►
From YouTube: IETF109-NMRG-20201119-0900
Description
NMRG meeting session at IETF109
2020/11/19 0900
https://datatracker.ietf.org/meeting/109/proceedings/
A
C
A
D
E
E
D
F
F
Okay,
perfect,
perfect,
so
hello,
hello,
everyone
again.
F
Okay,
so
my
name
is.
F
Okay,
let
me
just
just
to
signal
you
just
to
change
a
bit
just
like
sorry,
sorry,
for
that.
F
Are
you
sorry
for
that?
Just
just.
A
Just
while
jerome
is
fetching
the
right
slides
to
inform
you
that
it
seems
there
is
an
ongoing
problem
with
the
the
chat
I
mean
the
so,
the
chat
is
not
operative
so
far.
So
that's
why
we
have
empty
empty
chat
windows
in
mitako.
Okay,
so
I
think
this
is
being
dragged
down,
but
I
have
no
more
info
for
a
moment.
If
you,
I
don't
know
if
it
will
be
restored
during
the
meeting,
but
in
case
you
you
would
like
to
speak.
A
You
have
the
raised
end
icon
on
the
top
of
the
window
and
you
can
request.
I
mean
a
slot
to
speak.
Okay.
Thank
you.
F
F
You
can
use
a
formal
one
just
to
okay,
so
remind
you
that
we
are
also
following
not
all
of
the
iutf.
So
if
you
have
any
interior
per
party
concern
you
you
have
to
disclose
as
as
basically
to
avoid
any
issue
then
important.
So
we
are
also
following
a
code
of
conduct
that
you
have
the
link
if
you
want
to
to
check
the
full
details,
but
basically
be
respectful
and
know
that
the
station
is
also
recording.
So
all
what
is
be
presented
and
what
said
will
be
recorded.
F
So
I've
already
said
that
we
are
research
groups,
so
I'm
not
clicking
on
that.
So,
okay,
this
is
a
virtual
meeting,
so
we
don't
have
blue
shield.
This
is
automatically
recorded
by
the
antennas
attendance
list
and,
of
course,
when
you
don't
speak,
please
keep
your
audio
muted
to
avoid
any
noise.
When
others
are
speaking,
so
you
have
the
different
link
so
as
just
want
to
say
we
have
some
some
prime
for
the
jabber
and
for
the
chat
now.
F
Okay,
so,
as
you
may
know,
there
is
also
gather
tool
to
have
more
social
interaction,
try
to
have
more
social
interaction,
as
we
have
an
face-to-face
meeting,
so
we
will
be
there.
I
mean
the
chair
me
and
laura
around
after
the
meeting.
Also,
if
you
want
to
continue
some
discussions,
so
we
can
continue
also
with
other
process
participants.
So
normally,
when
you
will
leave
the
meeting
room
after
the
meeting,
you
should
be
proposed
to
enter
into
the
in
together,
and
so
we
can
continue
and
try
to
to
have
more.
F
Let's
say:
haddad
discussion
there.
F
Okay,
so
here
is
our
agenda
for
today,
so
we
we
are
knowing
so
briefly
doing
some
some
some
information
about
the
group
and
then
we
enter
into
three
different
parts.
The
first
part
is
about
ibm,
so
we'll
have
four
drugs
that
will
be
presented.
We
originally
plan
to
have
a
discussion
on
use
case.
Unfortunately,
it's
it's
cancelled,
but
we'll
give
a
brief
brief
news
at
the
end
of
the
slides,
then
we
have
two
parts
about
the.
F
F
So,
just
to
give
you
some
news,
so
okay,
future
meetings,
so
so
we'll
continue
mostly
with,
of
course,
the
online
meeting
so
virtual
meeting
next
one
will
be
in
january.
We
don't
think
it's
it's
already
available
to
have
in
december,
but
in
january
we
will
have
the
next
vehicle
meeting
and
so
for
the
next
iedf
meeting
it
will
be
so
in
march
and
maybe
in
may
2021
we
tried
to
have
an
entire
meeting
collocated
with
ieee.
F
I
am
conference,
so
it's
not
decided
yet
it
will
be
free
virtual,
this
conference
or
every
so
we
don't
know
yet.
Hopefully,
maybe
we
can
partially
move
there
when
we
we
will
try
to
have
it
online
and
to
be
collocated
with.
I
am
too,
although
the
idea
of
being
collected,
the
iem
is
also
to
other
people
participating
to
this
question
during
the
concurrence
so
yeah.
So
so
we
will
continue
relying
on
online
meeting
or
we
try
to
be
monthly
on
buy
monthly
meeting
depending
a
bit
on
the
situation.
F
F
F
Okay,
sorry,
just
as
I
mentioned
before,
a
brief,
a
brief
overview
of
the
the
discussion
regarding
use
cases.
There
was
a
presentation,
the
last
interim
meeting
and
so
jefferson
is
leading
a
business,
this
action
item
and
this
we
proposed
a
template
on
the
github
platform.
F
We
have
a
list
of
proposed
scales
now
that
are
just
summarized
here,
and
there
are
some
of
course,
pending
questions
that
we
have
not
yet
answered.
Basically,
one
is
what
will
be
the
outcome
of
the
use
case
for
what
it
will
be
used.
We
are
thinking,
of
course,
it
can
be
used
to
help
in
elaborating
working
another.
F
The
end
of
the
group
like
the
architecture.
What
are
what
would
be
the
learning
of
the
different
use
case?
One
question
is
also:
should
we
keep
the
documentative
use
case
are
separate.
I
try
to
have
a
single
document
so
currently
it
may
be
that
you
have
the
feeling
that
people
would
prefer
to
have
separate
at
that
time
now,
because
maybe
in
the
in
the
beginning
and
but
anyway,
yes,
our
discussion,
just
we
should
have
template
and
we
should
have
gained
a
common
level
of
description.
F
So
here
it
was
just
a
brief
introduction.
So
if
you,
if
you
have
any
question
before
we
start
with
the
with
the
presentation
of
the
different
ibm
draft,
please
let
us
let
us
know.
F
Okay,
so
I
think
we
can,
we
can
start
with
with
the
next
attempt.
So
laura
is,
is
your
turn
with
the
presentation
of
studies
of
concept
and
definition,
draft.
A
A
We
can
see
okay,
so
I
I
tried
to
be
quick,
because
this
is
just
an
update,
so
status
on
intent-based
networking
concept
and
definitions
draft.
A
This
is
a
research
group
document
and
I'm
presenting
on
behalf
of
the
co-authors,
alex
lissandranja,
so
recent
history
just
to
situate
a
bit
the
purpose
of
this
draft,
so
the
goal
is
to
contribute
towards
a
common
and
shared
understanding
of
idea
in
terms
concept
and
functionality,
which
is
why
it's
called
concept
and
definitions
this
this
work
has
been
around
for
several
years
now,
so
we
had
version
2
of
the
research
group
document
posted
mid-september.
A
The
main
updates
in
this
revision
was
to
address
a
review
provided
by
ali
razaki
earlier,
and
so
we
provided
changes
according
to
the
review,
we
also
provided
a
more
detailed
description
of
ibm
functionality,
which
is
on
section
six,
with
new
subsections
dedicated
to
intent,
fulfillment
and
internet
assurance.
A
Some
redefinition
of
I
mean
clarification
of
the
distinction
from
policy-based
management
added
a
couple
of
examples
of
intents.
This
was
requested
several
times
by
some
comments
and
we
also
have
a
towards
the
end
of
the
document,
kind
of
hanging
items
on
discussions
or
future
work,
and
so
it's
renamed
for
the
moment
into
additional
consideration
and
we
made
a
various
editorial
improvements
throughout
the
document.
A
Since
then,
there
have
been
a
research
group
last
call
which
jerome
handled
it
was
started,
beginning
october
and
last
for
the
whole
month
of
october
jerome,
which
is
also
the
the
shepherd
from
this.
For
this
document.
A
Mentioning
his
conclusions
mid-november
a
couple
of
weeks
from
now
and
so
forth,
for
what
will
be
the
next
step,
just
a
very
high
level
outcome
of
this
research
class
called
for
this
document.
We
we
got
very
large
support,
positive
expression
of
support.
A
I
contained
a
bit
more
than
18
different
sources
in
the
research
group,
what
sequence
that
expressed
that
the
document
is
ready
to
move
on,
and
we
also
received
in
parallel
to
that
some,
I
will
say
more
detailed
reviews
beyond
just
the
support,
and
so
we
thank
all
the
researcher
participants
for
this
support
and
especially
for
the
reviews.
A
So
this
is
the
status
so
far,
and
this
is
the
last
slide
next
step
what
we
will
do
so
first,
we
are
working
on
the
new
version
which
would
be
zero.
Three.
This
should
be
ready
in
the
next
couple
of
weeks.
We
are
a
bit
busy
with
the
atf
so
far,
but
after
that
we
will
go
full
speed
on
this,
and
the
main
focus
of
this
revision
will
be,
of
course,
to
address
the
comments
and
the
reviews
received
during
the
last
call.
A
There
are
different
levels
of
comments
and
reviews,
so
we
have
still
to
figure
out
the
impact
of
those
comments
and
reviews
and
the
changes
we
will
need
to
to
implement.
