►
From YouTube: CI WG demo: Big data Benchmarking and Surrogates
Description
Geoffrey Charles Fox (http://www.dsc.soic.indiana.edu/, gcf@indiana.edu)
Fox received a Ph.D. in Theoretical Physics from Cambridge University, where he was Senior Wrangler. He is now a distinguished professor of Engineering, Computing, and Physics at Indiana University, where he is the director of the Digital Science Center. He previously held positions at Caltech, Syracuse University, and Florida State University after being a postdoc at the Institute for Advanced Study at Princeton, Lawrence Berkeley Laboratory, and Peterhouse College Cambridge. He has supervised the Ph.D. of 73 students and published around 1500 papers (over 540 with at least ten citations) in physics and computing with a hindex of 82 and over 38500 citations. He received the High-Performance Parallel and Distributed Computing (HPDC) Achievement Award and the ACM - IEEE CS Ken Kennedy Award for Foundational contributions to parallel computing in 2019. He is a Fellow of APS (Physics) and ACM (Computing) and works on the interdisciplinary interface between computing and applications. He is involved in several projects to enhance the capabilities of Minority Serving Institutions. He has experience in online education and its use in MOOCs for areas like Data and Computational Science. He is active in the Industry consortium MLPerf.
Date: 09/04/20
Presenter: Geoffrey Fox
Institution: Indiana University
Midwest Big Data Hub
B
A
So,
as
many
of
you
know,
I'm
john
goodhue,
I'm
in
the
steering
committee
for
the
big
data
hub
infrastructure
working
group
along
with
christine,
and
I
think,
jim
and
others
each
time
we
meet,
we
we
do
do
a
lightning
talk
and
then
then
some
additional
business.
This
this
week
we
are
lucky
to
have
jeffrey
fox,
who
is
going
to
talk
about
a
new
project.
A
That's
just
starting
up
having
to
do
with
fair
data
in
artificial
intelligence,
which
is
convergence
of
two
important
topics
to
us
in
many
ways
that
jeffrey
needs
no
introduction,
but
I'll
I'll
give
it
a
try.
He
got
his
phd
many
moons
ago
at
cambridge
university
he's
currently
at
indiana
university,
but
his
path
from
there
to
here
has
included
caltech
where,
as
I
mentioned
earlier,
I
I
I
knew
him
as
the
co-inventor
of
the
cosmic
cube,
which
in
in
many
ways
is
the
grandparent
of
every
cluster
on
the
planet.
A
Syracuse
university
florida
state
institute
for
advanced
study,
lawrence
berkeley,
laboratory,
peterhouse
college
in
cambridge,
a
trail
of
phd
students
and-
and
you
know
more
than
a
thousand
pages
of
papers-
has
brought
a
lot
to
the
world,
and
I'm
I'm
pretty
interested
to
hear
what
he
has
to
say
so
I'll
stop
it.
There.
A
A
B
B
So
I'll
discuss
surrogates
first,
and
here
is
a
surrogate
and,
in
my
opinion,
surrogates
will
really
have
this
very
significant
impact
on
all
aspects
of
high
performance,
computing
and
general
computing,
because
they
allow
you
to
replace
large
simulations
by
a
a
network
that
has
learned
the
results
of
those
simulations
and
it
can
be
done
in
in
many
ways.
Here's
actually
one
a
pretty
simple
way
that
I
did
with
a
colleague
in
a
nanoparticle
simulation,
a
strong
ionic
structure
in
a
combined
area,
and
it
uses
this
one.
B
This
particular
one
used
a
fully
connected
network
shown
up
in
the
top
right,
and
I
wouldn't
be
talking
unless
it
gave
essentially
perfect
answers
and
those
answers
are
served
on
the
bottom
right
and
when
I
was
at
caltech,
we
discussed
speed
up
and
this
speed
up
is
a
million
at
least
once
you've
overcome
the
overhead
of
making
the
training
set.
So
you
have
to
run
five
thousand
simulations,
which
takes
whatever
time
five
thousand
simulations
run
take.
B
B
B
B
I
so
coming
back
to.
When
I
was
at
cambridge
there
I
had
some
there.
I
learned
that
you
should
never
have
a
high
parameter
approximation.
You
should
use
simple
low
order,
approximations
well.
