
four
five,
whatever
we
come
up
with
ten
and
and
then
this
these
are
the
profiles
that
should
have
should
be
backed
with
certain
set
of
one
or
more
use
cases
and
and
models.
So
then,
that
tile
that
dives
down
into
the
okay,
so
we
have
profile.
We
have
the
use
cases
behind
these
profiles
and
then
we
have
models
associated
with
these
profiles,
and
then
this
is
like
path
to
defining
that
black
compliance
of
the
of
the
edge
devices
right.
B
So
somebody
defines
the
develop,
develops.
The
specific
use
case
with
their
edge
device-
then
it
has
it-
should
fit
in
one
of
these
profiles.
It
should
oh,
this
is
my
my
I
am
in
this
category
right
and
and
I,
and
you
know
somebody
ultimately
could
claim
if
I
pass
compliance
based
on.
Ultimately
what
we
want
to
define.
It's
like
set
of
use
cases,
models
and
test
criteria
that
somebody
can
say:
okay,
I,
am
passing
this.
Compliance
for
the
infrastructure,
edge
profile
and
I
make.
D
B
I'm
certified
right
for
that
and
it's
just
defining
good
like
maybe
maybe
we
don't,
we
won't
know
for
sure
before
we
start
defining
the
actual
profiles
and
I
might
be
that
defining
specific
attributes
upfront
without
going
through
exercise
of
the
use
cases
and
an
edge
profiles,
then
maybe
we'll
come
up
with
with
more
attributes
or
less
attributes,
maybe
a
fee.
If
you
can,
you
know,
maybe
we
have
to
have
a
starting
point:
working
draft
okay,
these
are
the
attributes
that
we
wanna
cover
as
important
and
then
go
to
the
next
step,
which
is
defining
profiles.
B
B
B
A
I'm
saying
that
connected
is
actually,
he
did
again
how's
that
effect.
What
we
want
to
ratify
is
the
processing
of
the
model
if
the
processing
is
polling.
Okay,
Oh
disco
bit
between
the
device
and
something
else
I
don't
know
if
connected-
is
affecting
other
aspect,
but
we
care
about
again.
Of
course,
we
can
factor
got
other
thing.
Okay,
what.
B
B
A
Okay,
maybe
maybe
suggestion
to
kind
of
move
forward,
is
to
bring
exercise
of
describing
what
window
with
the
exist
at
the
toilet
appeal
or
the
appeal
in
this
side,
and
we
agree
on
just
to
see
that
we
say
we
describe
a
form,
a
mobile
phone.
We
describe
it
using
the
DS
and
we
see
whether
that
sufficient.
D
A
B
Yeah,
so
this
I
agree,
so
it
seems
like
maybe
in
this
document.
It
should
be
some
some
test
test.
Examples
right
for,
as
use
case
is
that
that
we
can
see.
Can
we
describe
all
of
them
that
we
that
we
using
these
attributes?
We
have
that's
why
I
thought
eventually,
the
the
defining
these
attributes
potentially
could
change
over
time,
as
we
start,
defining
more
more
detailed
use
cases
or
solely
the
best
profiles,
and
so
I
agree
I.
Think
that
to
me
it's
basically
what
we
have
discussed
today.
B
D
B
So
that's
great!
Thank
you.
Thank
you.
You
don't
so
the
next
one
will
be
in
two
weeks
and
by
the
way
our
goal
is
to
define
this
this
and
scope
and
edge
profile
document
each
profile
definition
document
it
make
it
like.
People
are
paid
plan
or
plan
a
record
or
make
it
into
artifacts
in
this
for
disk
water,
meaning
by
the
end
of
June,
and
basically
we
have
two
three
meetings:
two
or
maybe
three
meetings.
B
If
we
go
by
weekly,
so
I
think
if
there
is
a
time
we
can
do
that
fine
and
if
need
be,
we
could,
if
discussion,
gets
more
kind
of
engaged
or
continuous
in
a
sense
like
there
is
a
cohesion
of
opinions
and
we
need
to
discussed
more.
We
can.
We
can
obviously
introduce
in
between
meetings
if
you
want
to
make
progress,
but
you
know
definitely
the
goal
is
that
we
want
to
define
this
completeness
document
in
this
this
quarter.
B
B
B
Okay,
we
are
just
saying
that
we
don't
think
we
don't
necessarily
need
to
spend
burn
food
for
our
and
weaker.
It
seems
like
we
did.
We
discussed
what
we
wanted
to
discuss
seems
like
we
could
skirt.
You
know
conclude
the
meeting
right
now,
unless
somebody
else
wants
to
bring
up
something
else,
so
that
is
the
last
call
for
discussion
topic
and
then,
if
not
then
I'm
gonna,
you
know
conclude
the
meeting
nothing
for
me:
okay,
okay,
good,
so
we
are
planning
to
do
review
this
document
draft
next
next
meeting
sounds
good
yep.