But
this
is
our
work
first
step
and
then
once
this
version
is
ready.
A
A
On
the
other
hand,
if
to
address
the
reviews,
we
need
to
change
a
bit
more
text
refactor
some
of
the
aspect
restructure
the
draft.
This
may
be
a
more
major
revision
and
then
depend
we
will
discuss
with
the
shepherd
if
it
may
require
to
go
through
a
secondary
social
class.
Call
just
to
make
sure
that
this
these
important
changes
are
validated
by
the
research
group.
D
D
F
Yeah,
sorry,
sorry,
is
there
any
question
or
feedback?
So,
as
you
said,
laura
you
you've
got
a
lot
of,
let's
say
positive
feedback
mailing
list.
I
just
can
confirm
that.
So
now
it's
it's
up
to
you
to
see
how
you
will
address
a
different
comment
and
then,
as
you
say,
yes,
depending
on
the
type
of
modification
you
will
do,
then
we
will.
I
will
check
if
this
is
necessary
to
go
for,
let's
say
kind
of
formal
and
last
call.
A
Yeah
I
mean
I'll,
say
there
are
some
reviews
that
need
more
attention
to
to
really
see
what
changes
we
need
to
make
to
the
document.
So
that's.
C
A
We
couldn't
reply
until
now,
but
this
is
really
on
our
table.
Yeah.
F
I
think
it's
good
that
you
receive
a
review
from
let's
say
new
readers.
So
basically
you
have
also
an
external
point
of
view
and
somebody
that
was
not
aware
of
the
work
until
now.
It's
important
because
these
documents
somehow
give
some,
let's
say
basis
or
a
fundament
for
this,
for
the
ibn
work
in
the
group.
So
it's
important
that,
although
the
understanding
from
external
reader
is,
is
well
well
certain
well.
F
H
F
H
D
H
H
H
The
three
classifications
have
been
proposed
in
our
draft
following
the
classification.
Workflow
include
carrier
solution
and
data
center
solution
and
enterprise
solution.
Now,
ietf
108
there
is
a
poc
about
a
multiple.
A
multi-layer
approach
for
ibm
solution
has
been
successfully
used
as
an
example
for
our
proposed
classification
methodology.
H
H
H
H
We
expanded
it
in
section
6.1
and
as
we
provided
intent
classification
examples
based
on
the
ietf
108
poc,
a
multi-layer
approach
for
ibm
for
the
carrier,
use
case
and
the
dc
use
case.
We
add
new
ad
sections
at
the
section
6.3.3
and
6.4
and
0.3,
and
then
we
up
updated
intent
descriptions
with
additional
information
on
how
intense
manifest
from
an
operational
point
of
view
in
sections,
6.3,
0.1
and
6.4.16.
H
H
Please
we
have
do
something
after
the
we
submit
the
version
1
and
we
will
submit
version
2
after
the
meeting
I
think
about.
We
continue
this
discussion
on
pocs
integration
into
draft
with
barbara
and
the
water,
and
we
clarify
clarification
on
requirements
for
different
intent,
tabs
based
on
context
in
our
draft
in
section
4.2
and
addressing
the
benefits
of
intents
to
network
requirements
in
section
4.3,
and
there
is
outstanding
about
we
will
add
a
scope
section
within
introduction
for
in
identify
identifying
the
scope
and
the
priorities
of
projects
and
then
will
include
a
definition
section.
H
Acknowledge
acknowledgements,
we
thank
for
all
the
reviews,
suggestions,
comments
and
proposals,
tests
provided
by
the
following
members
list
in
alphabetical
order:
maddie
benshuff,
bernie,
kapenter,
capender,
lauren
and
alexander
klem,
yahya
pedro
daniel
king
and
georgian
and
charlene
we
sent
to
barbara
and
the
water
moka
and
for
contributing
with
their
a
multi-level
approach
to
ibm.
Poc
demonstration.
H
A
I
will
say,
interaction
with
walter,
barbara
and
molka,
on
making
kind
of
example
with
their
with
their
demo.
So
you
think
your
document
will
be
ready
to
proceed
after
you.
You
issue
the
version
two
towards
research
group
last
call,
or
do
you
expect
I
mean?
F
D
I
H
Yes,
okay,
we
can
have
some
the
discussion
about
the
the
intent
classification.
H
F
A
A
Otherwise,
yeah,
okay,
we
cannot
hear
you
and
I
don't
know
jerome
if
you
can
unmute
her,
but
at
least
on
my
side
I
cannot
no.
I
cannot.
A
Okay,
so
just
because
I
think
we
are
drifting
a
bit
and
we
have
a
full
agenda,
I
suggest
we
we
continue
with
the
next
draft
and
eventually,
if
olga
managed
to
raise
our
comments,
we
can
hear
that
still
in
the
ibn
slot
occasion,.
F
Okay,
yeah
sure
so
next,
next
time
on,
the
android
is
the
transpose
licensing
presentation
by
by
lewis.
J
Okay,
thank
you
so
essentially.
Well,
I
I
will
present
the
the
update
of
the
examples
of
the
drug.
To
be
honest,
there
is
a
very,
very
few
updates
and
mostly
related
to
terminology.
Essentially
the.
What
we
have
done
here
is
to
align
with
the
terminology
that
is
being
used
in
this
working
group,
where
we
have
transitioned
from
the
concept
of
transporter
slice
to
the
concept
of
itf
network
slice.
So
I
I
will
go
briefly
to
the
to
the
point,
but
essentially
this
is
the
the
main
message
so
next
slide.
Please.
J
J
So
here,
maybe
is
is
important
to
say
or
relevant
for
us
at
this,
the
the
propo
the
proponents
of
this
work
that
so
how
we
we
do
see
this
as
a
complement
to
the
working
in
this,
in
the
sense
that,
with
this
approach,
this
intent-based
approach
would
be
complementing
what
could
be
related
to
the
concept
of
itf
network
slice
controller,
so
basically
yeah
adding
this
other
piece
on
top
of
that
controller,
dealing
with
the
intents
for
the
creation
and
approvation
of
the
slice,
but
also
whatever
has
to
do
with
the
slice
in
the
itf
scope
next
slide.
J
J
Just
as
a
refreshment
of
what
is
the
the
purpose
of
the
draft
so
is
to
leverage
on
the
ibm
technologies
for
requesting
itf
network
slices
for
essentially
assisting
the
customer,
and
this
customer
could
be
a
high
level
system.
Maybe
the
the
system
from
the
3dbp
could
be
a
clear
example
so
to
request
and
and
manage,
let's
say
the
the
the
negative
slice.
J
So
so
we
do
foresee
the
the
role
of
this
intent
based
system
as
a
one,
managing
the
descriptive
request
from
the
customer
and
essentially
interacting
later
on,
with
the
idea
of
network
slice
controller
in
order
to
have
the
slides
ready
in
the
network.
The
benefits
of
following
this
intern-based
approach
is
is
clear
for
us,
but
with
the
portability
of
the
solutions
across
implementation
and
networks.
So
having
a
disintend-based
mechanisms,
we
could
guarantee
that
portability
in
an
uneasiest
way
having
a
simple
way
of
expressing
transporters.
J
One
itf
never
slice
needs,
and
this
buddha
is
the
political
customer
for
instant
vertical
industries
and
so
to
re
to
be
not
so
specialized
in
the
particular
technology,
but
essentially
working
with
these
descriptive
manners
of
requesting
the
slice
and
necessary
they
focus
on
the
what
that
would
be
the
idea
of
never
slice
and
not
the
how
so
the
essential
of
the
idea
intent
basis
system.
J
So,
as
next
steps,
we
will.
Our
intention
is
to
keep
working
on
developing
the
internet-based
capabilities
for
itf
network
slices.
So
further
of
the
content
that
we
already
have
in
the
in
the
draft
also
very
important
to
align
with
the
ongoing
propositions
for
internet
use
cases,
as
discussed
in
in
the
interim,
the
last
interim
meeting.
J
So
essentially,
we
finally
define
a
kind
of
template
for
use
cases
or
also
adapt
the
description
of
this
use
case
to
that
kind
of
template,
or
once
I
mean,
or
even
work,
more
deeply
in
the
matching
with
the
internet-based
architecture
and
so
on
so
far
so
essentially
trying
to
align
and
synchronize
with
the
other
activities
here
in
such
a
way
that
could
be
seen-
let's
say
as
as
not
as
an
independent
proposition,
let's
say,
but
as
something
integrated
in
the
overall
scope
of
interbase
within
the
network
management.
J
Research
group-
and
this
is
for
the
draft
itself
so
for
the
for
the
community
for
the
group
just
to
request
comments
and
inputs
for
growing
new
versions
of
this
draft,
for
updating
for
new
versions
for
sure,
and
we
are
considering-
maybe
for
the
next
meeting,
to
ask
for
adoption
as
a
an
intern
use
case.
So
we
remind
that
there
is
a
number
of
word
guardians
defined,
and
one
of
them
is
the
the
definition
or
the
selection
of
use
cases.
J
J
A
Yes,
can
you
hear
me
yeah,
yeah,
okay,
thanks
reese
and
orders
for
for
for
the
work
and
the
updates.
I
have
a
comment
which
is
more
a
bit
of
a
surprise
and
a
question.