This
neural
net
has
65
000
parameters,
and
so
it's
pretty
complicated,
but
it's
highly
robust,
deep
learning
networks
are
incredibly
robust
compared
to
more
traditional
approximation
methods,
and
here
we,
I
won't
bother
with
this
one.
As
I
have
too
many
slides,
this
shows
you.
B
It
has
73
companies,
10
universities,
including
all
the
major
companies
and
his
mission,
is
to
build
fair
and
useful
benchmarks.
Although
fair
as
used
here,
is
not
fair
as
we
we
wanted.
We
want
to
build
in
this
doe
project
and
it
has.
This
is
a
very
large
effort
by
some
standards.
It
has
50
ftes
working
on
it,
spread
over
these
73
companies,
and
it
is
basically
established
by
these
companies
to
document
the
performance
of
important
ai
and
there
are
a
whole
set
of
working
groups.
I
go
to
many
of
these.
B
B
B
B
And
well
these
I
like
this
picture
here
whether
which
is
part
of
their
publicity,
I
should
say
mlperv-
is
likely
to
be
renamed
ml
commons
and
have
a
more
formal
non-profit
like
status,
but
they
want
to
combine
benchmarks
and
relevant
view.
Data
sets
and
also,
I
think,
relevant
best
practices,
because
I
see
this
collection
of
benchmarks
so
because
every
entry
there
has
a
data
set
and
a
reference
implementation
and
a
submitted
implementation.
B
B
Every
single
submission
is
a
deep
learning
submission,
not
not
other
parts
of
machine
learning,
and
we
also
want
to
involve
things
like
end-to-end
benchmarks,
involve
the
edge
to
the
data
center
issues
and
again,
like
the
ml
perf,
involve
inference
and
training
for
the
science
case,
and
there
is
sort
of
interesting
similarities
because
science
has
images
and
but
things
like
say
the
field.
I
was
originally
trained
in
particle
physics.
That
analysis
doesn't
have
an
obvious
analogy
in
industry
and
also
surrogates
are
not
used
in
industry.
B
Here
is
some
early
results
from
tony
has
actually
already
got
some
science
benchmarks
because
he's
part
of
his
group,
but
rather
for
that,
has
actually
got
a
core
effort
on
producing
better
science
benchmarks
and
all
science
benchmarks
have
open
source
data
sets
labeled
data
plus
these
reference
implementations,
so
they're
pretty
useful,
and
there
is
a
link
here
which
you
can.
These
slides
are
available.
They're
all
online
on
as
a
google
slide,
so
you
can
capture
that
link.
B
Here
is
an
illustration
of
why
we
need
some
best
practice,
so
there
was
over
the
summer.
I
happened
to
be
looking
at
hydrology
and
there's
a
paper
of
a
pretty
useful
review
paper
in
hydrology,
which
says
there
are
129,
deep
learning
papers,
studying
rainfall,
runoff
and
terrain
structure
and
stuff
like
that.
B
So
this
illustrates
that
advancing
in
these
fields,
like
deep
learning
for
hydrology,
it's
pretty
important
to
have
some
reference
implementations
of
related
problems
which
in
this
case
a
geospatial
time
series
which
are,
I
said,
the
side
comment,
I
would
say
deep
learning
is
revolutionizing
the
geospatial
time
series
a
really
dramatic
improvement
of
the
previous
methods
anyway.
So
this
is,
I
think,
this
to
this
particular.
This
just
illustrates
why
best
practice
is
important.
B
B
All
right
so
here's
this
so
this
slide
here.
Let
me
talk
about
one
important
aspect
of
this,
so
ml
commons
or
mlpervis
produ
has
a
project
which
I've
been
working
with
them
on
called
ml
box,
and
mr
box
is
the
universe
which
I
call
the
interoperable
interface
to
the
data
sets
and
and
deep
learning
networks
which
are
which
are
associated
with
every
one
of
those.
Mr
perth
instances,
and
so
we
want
to
take
in
our
box
and
I've
been
working
with
ml
perf.
B
That's
an
agent-based
computation,
and
I
call
that
machine
learning
for
hpc
and
of
course
we
also
know
hpc
will
accelerate
machine
learning.
I
recommend
that
we
look
at
mlperf
and
we
need
to
include
surrogates
there
and
we
want
to
not
only
do
benchmarking.