The
first
comment
is:
I
saw
that
there
is
this
terminology
update
an
alignment
with
teas
about
calling
that
itf
network
slice.
A
When
I
read
that
I
thought,
are
you
trying
to
put
intent
for
the
you
see
the
itf
network
we
used
to
have
in
the
physical
iitf
meetings,
because,
for
me,
ietf
network
is
the
iatf
network.
So
I
was
a
bit
confused
when
I
saw
that
first,
so
I
don't
know
if
the
terminology
will
evolve.
I
know
it's
it's
done
in
tears,
design,
team
etc,
but
I
found
it
a
bit
confusing
and
I'm
not
even
sure
that,
externally
to
iitr
it
will
be
well
understood
what
what
an
ietf
type
of
network
is.
A
Yeah,
I
mean
everyone
else
around
iitf
calls
that
transport
network.
I
know
it
could
be
too
limiting
but
yeah
it's
terminology
discussion,
but
this
is
just
a
a
bit
of
a
surprise
for
me
and
but.
C
A
Question
is
more
on
the.
I
think
I
I
I
concur
to
what
you
are
proposing
to
continue
developing
the
draft
and
eventually
becoming
a
surgical
document
on
the
use
case.
I
think
it's
a
good
good
way
forward.
I'm
just
wondering
in
terms
of
scope,
because
I
remember
that,
for
instance,
amina
was
also
having
some
proposal
around
slice,
lifecycle,
automation
and
other
things.
If,
because
I
think
your
scope
is
quite
well
aligned
with
again
iatf
technologies
and
especially
what
what's
being
developed
in
teas,
so
I
see
the
value
in
that.
A
I'm
just
wondering
if
you
will
be
open
to
consider
a
broader
scope
than
just
ietf
network
slice
so
that
it
can
encompasses
also
other
types
of
I
mean
that
the
internet
based
interface
or
the
internet
based
characteristics
could
also
be
reused
by
other
sdos
or
for
other
types
of
slides.
Just
to
consider.
If
this
will
be
part
of
what
you
you
would
agree,
or
you
will
prefer
to
stick
to
to
this
specific
scope.
J
Okay,
thanks
lauren,
sorry,
the
first
comment
the
other
day.
I
I
also
I
didn't
like
either
the
the
terminology.
It
was
more
comfortable
with
the
transport
slides,
but
well.
This
was
the
the
consensus
reach
until
now,
at
least
so
we
need
to
live
with
that
and
probably
we
maybe
we
can
find
a
better
name
for
the
for
this
draft.
In
order
to
put
that
confusion,
so
I
agree
with
you
that
probably
is
confusing
outside
the
itf
even
inside
idea,
but
by
the
way
I
didn't
change
the
the
name
of
the
draft
itself.
J
I
mean
the
the
code,
the
draft
sla,
if
and
transporter
slides,
and
so,
but
just
simply
the
title,
so
they
are
in
the
second
point,
I
I
see
pros
and
cons
or
maybe
more
than
cons
what
I
would
like
to
to
to
to
to
get
be,
let's
say
close
to
the
working
piece
so
opening
the
scope,
I
think,
is
interesting
and
probably
necessary
in
order
to
to
be
sure
that
it's
consistent
with
whatever
effort
that
could
be
done
outside
and
maybe
we
can
think
of
3gpp
or
in
in
some
other
bodies.
J
So
from
that
point
of
view,
I
think
would
be
beneficial,
but
I
think
that
also
could
be
somehow
interesting
or
beneficial
to
to
keep
the
link
with
the
working
test
that,
in
that
sense,
somehow
I
don't
know
if
reducing
the
scope
of
narrowed
on
the
scope
or
at
least
zooming
on
what
could
be
the
interaction
with
the
itf
network
slice
controller
below,
because
so
how
this
helped
us
to
to
build
a
a
complete
story.
So
linking
with
what
we
are
doing
in
this
as
well.
J
What
we
do
is
we
expect
to
do
in
this
so
yeah
I
mean
we
can
discuss
I'm
open
to
to
go
in
that
direction,
but
maybe
keeping
the
link
with
the
with
the
ground
being
done
in
in
this,
because
I
think,
would
be
beneficial
for
for
us
to
having
the
complete
story.
F
Okay,
I
think
that
danian,
you
have
a
little
a
comment
or
a
question.
F
F
F
So
maybe
maybe
for
my
from
my
side
surgery
for
speaking
luis.
Can
you
make
sure
to
well
understand
on
your
next
point?
Your
last
point
on
the
next
step
slide.
So
the
idea
is
that
for
to
when
you
say,
ask
for
admission
of
the
graph
as
an
energy
intense
case.
It's
you
say
that
in
a
next
meeting,
so
it
mean
it
means
that
you
want.
You
want,
maybe
next
virtual
meeting
you
you
expect
to
have
something
more,
let's
say
concrete
regarding:
oh,
we
will
push
you
in
use
cases.
J
So
that's
what,
for
instance,
what
we
have
discussed
in
some
other
meetings,
this
idea
of
the
platform
use
cases
or
even
better
matching
with
the
intern
based
architecture.
This
is
something
that
is
not
yet
in
the
draft.
So
probably,
if
we
evolve
the
work
in
this
for
coming
months,
maybe
we
can
somehow
align
better
the
this
proposition
with
the
the
overall
intent
working
here
and
and
with
that
maybe
have
a
more
solid
draft
to
be
adopted.
F
F
F
F
F
K
And
next,
the
network
environment
intent
refers,
refers
to
based
on
user
or
network
operators,
demand
for
network
states
and
automatically
collect
network
state
information
on
the
mind
and
the
in
order
to
illustrate
it
more
clearly,
we
will
present
the
detailed
process
of
each
within
each
divided
part
and
we
will
take
the
environment
in
terms
of
busy
network
performances
as
an
example
in
the
recognize
or
generate
intent,
function
and
the
translator
or
refine
function.
Network
environment,
intense
needs
to
be
translated
into
actions
and
requires
taken
against
the
network.
K
So
these
functions
need
to
determine
the
network
busy
time
of
the
day
based
on
historic
data
learned
by
ai
and
then
determining
the
content
need
to
be
measured,
and
then
we
will
take
the
lapping
part
as
a
whole,
namely
the
part
consists
of
learn
or
plan
or
ringer
function,
configure
our
provision,
function,
monitor
or
observe,
function,
validate
function
and
endless
aggregate
function,
and
in
this
part,
based
on
the
required
measurement
content
and
the
equipment
support
degree,
it
determines
the
corresponding
environment
scheme
and
verifies
whether
the
environment
scheme
is
visible
and
the
next
slide.
Please.
K
And,
at
the
same
time,
the
light
pink
part
also
needs
to
determine
whether
the
network
is
busy
according
to
the
current
network
stage,
while
the
busy
time
threshold
is
exceeded.
This
part
performs
automatic
network
deployment,
such
as
in
cli
mode
and
to
meet
the
requirements
of
the
network
environment
of
intent.
These
parts
also
need
to
collect
measurement
data
on
the
mind
and,
of
course,
this
data
is
collected
automatically
as
and
the
abstract
function
and
report
function
from
the
last
part.
K
K
In
this
page,
I
will
explain
the
advantages
of
convening
network
environment
with
intent.
First
of
all
is
automation,
is
it
can
realize
automatic
network
measurement
and
automatic
analysis
of
measurement
results?
K
Second,
is
in
intelligence,
it
can
realize
the
intelligent
anomaly,
detection,
normally
prediction
and
automatic
repair
when
the
environment
results
is
not
in
conformity
with
the
expectation
and
the
finally
is
the
closed
loop
control.
Due
to
the
closed
loop
operation
is
available
in
the
intent-based
network.
K
K
F
K
A
Yeah
yeah,
thank
you
for
for
the
work
and
and
the
the
proposal.
I
think
there
are
interesting
aspects
if
you
understand
correctly
what
you
want
to
do,
but
I
was
also
a
bit
confused
with
some
of
the
way
you
you,
you
present
things.
A
What
I
understood
is
that
as
an
operator
or
as
a
customer,
it
may
be
interesting
in
getting
some
reporting
information
or
what
you
call
measurement
information
corresponding
to
some
to
some
expectations
and
then
what
that
this,
this
needs
of
of
of
getting
some
specific
information
will
trigger
different
measurements
or
monitoring
activities
inside
the
network,
and
the
other
things
you
are,
I
think
trying
also
to
to
put
together
is
that
you
may
want
to
to
detect
if
the
the
measurement
or
the
monitoring
can
fulfill
the
the
expectation
of
the
measurement
provided
by
the
operator.
A
And
the
other
thing
is
that
if
some
some
issues
are
in
fact
happening
and
and
to
be
able
to
detect,
in
fact
from
this
measurement
expectations,
if
there
are
deviations
from
from
what
you
expect
to
see.
A
So
for
me,
what
I
was
a
bit
confused
is
that
you
have
at
least
two
or
three
phases
or
aspects
that
you're
looking
after
in
the
draft,
but
the
way
they
are
articulated
right
now
it
was
a
bit
confusing
for
me,
so
I
see
potential,
but
I
think
we
may
work,
maybe
together
to
to
try
to
clarify
the
different
steps
or
where
is
the
intent
and
where
are
the
other
functionalities
coming
into
play.
A
So
this
was
just
kind
of
observation
that
I
think
there
are
potential,
but
maybe
need
some
clarification
in
what
you
really
want
to
achieve
and
how
you
see
this
to
work.
B
Yes,
I
I
agree
with
you
around
in
his
comments
that
this
is
somehow
mixes
the
idea
of
measurement
with
the
idea
of
action
building
a
closed
loop
or
whatever
the
nature,
and
this
is
a
little
bit
misleading.
That's
true!
This
is
something
that
I
think
is
really
vulnerable
here
and
it's
connected
with
some
some
ideas
that
we
are
starting
to
work
with.
B
It's
precisely
the
idea
when
you
have,
when
you
have
a
catalog
or
a
or
a
database
or
whatever
you
call
it
of,
for
example,
measurement
tools
and
measurement
capabilities,
and
what
you
express
is
precisely
this
is
that
you
request
the
measurement
capability
and
not
necessarily
the
tool,
because
the
tool
may
have
to
be
adapted
to
the
particular
environment
in
which
you
are
going
to
deploy
it
and
well
what?
B
Personally,
I'm
trying
to
connect
this
with
an
idea
that
I'm
insisting
very
much
recently,
which
is
about
the
repeatability,
how
you
guarantee
that,
whatever
you
do
and
whatever
you
report
to
others,
whether
it
is
in
on
a
paper
or
a
test
recall
or
an
experiment
report
or
whatever,
is
something
that
can
be
repeatable
by
others.
B
These
an
approach
like
this
probably
would
translate
in
in
a
higher
level
of
expressing
the
repeatability
requirements,
and
I
think
this
is
if,
if
we
narrow
down
a
little
bit
the
scope,
this
will
be
quite
interesting.
K
Okay:
okay,
thank
you
for
your
suggestions
and
we
will
discuss
it
on
relay.
Thank
you.
F
Okay,
thank
you
again,
chen.
We
have
to
move
to
the
to
the
next
time
on
the
agenda,
so
we
will
start
with
the
part
related
to
artificial
intelligence.
Lauren.
You
are
the
the
first
one
in
this
part.
H
F
A
Okay,
so
so
this
is
in
the
area
of
artificial
intelligence
from
the
network
management,
so
we're
a
bit
switching
topics,
and
today,
I'd
like
to
give
a
presentation
on
artificial
intelligence
for
network
and
service
automation,
and
especially
to
look
into
the
standard
enablers
and
in
some
research
directions.
A
I
will
give
only
a
short
overview
of
the
research
topics,
because
it's
quite
big-
and
I
think
it
will
deserve
a
an
order,
full
presentation
and
just
to
to
stress
also
that
this
is
part
of
a
larger
activity
inside
nokia,
and
this
is
only
a
an
extract
of
this
work
that
we
propose
in
the
scope
of
energy
and
if
you
are
interested
to
learn
more
on
this
activity
or
if
you
want
to
collaborate
with
us,
because
we
have
also
related
actions
and
activity
in
order,
sdos
don't
hesitate
to
contact
me
after
the
meeting
or
offline.
A
We
also
hope
that
the
content
presented
here
will
help
to
provide
input
to
the
ai
research
challenges
document
that
we
that
journal
will
present
just
after
this,
and
also
I
have
a
lot
of
slides
sorry
for
that.
But
we
wanted
to
present
the
overall
content.
So
I
will
go
a
bit
quick
on
the
first
slide,
which
talks
about
the
background,
but
you
can
look
at
them
after
and
in
order
to
keep
some
time
for
for
the
core
part
of
the
presentation.
A
A
In
fact,
we
see
continuous
innovation
and
progress,
especially
in
the
technology
of
ai
theories
and
breakthroughs
in
a
new
algorithm,
etc,
and
also
a
continuous
improvement
in
the
performances
of
all
these.
These
different
techniques
in
the
area
of
data
more
and
more
data
being
collected
also
from
different
types
of
sources.
A
Data
also
becomes
available,
and
we
say
everywhere
in
different
area
of
the
operator
infrastructure.
These
data
are
also
becoming
more
and
more
aggregated.
It's
not
only
a
single
types
of
counters
that
you
can
collect,
but
it
comes
from
technological
devices,
but
also
from
customer
records,
etc.
A
So
really
different
types
of
data
being
put
together
to
to
to
be
used
as
sources
for
the
machine,
learning
algorithm
and
also
more
data
awareness
in
the
different
organizations
about
the
value
of
the
data
and
how
to
organize
operations
and
activities
in
the
different
organization
to
be
data-driven
and
have
this
data
around
us
everywhere.
A
On
the
compute
side,
this
is
pretty
obvious.
I
mean
we
are
still
quite
pushed
by
continuous
improvements
in
performances
of
the
compute
architecture,
but
also
the
fact
that
there
are
many
different
computing
architecture,
in
fact,
in
support
to
especially
machine
learnings
and
a
continuous
improvement
of
the
performance
of
all
the
different
architectures
for
the
tooling.
A
This
is
very
important
in
the
area
of
ai
and
machine
learning,
a
maturing
ecosystem
of
different
platforms,
languages
and
libraries
that
are
most
of
the
time
publicly
available
in
their
open
source,
very
helpful
to
support
the
research
and
the
progress
and
innovation
in
that
area,
and
also
very
important.
This
aspect
of
pipeline
automation,
so
not
only
providing
the
tools
and
libraries,
but
also
to
ease
a
bit
the
work
of
the
developers
and
the
integrators
in
creating
data
and
the
different
automation
that
occur
in
the
in
the
ml
development
process.
A
Finally,
on
the
business
side,
this
is
a
bit
more
disruptive,
especially
different.
I
will
say:
models
how
to
deploy
and
operate
machine
learning,
machine
learning
as
a
service
air
as
a
service
through
apis,
and
we
see
also
more
and
more
in
the
telco
space
collaboration,
especially
direct
collaboration
between
web
scale,
actors
and
operators,
and
sometimes
without
deco
vendors
in
the
loop.
So
this
is
a
quite
a
change
in
how
networks
are
being
operated
and
who
is
involved
in
this
new
era.
A
So
there
are
drivers,
but
we
also
identify
different
gaps,
especially
that
when
we
say
artificial
intelligence.
In
fact,
most
of
the
time
we
refer
only
to
machine
learning
and
in
machine
learning,
especially
to
deep
learning,
which
was
really
the
mainstream
and
what
is
driving
the
activity
in
that
area.
A
The
second
main
gap
is
that
so
far
the
ie-based
solutions
have
been
demonstrated
in
isolation
or
especially,
towards
specific
use
cases.
So
this
is
good
because
it
shows
that
it
works,
but
we
also
a
bit
lacking
all
aspects
about.
I
mean
collaboration
or
coordination
between
different
ai-based
solutions.
A
It's
a
very
limited
aspect
of
generalization
and
automation
across
different
technology
across
different
use
cases,
and
also
we're
a
bit
lacking
this
reusability
factors
due
to
the
lack
of
architectural
principles,
especially
in
the
operator
environments.
A
I
mean
sometimes
what
work
for
a
use
case
is
difficultly
replicable
in
in
other
areas,
and
also
the
third
gap
related
to
this.
To
this
discussion
today
is
the
limited
use
or
reliance
on
standards.
A
A
But
we
see
also
values
in
in
having
standards
being
developed,
especially
for
the
telecommunication
environments,
so
still
in
the
infancy
age
of
this
ai
standards
for
networks,
but
also
that
the
standard
needs
to
position
itself
to
be
able
to
provide
value
with
multi-vendor
and
open
solution
in
that
area,
and
also
the
fact
that
without
standards,
we
will
miss
some
key
properties
like
interoperability
and
reusability,
of
the
what
what
is
currently
delivered
via
this
kind
of
de
facto
software
environment.
A
In
fact,
if
you
bring
the
two
together
for
us,
we
have
broadly
two
two
main
domains
or
cases
of
applications
which
are
what
we
call
ai
assisted
management
and
automation,
which
is
where
you
you
take
current
operations,
current
functionalities
and
you
try
to
inject
ai
into
that
or
make
them
ai
based.
And
so
you
increase
the
efficiency
and
the
support
to
human
driven
operation.
A
But
you
see
this
is
what
you
do
today
and
you
do
it
with
ai,
the
other
main
scope
with
the
ai,
what
you
call
air,
enable
or
air
and
power,
which
is
really
to
try
to
discover
unveil
new
areas,
new
way
of
doing
things
that
were
not
possible
without
without
ai,
either
due
to
the
complexity
or
due
to
it,
was
too
challenging
to
process
the
amount
of
data
etc.
So
it's
really
to
not
do.
A
What
you
can
do
today
is
to
do
more
things,
new
things
and,
in
fact,
as
a
conclusion,
the
we
need
really
to
have
an
advanced
integration
and
embedding
of
the
capabilities
which
is
needed
to
maximize
the
use
and
efficiency
of
ai,
assisted
and
air,
enable
management
automation.
So
we
really
think
it's
important
not
only
to
work
on
the
ai
techniques,
but
also
on
how
they
get
integrated
and
how
we
embed
them
with
the
network
environment
in
terms
of
a
bit
coverage
this
high
potential
areas.