We
want
to
use
that
to
produce
best
practice
and
tutorials,
and
I
encourage
you
to
join
us
and
help
us
in
this
in
this
in
this
task,
and
I
think
I
have
a
on
the
next
slide
some
details
about
what
to
do.
B
If
you
want
to
join
mr
path,
you
just
go
to
the
google
group
for
mlperf
and
request
join
then.
If
you
want
to
go
to
a
particular
working
group
like
hpc
or
science
data,
you
just
go
to
the
associated,
there's
a
link
for
every
working
group
on
the
mlp
site.
You
just
click
it
and
then
you
will
request
to
join
and
we
also
would
like
to
help
designing
and
building
surrogates.
B
C
I
I've
got
a
question
thanks
very
much
jeffrey
that
was
a
great
talk
and,
and
just
a
good
good
overview
of
a
really
exciting
area
w.
Just
if
you
could
just
speak
to
the
data
for
a
moment
when
you,
when
you
talk
about
some
of
your
benchmarks,
is
the
group
in
general
sort
of
grappling
with
how
to
make
data
available
for
some
of
the
benchmarks.
C
B
C
Yeah,
I'm
aware
of
some
of
them
and
that's
why
I
thought
this
was
especially
given
the
focus
of
this
group,
which
is
around
sort
of
big
data
and
and
cyber
infrastructure
that
helps
to
address
some
of
these
things
and-
and
I
think
you
know
just
kind
of
wanted
you
to
point
to
those
that
those
the
work
that's
being
done
to
sort
of
address.
Some
of
the
challenges
that
you
see
well,.
B
Like
I
mentioned,
mr
box
is
trying
to
get
some
universal
access
once
the
data
has
been
placed
where
it
has
to
be
placed,
because
I
don't
think
it's
going
to
be
possible
to
have
a
universal
access
mechanism,
because
one
group
of
people
will
will
find
globus
very
convenient
and
another
group
will
not.
I
mean
europe
finds
problems
because
globus
probably
uses
us
us
cloud
sources
which
is
against
your
eu
rules,
and
companies
can't
afford
the
license
which
globus
has
and
things
like
that.
So
that's
that's
a
minor
part,
but
I
think
more
interesting.
B
Actually
I
want
to,
as
I
want
to
make
this
a
good
educational
resource.
I
want
I'm
working
on
getting
a
good
high
performance,
jupiter
notebook
interface
to
all
these
data
sets,
which
is,
in
my
opinion,
I
might
most
of
my
work.
These
days
is
done
from
jupiter
notebooks.
Actually,
most
of
it
is.
You
uses
google
codelab
for
most
of
the
initial
work,
but
so
I
think
it's
very
important.
We
provide
a
good,
convenient
interface.
B
A
B
So
there
are
all
sorts
of
legal
issues,
but
I
think
I
mean
I
mean
I
think
ml
perf
slash.
Mr
commons,
is
the
best
community.
I
have
seen
for
a
long
time.
The
people
in
it
are
really
exciting
and
they're
all
industry,
nearly
all
industry.
People,
though
I
have
a
good
friend
now,
a
good
friend
from
harvard
who
I
met
through
this
process,
who
works
on
works
on
inference
and
edge
devices
in
the
language
area.
B
But
these
people
are
very
free
within
with
their
wisdom
and
ex
and
expertise,
and
I
find
their
working
groups
to
be
very
fruitful
because
they
are
the
people
involved
with
mr
puff
are
people
from
industry
who
have
both
practical
experience
and
they
have
somehow
the
time
to
do
these
quite
difficult
leading-edge
results,
because
these
benchmarks,
like
there's
dlrm,
which
is
facebook's
recommender
engine,
that's
a
very
sophisticated
piece
of
code,
as
are
the
obviously
these
giant
image,
processing
and
language
processing
networks
and
the
people
know
about
that.
And
there
are
there's
huge
emphasis.
B
There's
huge
effort
from
say,
nvidia
all
about
hpc.
I
mean
the
hpc
component
of
this
effort
is
very
high
because
there
is
an
explicit
hpc
working
group,
but
that's
really
not
an
hpc
working
group.
Although
that's
what
it's
called.
It's
really
a
doe
lab
supercomputer
working
group
and
the
eight
the,
but
every
single
working
group
is
all
devoted
to
hpc,
because
you
can't
get
good
performance
without
hpc,
whether
it
be
a
tpu,
a
gpu
or
a
or
a
scanless
chinese
accelerator.