A
I
will
not
explain
each
of
these,
and
in
fact
it's
not
exhaustive.
It's
just
to
highlight
that.
In
fact,
you
can
apply
ai
techniques
virtually
everywhere
in
in
the
networking
environment,
and
you
can
have
different
types
of
problems:
different
benefits,
different
types
of
techniques
you
can
apply,
and-
and
so
the
goal
and
kind
of
an
underlying
challenge
that
we
start
to
see
here
also
is
it's
to
find
solutions
or
capabilities
that
are
really
reusable
and
replicable
so
that
you
can
apply
it
end-to-end.
A
So
for
all
the
management
tasks
and
all
the
operations
that
operators
are
doing,
and
not
only
always
specialize
in
use
case,
based
techniques,
so
a
kind
of
counterpart
with
that
it's
kind
of
a
problem,
technique,
mapping-
and
it's
absolutely
not
prescriptive.
A
It's
in
fact
to
consider
a
typical
panel
of
techniques
that
you
could
use
to
address
some
of
the
problems
and
ai
tasks,
but
we
don't
see
any
limit
here.
I
mean
it's
just
a
kind
of
indicative
mapping,
because
maybe
tomorrow
someone
will
use
a
technique,
use
never
used
before
for
for
this
kind
of
problems
and
may
trick
it
in
a
certain
way
and
it
will
work
and
have
very
great
results.
So
this
we
don't
know,
but
it's
just
that
for
some
of
the
ai
type.
A
What
you
see
a
bit
on
the
left
is
that,
depending
on
the
type
of
I
will
say
task
that
you're
after,
if
you
want
to
get
some
explanation,
a
diagnosis
this
is
essentially
when
you
are,
you
need
to
have
partner
or
even
detection
and
recognition.
A
A
Then
we
have
also
tried
to
categorize
some
set
of
key
challenges
for
ai
and
network
management
or
ai
networks
together.
So
first
I
mean
just
you
to
be
able
to
specialize
in
our
own
field.
A
So
first,
it's
that
networks
are
hard.
Ai
problems,
networks
are
distributed,
dynamic
environments,
usually
heterogeneous
in
terms
of
technology
and
and
types
of
operations
and
choices
from
operators,
and
also
we
also
face
more
and
more
the
kind
of
challenge
related
to
encryption.
We
have
also
complexity
in
the
data
sets
that
we
face
in
network
environment.
It
can
be
laws,
it
can
be
performance
counters,
it
can
be
customer
records,
etc.
A
So
very
types
of
data
sets,
and
also
very
technology,
specific
or
domain
specific
and
complex,
and
the
really
and
the
relative
complexity,
also
in
the
algorithm
to
deal
with
this
complexity
in
the
data
sets
and
also
we
have
high
dimensional
spaces
in
the
network
data,
and
they
are
also
interogenes
and
diverse.
A
I
mean
if
you
go
to
the
access
network,
if
it's
fixed
access
or
mobile
access,
3g
4g
5g,
it's
different
types
of
of
data
types
and
also
diverse,
because
you
have
different
formats
for
different
applications,
different
vendors,
etc,
and
the
last
point,
which
is
starts
to
have
a
bit
less
of
a
challenge
recently,
but
access
to
network
data
can
also
be
sometimes
difficult
and
challenging,
and
you
have
a
lack
of
reference
label.
Data
sets
in
many
different
applications.
A
A
second
challenge
is
on
the
confidence
in
ai,
so
ai
technology
introduced
new
challenges
and
specificities
such
as
you
can
have
performance
the
gravitation
over
time.
If
your
the
profile
of
your
data
changes,
you
need
to
retrain
at
least
minimum
retrain
on
your
model.
Sometimes
you
may
even
need
to
change
it.
A
Data
sensitivity,
database,
adversarial
inputs,
some
issues
related
to
governance,
to
address
properly
security
and
privacy
on
the
data
used
by
the
machine
learning
algorithm
shift,
also
from
a
kind
of
deterministic
and
probable
algorithm
that
we
have
in
many
applications
in
network
environment,
to
the
typical
machine
learning
paradigm,
which
is
based
on
stochastic
and
statistical
paradigm,
and
also
challenges
related
to
explainability
accountability
of
how
the
decision
process
was
made
by
those
algorithms
that
we
don't
necessarily
understand
the
inner
workings.
A
Human
eye
interactions
are
the
third
challenge
need
to
support.
I
mean
we
can
have
different
levels
of
intelligence
provided
by
the
ai
and
then
different
level
of
support
to
the
human
operator
from
acc
to
empower
it,
as
you
remember,
we
explained
in
the
scope
so
how
to
assist
or
extend
the
human
operator
in
the
service
design,
the
parametrization
and
operation
different.
A
Finally,
our
fourth
challenge
we
call
ai
diversity
is
that
I
mean
the
eye
techniques
are
not.
I
mean
it's
a
broad
set
of
different
techniques
with
quite
different
characteristics,
and
then
they
will
have
different
requirements,
and
standardization
needs
also
the
overall
design,
integrating
and
averaging
individual
ai
properties
needs
to
be
put
forward.
A
Then
what
we
see
as
very
important
to
to
understand
this
journey
is
what
we
call
a
joint
evolution
of
ai
and
ops,
so
we
don't
consider
only
ai
in
isolation
and
network
management
management
or
network
operation
isolation.
We
really
think
that
it's
a
joint
evolution,
so
we
want
to
go
from
what
we
see
on
the
left
side,
which
is
with
the
situation
today.
What
we
call
raw,
ai
and
automated
ops
to
the
kind
of
ultimate
stage
in
25
and
beyond
intuitive
ai
and
autonomous
ops,
and
to
go
through
that.
A
In
fact,
we
can
consider
various
criterias
what
you
see
on
the
left,
ai
and
data
scale,
and
adoption
practice
and
integration,
competence
and
security
standards,
regulation
etc.
So
what
we
want
to
insist
in
this
slide
and
again,
I
will
not
go
into
the
details
due
to
lack
of
time,
but
it's
that
it's
not
only
driven
by
ai
techniques
or
ai
technology
development.
This
is
just
the
first
row
what
the
ai
can
do
and
what
the
type
of
data
you
need
to
have,
and
also
we
see
there
are
still
evolution.
A
There
are
other
things
which
are
more
on
the
the
penetration
of
of
ai
in
the
in
the
infrastructure,
also
the
no
ho
the
practice
of
using
and
developing
ai,
specifically
for
networks,
but
also
what
I
said
about
this
confidence
and
security,
making
sure
that
we
can
safeguard
the
ai
operation
and
adopt
it
at
wide
scale,
etc,
and
the
very
importance
of
standards
and
regulation
standards
for
interoperability
and
regulation
in
order
to
make
sure
we
comply
with
the
right
frameworks
and
and
policies
in
terms
of
addressing
privacy,
security
and
other
related
ethical
issues.
A
The
the
approach
here
is
it's
to
show
a
bit
where
we
see,
I
mean
it's
not
that
you
will
have
ai
only
in
one
of
a
small
area
in
the
network.
It
will
really
be
pervasive
everywhere,
as
you
can
see
at
the
bottom,
we
can
have
it
in
the
different
segments.
Different
domains
also
capturing
the
specificities
of
ai
being
deployed
in
such
environment.
A
If
you
deploy
on
a
smartphone
or
a
sensor,
it's
not
the
same
type
of
ai
and,
and
we
say
computing
environment
that
you
can
get
if
you
deploy
it
in
a
big
data
center
and
also,
if
you
go
up,
I
mean
vertically,
you
have
ai
in
the
data
plane,
but
also
you
will
have
application
of
ai
for
the
control
plane
and
also
the
management
plan,
which
is
the
main
area
of
concern
for
for
network
management
and
network
operations.
A
And
then
in
the
different
effort
management
domain,
you
will
have
ai
dedicated,
for
instance,
if
it's
operating
in
the
in
the
run,
access
or
if
it's
operating
in
the
core
or
it's
up
everything
in
the
transport.
So
you
can
also
capture
different
specificities
of
the
operations
there
and
also
the
end-to-end
service
management
and
what
you
see
also
sorry,
this
kind
of
arrows.
A
Really,
I
will
show
some
slides
on
that,
but
this
is
really
the
core
of
what
we
think
should
be
the
enablers
to
to
put
forward
in
order
to
really
make
sure
we
can
integrate
and
support
fully
functionality
and
operations
based
on
ai.
The
second
aspect
is
support
for
deployment
and
diversity.
A
This
aspect
is
very
important,
because
the
first
one
is
really
to
say
what
we
need
to
develop
to
make
it
work
to
make
it
integrated.
The
second
one
is:
if
you
develop
enablers,
you
want
to
make
sure
that
you
can
reuse
them
and
apply
them
to
as
many
scenarios
as
possible
as
many
technologies
as
possible
as
many
use
cases
as
possible.
So
you
don't
want
to
develop
an
enabler
that
will
be
too
specific
ins,
for
instance,
to
a
a
given
technology
or
given
protocol
or
a
given
deployment.
A
You
want
to
make
sure
that
you
can
use
it
as
broadly
as
possible,
and
so
we
see
also
a
role
that,
when
we
design
these
enablers,
when
we
specify
them-
and
we
say
the
research
topics
we
need
to
investigate-
we
need
also
to
make
sure
that
our
design
is
extensible
and
and
portable
the
the
final
aspect,
trust
and
adoption.
This
is
really
because
it's
not
just
because
you
have
good
technology
and
you
can
apply
it
in
many
areas
that
it
will
get
adopted.
A
With
this
change
going
to
more
and
more
ai-based
operations,
we
need
also
to
support
the
user,
the
operator
different
verticals
that
will
reuse
this
technology
in
having
the
right
frameworks
in
place,
a
trust,
building
framework,
a
security
framework
and
all
these
different
techniques.
You
know
they're
really
to
support
and
make
sure
that
when
you
use
the
ai
technologies
and
they've
been
embedded
in
your
infrastructure,
you
can
know
how
they
operate.
You
can
audit
them.
You
can
have
explanation,
etc.
A
So
the
way
we
have
structured,
we
propose
to
structure
the
some
of
the
standardization
aspects
and
research
aspects
for
network
and
service
automation
using
ai
is
with
this
so-called
enabling
areas
we
have
identified,
six
of
them
execution
data,
interai
actions,
governance
and
inner
ai,
and
for
each
of
them.
A
In
fact,
we
try
to
describe
what
we
consider
to
be
important
and
how
this
contributes
to
being
an
enabling
area
for
ai-based
operation.
So,
for
instance,
of
this
execution
or
existing
environment
is
really
to
consider
a
versatility
of
and
this
continuum
of
where
the
applications
can
be
deployed.
We
can
think,
for
instance,
of
some
of
the
the
work
being
developed
in
the
coin
energy
about
this
data
discovery,
data
ability
and
the
contra
part
on
the
execution
and
compute.
Where
is
the
compute?
A
What
is
the
form
of
compute
and
how
you
can
deploy
your
applications
on
top
of
this
distributed
computing
environment?
A
So,
really
finding
the
good
combination
between
deploying
the
air
in
terms
of
operations
and
also
the
access
to
data,
and
we
relate
that
to
a
set
of
different
standard
needs
so
having
the
ability,
for
instance,
to
express
requirements
and
constraints
for
the
different
deployment
options
of
the
applications.
Can
I
run
more
or
less
distributed?
A
What
are
my
requirements
in
terms
of
computing
processing,
powers,
storage,
etc,
and
the
counterpart
is
what
my
infrastructure
is
providing
as
capabilities
and
in
fact
we
need
to
match
those
two
sets
of
capabilities
and
constraints
in
order
to
make
it
to
be
able
to
support
a
wide
variety
of
deployment
scenarios
and
to
automate,
in
fact,
those
scenarios
for
the
data.
Enablers
it's
a
bit.
The
same
situation
is
in
fact
to
ensure
that
we
have
access
to
the
right
data
at
the
right
place
at
the
right
time,
and
there
are
many
challenges
related
to
that.
A
Sometimes
data
is
not
available.
Our
data
is
meaningful
only
for
a
certain
amount
of
time,
so
we
need
also
to
be
able
to
localize
the
ai
execution
with
where
the
data
is
collected
or
where
the
data
is
meaningful.
There
is
also
an
issue
in
the
representation
of
the
data.
We
have
already
good
data
collection
platforms,
but
we
need
also
to
work
more
on
the
description
of
what
the
data
is
the
so-called
metadata,
especially
if
we
want
to
be
able
to
automate.
If
I
don't
want
the
human
to
say.
A
I
understand
that
this
is
this
performance
counter
and
you
need
to
plug
it
as
input
to
this
machine
learning.
If
you
want
a
machine
to
do
that,
we
need
to
have
descriptors
of
what
the
data
is
even
going
towards.
Maybe
some
ontologies
describing
this.
So
we
see
a
role
for
the
standards
and
some
research
aspects
related
to
that
also.
The
fact
that
the
data
patterns
are
dynamic
and
will
change
over
time,
especially
in
our
networking
environment,
so
limited
validity
of
the
learn
model,
limited
generalization
of
the
inferent
knowledge.
A
So
this
was
considered
an
important
area
of
research
enter
ai.
F
Excuse
me:
excuse
me,
I
think,
maybe
to
to
move
forward
maybe
to
some
conclusion,
because.
A
A
bit
longer,
I
will
accelerate,
so
we
have
still
three
other
neighbors.
So
italia
is
really
to
consider
everything
related
to
interactions
interpolation,
where
you
have
multiple
ais
being
developed.
If
you
remember
the
diagram,
I've
shown
you,
we
have
a
lot
of
different
ai
instances
everywhere
for
different
applications,
but
you
will
have
also
to
orchestrate
how
they
operate
together,
and
there
is
a
lot
of
standard
needs
and
research
aspects
to
investigate
there,
actions
very
important
one.
This
is
really
for
closing
the
loop
of
automation.
A
If
you
have
an
output
of
an
ai
or
machine
learning
algorithm,
it
can
be
a
score.
It
can
be
a
vector,
it
can
be
different
forms,
but
you
cannot
use
it
right
away
out
of
the
ai.
So
if
you
want
to
automate
the
usage
of
this
output,
you
need
to
plug
it
to
the
right
actions
to
the
right,
orchestration,
etc.
A
So
this
is
very
important,
also
and
related
to
data,
this
aspect
of
describing
what
is
the
output
and
what
is
the
action?
I
can
activate
governance.
This
is
a
very,
very
important
area,
as
you
can
see
a
lot
of
different
considerations
and
standard
needs.
This
is
ready
to
consider
usability,
trust
and
integration
of
the
I
component
in
a
network
and
service
automation
operation
in
rai.
A
A
We
also
have
other
additional
important
considerations
at
the
architectural
level
going
towards
machine
reasoning,
security,
which
is
a
very
important
aspect
that
we
consider
transverse
to
to
these
different
enablers.
We
have
considered
an
ethics
and
regulation
which
more
and
more
becomes
integral
part
of
the
technology
development.
A
So
the
color
code
is
a
map
to
the
different
enabling
areas,
but
you
can
see
also
that
across
the
different
enabling
areas
you
have
a
lot
of
relationship
and
interdependence
between
the
different
topics
and,
finally,
just
to
summarize,
because
I
think
we're
already
quite
overtime,
this
is
a
bit
to
frame
what
we
believe
could
be
good
directions
to
to
push
forward.
Go
deep.
This
is,
in
fact,
to
maximize
automation
by
really
enabling
the
integration
of
the
full
range
of
ai
capabilities.
A
Second,
to
go
wide
again,
if
you
have
excellent
enablers,
excellent
ai
technologies,
but
they
are
too
specialized,
then
you
are
missing.
Something
so
go
wide
is
in
fact
to
be
able
to
make
sure
that
you
can
support
the
broadest
application
deployment,
diversity
for
the
telco
and
also
vertical
scenarios,
so
to
ensure
portability,
reusability
go
safe,
is
in
fact,
this
aspect
of.
If
you
want
to
foster
an
adoption,
you
need
to
make
sure
that
you
can
provide
support
the
human
in
understanding
how
the
ai
works,
how
it
is
integrated.
A
F
F
Comment
laura.
You
can
hear
me
yeah.
F
Okay,
now
because
I
got
an
error
message
and
I
was
thinking
okay,
I
think
it's
it's
really
interesting
yeah.
Unfortunately,
we
don't
have
much
time
to
discuss.
I
think
yeah
there
is
a.
Is
there
also
some
overlaps,
as
you
say,
with
what
you
are
doing
in
the
other
document?
F
You
know
in
the
group
and
also
some
that
we
can,
I
think,
really
also
benefit
from
each
other
from
the
different
ideas,
and
I
think
it's
it's
really
now
we
don't
have
any
time
to
without
it,
but
I
would
really
like,
if
we
can
in
the
next
meeting,
if
we
can
already
arrange
more
time
to
to
discuss
a
bit
more
what
you
present
here
and
also
a
very
dedicated
meeting
to
ai
injury
meeting,
because
I
think
it's
it's
too
short,
now
too
much
content.
F
So
maybe
you
move
to
the
next
to
the
next
ethan
and
of
course
anyway,
we
can
also
discuss
then
on
gather.
I
think
we
have
time
so.
The
next
is
my
presentation.
So
let
me
share
my.
D
F
So
I
will
tell
you
to
keep
it
sharp.
So,
yes,
I
represent
the
status
of
the
document
and
research
challenge
in
a
ai
for
network
management,
so
just
really
a
quick
update.
So
as
you,
so
we
have
here
the
link
for
the
google
doc.
F
F
F
There
are
some
comments
that
have
been
raised
because
I
mean
this
is
one
was
regarding
what
is
the
specificity
of
the
challenger
into
network
management?
Why
existing
techniques,
let's
say
for
real-time
analysis
of
let's
say
big.
C
F
F
We
had
virtual
meetings
in
november
to
discuss
mostly
the
scope
and
rgt
us
document.
F
Mostly,
we
say
that
we
should
not.
We
should
not
focus
on
particle
application,
domains
or
problems
and,
as
we
already
said,
we
don't
want
to
have
this
document
to
be
a
collection
of
use
case.
Although
one
problem
we're
getting
that's
one
issue
was
getting
ai.
What
we
think
what
we
mean
by
ai
in
document,
it's
very
important,
to
clarify.
F
We
mostly
mostly
have
in
mind
machine
learning,
but
still
there
is
some
some
challenges
that
you
could
describe
in
regarding
other
techniques
from
ai,
so
we
should
be
clear
on
what
are
the
boundaries
of
ai.
If
you
want
to
integrate
everything
in
the
camera
or
not,
we
don't
have
the
answer
yet,
but
you
have
to
be
clear,
also
use
really
a
precise
time,
because
the
document
should
be
also,
let's
say
not
only
for
us,
but
also
for
ai
expert
and
should
be
developed
in
conjunction
with
ai
expert.
F
We
also
want
to
highlight
the
in
the
comments
before
addressing.
We
have
challenges
I
like
why
we
have
these
challenges.
What
makes
the
basically
network
management
problems.
F
Why
is
this?
These
problems
will
require
artificial
intelligence
because
of
course,
there
are
maybe
different
answers
to
that,
because
we
have
too
many
to
complex
data
too
much
towards
the
data
we
need
fast
decision
and
so
on,
and
we
cannot
just
rely
on
human
response.
So
there
are
some
introduction
content
that
has
been
proposed
and
although
maybe
these
are
also
new
problems
that
we
can
solve
with
ai,
that
we
don't
think
before
and
also
I
think
it's
something
that
you
are
highlight.
Lauren
your
presentation
before.
F
We
really
discuss
what,
which
should
be
the
value
of
the
document
in
order
to
refine
the
scope
and
ability
of
the
document
we
what
has
been
raised
that
which
the
idea
that
what
we
see
today
is
that
there
is
a
lot
of
a
test
and
trial
method
in
the
fantasy
research
paper.
For
example,
we
just
take
different
methods
of
ai
to
apply
to
our
problem
and
say:
okay,
this
method
works
better
or
we
add.
F
We
want
something
we
want
basically
to
have
that
that,
rather
than
having
a
trust
and
try
method
help
people
to
really
identify
what
are
the
issue
when
they
want
to
apply
ai
to
their
problem
in
order
to
have
a
better
guidance
on
how
they
can
apply.
Aha,
so
basically,
we
want
to
make
the
breed
between
a
network
engine
problem
and
isolation.
So
that
is
a
proposition
now.
Everything
is
still
discussed.
F
There
are
different
obstacles,
so
we
discuss
a
lot
about
what
ai
techniques
for
world
problems.
But,
of
course,
it's
just
only
one
one
issue
when
you
that
trying
to
find
the
right-
let's
say
ai
algorithm,
for
your
problem-
it's
only
one
issue
and
there
is
other
other
obstacles,
for
example
organic
explainability,
education,
auto
usea,
ai
and
so
on.
So
it's
not
about
technical,
let's
say
issue
as
well:
we
don't
want,
as
I
said
before,
we
don't
want
to
document
the
set
of
use
cases.
F
We
don't
want
to
be
as
documented
as
a
survey
between
what
has
been
used
for
the
different
network
management
programs.
Otherwise
it
will
be
always
unless
we,
I
think
we
will
never
reach
a
consensus.
So
what
should
be
the
right
technique
for
others
in
each
problem
and
we?
F
Finally,
I
tried
to
put
in
the
document-
let's
say
a
scope
or
tentative
scope
following
this,
this
meeting
that
here's
here
it
just
the
next
price,
we
can
see
the
full
proposal
in
the
google
doc,
but
basically
we
want
also
to
have
a
a
method
on
top
of
the
challenges
in
order
to
identify
those
challenges,
rather
than
saying
that
we
have
challenges.
Of
course
we
we
know
that
we
have
that.
F
Maybe
we
would
also
have
a
method
and
check
if
we
cover
all
important
or
all
relevant
challenges,
and
that
is
also
the
idea
of
the
document
or
a
method.
I
want
that
and
yes,
the
objective
of
the
documents
for
the
reader
is
justify
set
of
challenges,
so
I
mean
it's
important
that
the
challenges
are
justified
and
just
actually
in
that,
because
we
things
are
important
and
maybe
sometimes
better
also
or
it's
easier
to
define
what
the
document
should
not
be
it's.
F
We,
it's
not
a
yeah
again
a
use
case
document,
and
we
don't
expect
in
the
ground
to
provide
answer
solution
to
all
challenges
that
we
will
introduce
next
step.
I
would
like
to
have
in
the
next
couple
of
weeks
to
have
the
scope
fixed.
So,
of
course
you
can
give
now
as
a
feedback
on
the
scope
of
the
document
that
was
on
the
mailing
list.
F
We'll
have
more,
maybe
interaction
more
time
on
the
mailing
list
as
well,
and
then
yes,
on
the
scope
of
the
document,
this
fix,
we
can
reframe
the
method
to
identify
the
challenge
and
although
we
also
supported
that
challenge
are
described
in
a
really
different
level
of
detail,
so
we
need
some
kind
of
a
korean
description
to
run
content
into
the
different
document.
C
F
See
what
will
be
the
people
preference
and
with
that
I'm
done
so
I
think
if
you
have
any
quick
question
or
comment,
we
can
take.
F
F
Okay,
so
if
there
is
a
any
comment
now
I
think
we
can,
we
can
move
to
the
next
point,
the
agenda
that
that
is
so
multi
multi-level
configuration.
The
presentation
will
be
given
by
luis.
F
F
I'm
really
sorry
that
okay.
J
No,
no
problem,
okay,
so
the
first
to
mention.
Well,
I
will
present
that
one
on
behalf
of
my
co-authors,
ian
bogdanovich
and
sufan
liu
from
from
bolton-
and
maybe
you
can
recall
the
this
topic-
was
authorized
in
in
the
mailing
list-
some
propositions
from
the
end
and
then
we
start
discussing
on
top
of
that
and
the
result
was
this
document
that
I
will
present
right
now.
So
next
slide.
Please.
J
So
essentially,
the
motivation
will
be
how
to
to
handle
the
the
situations
that
we
see
in
in
in
operational
networks,
where
essentially,
a
a
number
of
device
configurations,
grow
and
grow
along
the
time,
and
it's
quite
difficult
to
to
have
a
clean
configuration,
a
base
configuration
and
essentially
we-
we
have,
let's
say
residual
configurations
alone
so
which
are
motivated
by
different
reasons,
for
instance
debugging
these
sessions
in
order
to
solve
problems
that,
where
the
the
people
operating
the
network
start
configuring
things,
dynamic,
evolution
of
protocols
or
peers
or
whatever,
where
you
maybe
could
been
person
b
sessions
that
later
on,
are
left
there
without
traffic
or
something
things
of
this
kind
of
things.
J
Okay,
so
essentially
all
the
configuration
information
is
finally
present
in
a
single
file,
a
plain
file.
So
all
this
multiplex
there
somehow,
which
is
hard
to
manage,
and
even
though
there
are
solutions
proposing
central
databases
with
all
the
network
configuration
files
that
could
be
persistent
on
timeline.
So
this
does
not
resolve
the
problem
at
all.
So
we
would
like
to
look
at
this
from
a
different
angle.
That
is
the
one
that
we
will
present
in
the
next
slides.
So
next
one.
J
F
J
J
We
will
go
now
now
to
the
next
two,
so
the
service
assurance
and
our
slicing
to
have
more
detail
on
that
service
assurance.
You
all
of
you
are
aware
of
the
the
draft
from
benue
that
has
been
presented
several
times
here
about
the
service
and
syrian
architecture.
So
basically,
this
this
graph
proposed
a
cross
control
loop
that
could
modify
the
the
configuration
in
real
time
the
running
configuration
according
to
different
intents
and
operational
states.
J
Regarding
slicing
here,
the
the
concept
also
applies
straightforward,
in
the
sense
that,
in
the
slice
concept,
we
will
have
either
slices
being
hierarchical
so
covering
more
than
one
layer
or
even
on
the
same
layer,
basically
taking
profit
of
the
recursivity,
so
one
slice
could
be
sliced
at
the
same
time.
Okay
and
now
also
the
the
the
sequential
slicing
so
the
stitching
of
the
slices,
and
so
the
multi-level
configuration
here
could
be
easily
us
as
a
point
in
order
to
understand
that
each
of
the
slices
that
we
will
deploy
will
be
essentially
one
of
these
levels.
J
J
So
here
you
can
see
a
an
easy
example,
so
you
we
could
have
a
baseline,
an
underlying
infrastructure.
That
could
be
the
figure
at
the
bottom
and
on
top
of
that,
trigger
to
deploy
different
slices.
So
do
each
of
these
slices
will
be
represented
by
these
different
levels
of
configuration.
So
we
have
there,
the
blue
on
the
left
and
the
red
on
on
the
right.
So
this
would
be
essentially
two
level
of
configurations
on
top
of
the
base
configuration.
J
That
would
be
the
one
marking
black
at
the
bottom,
so
essentially
the
black
one
would
be
the
persistent
in
case
of
whatever
outage
or
issue
in
the
network.
We
will
recover
that
black
one
while
the
red
and
the
blue
could
be
later
on
deployed
on
top
of
the
of
the
black
in
an
easier
way.
Next
slide,
please.
J
So
the
proposal
essentially
is
a
visually.
You
can
see
what
would
be
the
idea
on
the
figure
so
essentially
having
these
different
levels
of
configuration,
and
you
can
associate
that
to
multiple
panes
of
glass
in
such
a
way
that
you
could
start
doing
incremental
things
linking
these
things.
Associating
these
things
to
the
multiple
levels
in
such
a
way
that
you
can
grow
or
scale.
J
I
mean
a
skill
up
or
scale
down
into
the
configuration,
depending
on
yeah
as
long
as
the
time
and
being
sure
that
the
persistent
information
is
is
kept,
and
so
how?
Even
though
we
would
keep
you
having
this
plain
figure
to
start
linking,
let's
say
these
multi
levels
and
such
however,
we
can
operate
in
a
in
a
more
easiest
way.
Having
so
divided
the
problem
essentially
and
having
this
incremental
approach,
each
level
can
be
handled
in
an
independent
manner.
J
So
we
will
minimize
the
impact
in
other
levels,
and
this
applies
fully
for
the
case
of
a
slicing,
for
instance,
and
each
level
will
be
autonomous
and
comparative,
and
this
following
applies
a
very,
very
straightforward
to
the
service
assurance
case
as
well
as
the
other
two.
The
mergings
and
and
the
cello
touch
for
the
mergers,
for
instance,
you
can
see
over
there
or
the
migration.
J
J
So
we
have
presented
this
to
the
routine
market
group
as
well
as
also
here,
so
we
expect
to
work
in
completing
the
description
of
the
proposals
and
identify
viable,
viable
solutions
and
approaches.
So
this
is
yet
an
idea.
We
need
to
land
this
idea
to
something
much
more
concrete
and
for
sure
we
would
like
to
collect
feedback
and
comments
from
from
the
research
group
and
for
sure
we
will
prepare
a
new
version
for
next
itf.
F
G
Hi,
just
a
quick
question
is
it
might
be
worth
presenting
this
also
in
net
mod,
and
I
wanted
to
ask
whether
you're
aware
of
the
nmda
architecture,
the
network
data
store
architecture
used
for
netconf,
because
that
already
had
the
idea
of
separate
data
stores
representing
different
types
of
data
within
network
configuration,
and
it
may
be
worth
looking
at
whether
you
could
use
separate
data
stores
to
manage
these
different
layers
and
then
the
interesting
questions
become
about.
G
How
do
you
merge
that
that
configuration
data
together
and
a
separate,
interesting
question
is
about
how
do
you
map
the
operational
state
data
back
again,
given
that
when
the
underlying
device
is
going
to
see
a
merged
view
of
that
configuration?
But
that
might
be
an
approach?
That's
worth
considering.
L
Yes,
hey
hey
robert
on
the
nmda
data
stores,
it
is
fine
as
long
you
don't
try
to
create
management
separations,
because
the
idea
there
is
that
you
can
have
a
completely
separate
owners
from
the
management
perspective
of
parts
of
the
configuration.
So
we
looked
into
that,
but
we
haven't
finalized
it.
But,
based
from
from
the
initial
looks
what
we
saw
that
there
are
some
there
are
some
missings
in
the
nmda
that
allowed
complete
the
figuration
management
operation.
G
So
so
dan,
just
to
reply
to
that,
is
that
things
like
the
nakam
side,
the
access
control
that
you
have
a
particular
concern
about.
L
F
Okay,
thank
you.
Thank
you
for
the
the
comments
and
questions
and
thank
you
luis
again
for
the
presentation.
So,
let's
now
move
to
the
the
last
meeting.
That
is
an
update
on
the
concept
of
digital
network.
So
I
think
the
promotion
is
done
by
a
zoo
and
jiggle
right.
B
E
Extra
hi
cheers
diego,
I
think
I
can
prepare
most
of
them
and
how
ask
dave
to
to
free
to
present
slides
three
or
n4
okay.
E
E
Okay
in
meeting
108,
we
introduced
the
initial
version
of
the
draft
in
including
the
definition
of
benefits
and
the
challenges
of
digio2
network.
Since
last
update,
there
are
three
major
updates
from
initial
version.
E
The
first
two
updates
were
provided
by
diva
locus
and
antenna
of
pasteur
from
telefonica,
and
they
are
about
adding
refine
any
annual,
fixed
element
to
the
klm
names,
orchestration
and
added
two
more
benefits
in
section
three
and
the
third
and
the
third
updates.
Is
we
added
a
new
section
for
reference
architecture
of
digital
network?
E
So
diego,
I
do
your
flow.
Thank
you.
B
B
B
B
Brief
of
the
of
the
twin
so
well
on
one
hand
you
you're
able
to
to
warranty
repeatability,
so
you
can
replicate
network
conditions
on
demand,
don't
necessarily
exactly
the
same
that
are
happening
in
a
certain
moment,
so
you
can
have
several.
As
I
said
several,
let's
say
individual
twins
and
you
have
the
possibility
of
reproduce
by
replaying
the
section
of
events
under
control
variations.
It
is
that
you
have
a
much
better
with
it
with
the
train.
B
You
have
a
much
better
way
of
managing
or
dealing
with
different
choices,
alternatives
or
or
or
situations
that
may
happen,
and
for
sure
to
for
these
to
be
useful
is
something
that
you
need
a
good
network
model
that
allows
you
to
make
this
orchestration
in
a
reasonable
time.
Next,.
B
One
so,
apart
from
that,
when
talking
about
the
digits
of
doing
we,
we
found
and
some
internal
discussions,
that
there
were
other
at
least
two
potential
benefits
for
the
moment.
We
probably
I
don't
know
if
we
with
the
as
the
work
go
on,
we
will
identify
another
ones.
B
Basically,
one
is
about
privacy
preservation,
everything
that
happens
in
the
dream,
every
every
processing
that
you
may
have
in
the
train,
with
the
external
ally,
with
external.
B
B
The
data
apply
the
knowledge
without
violating
privacy
requirements,
and
and-
and
this
is
something
that
as
well,
is
connected
as
well
with
the
idea
that,
for
example,
when
dealing
and
this
a
concern
in
natural
management
right
now
is
about
what
happens
when,
when
end-to-end
encryptions
become
completely
progressive
and
via
and
the
network
management
mechanisms
are
become
a
little
bit
a
little
bit
quite
blinder
than
they
are
right
now,
and
so
with
this
idea
of
adding
or
enhancing
the
training,
with
the
possibility.
B
B
We
are,
we
are
using
it
in
well,
we
are
developing
a
project
for
doing
with
cyber
ranges
in
which
people
may
play
at
the
different
conditions
and
security,
evaluations
and
all
the
like,
and
we
have
as
well
a
devaluation
of
different
chests
and
I
hand
it
over
to
you.
Sorry.
A
Yeah,
you
may
need
ready
to
speed
up
because
the
session.
B
E
Okay,
okay,
let's
might
take
the
turn,
let
me
type
I'll
take
this
slides
shows
the
reference
tech
architecture
of
gtn
system
and
it
can
be
designed
as
a
three
layer
system.
The
bottom
layer
is
a
physical
network.
E
Oh
yes,
and
our
bottom
layer
is
physical
network.
Physical
network
can
be
of
wears
type,
including
telecommunication
operator
network,
the
sensor
network,
campus
network,
etc,
and
all
network
elements
and
and
equipments
in
physical
network
exchange,
massive
network
data
and
the
control
messages
with
digital
twin
entity
via
southbound
interface.
A
middle
layer
is
network,
digital
twin.
E
E
Then,
after
four
or
four
full
verification,
the
change
controls
can
be
deployed
safely
to
the
physical
network
under
the
top
layer
is
a
network
application
layer
of
both
conventional
and
innovative
applications
can
run
against
the
dtm
platform,
with
low
cost
and
less
service
impact
on
real
network
network
management.
Network
applications
provides
requirements
to
network
digital
training
entities
while
northbound
interfaces
next
page.
Please.
F
E
Okay:
let's
go
to
the
final
final
page,
okay,
so
for
next
next
step,
we
are
going
to
consider
the
consider
dynamic
data
collection
through
day
day
and
orchestration
with
illustration,
by
arches
by
spider
project,
and
we
have
to
analyze
requirements
on
flow
problems,
and
it
is
chased
by
another
inspire
file,
g
plus
project
and
well
to
investigate
more
use
cases
and
requirements
and
to
define
basic
suspense
and
not
bound.
Interfaces
of
dtm
system
welcome
to
join
our
work
and
any
comments
are
welcome.
Thank
you.
F
Thank
you.
Thank
you,
indigo
for
the
presentation.
Unfortunately,
we
cannot
take
the
comment
and
question
now
I
have,
since
there
is
one
on
there
and
jabber,
so
we
may
also
have
a
look,
so
let's
continue
the
discussion
over
the
mailing
list
for
the
sake
of
time
so,
and
also
I
want
to
thank
you
all
for
presenting
and
participating
to
this
session,
so
we'll,
of
course,
publish
minutes
and
we'll
continue
the
different
discussions
that
we
unfortunately
have
to
pick
been
shortened
today
over
the
mailing
list
and
other
meeting.
Thank
you
all.