It's
all
hpc.
D
Thank
you
for
that
great
presentation,
yeah
and
I
had
a
couple
questions
I
I
could
probably
take
the
whole
rest
of
the
time
so
I'll
just
try
and
be
disciplined
and
not
self-indulgent.
In
my
questions,
I
was
wondering
if
the
any
of
the
working
groups
have
explored
the
differences
and
storage
types,
especially
shared
storage
versus
attached
storage
and
ssds.
B
Well,
I
would
say
actually
when
I
was
in
the
best
practices
working
group
today-
and
there
was
some
complaint
from
red
hat-
that
the
storage
issues
were
not
being
properly
addressed,
and
I
would
say
that
is
going
to
get
more
attention.
But
it's
just
slightly
related
to
the
rules,
because
the
current
rules
are
quite
careful
of
specifying
where
the
data
is
allowed
to
be
and
what
you
actually
time
and
they
don't
because
they
want
the
timing
to
reflect
the
machine.
B
The
deep
learning,
not
necessarily
the
time,
spent
getting
the
data
from
point
a
to
point
b,
but
I
suspect
what
will
happen
is-
and
I
certainly
think
we'll
do
that
in
the
science
data
working
group.
We
will
make
the
metrics
composite,
so
the
metrics
will
include
various
steps
in
the
process
because
we
know
we
all
know
that
the
end
am
process
of
going
from
taking
the
data
to
the
result
is
a
very
complex
process
with
many
steps,
and
I
think
we
intend.
B
E
D
Great
and
that's
that's
my
interest
in
personal
research
area
as
well,
and
that's
why
I
ask
is
that
we
also
see
as
infrastructure
providers
a
lot
of
requests
to
say
this
isn't
running
fast
enough.
I'm
gonna
need
a
whole
bunch
of
ssds
locally
to
and
it's
like.
Well,
okay,
I'm!
I
could
buy
those,
but
I
would
like
to
understand
if
that's
going
to
help
your
problem.
B
Yes,
if
you
joined
the
the
science
data
working
group,
we
could
certainly
have
that
discussion.
I
think
that's
very
certainly
in
scope,
because
we
want
to
do
end-to-end
and
and
such
discussions
about
where
what
discs
of
what
performance
you
need
in
the
various
stage
of
this
analysis,
and
also
just
some,
I
don't
see
that
much
work
on
how
to
run
juvenile
notebooks
to
run
over
cut
large,
large,
deep
learning
environments
with
different
network
different
storage
devices
and
where
they
have
to
be.
It's
not
very
well
discussed.
E
D
B
Yeah,
so
again
we
can
probably
do
that.
I
think
I
am
the
main
the
person
pushing
fair
principles
on
mo,
I
joined
the
ml
box
working
group
to
try
to
see
how
to
add
fairness
to
what
they
were
doing
and
unfortunately,
my
you
know
my
students
disappear
over
the
summer
and
go
to
work
for
microsoft
and
facebook
and
people-
and
I
just
got
my
student
back
and
so
he's
now.
He
will
be
working
on
that.
Hopefully,
this
semester.
D
A
B
B
There
was
an
interesting
paper
from
north
carolina
biology
paper,
which
showed
that
they
actually
had
they're
the
largest
training
sample.
I
know
for
a
circus,
they
had
a
hundred
thousand
and
they
showed
that
their
sim,
their
their
deep
learning
network,
which
they
trained
learnt
scenarios
which
they
didn't
train
on,
and
it
was
so.
It
was
sort
of
interesting
they
found
some.
B
You
just
take
your
network
and
randomly
remove
some
fraction
of
the
of
the
nodes,
and
there
was
some
set
of
papers
actually
from
cambridge,
where
there
was
a
phd
cambridge
in
england,
where
I
came
from
a
phd
from
two
or
three
years
ago,
studying
such
a
drop-out
methods
to
get
errors,
but
I
don't
there
is
no
there.
This
is
a
wide
open
research
area
and
obviously
important.
B
By
the
way
I
have
another,
I
mean
I've
been
having
a
big
debate
within
this
air
because
they
keep
test.
Oec
keeps
turning
down.
My
proposals
is
to
whether
surrogates
are
cyber
infrastructure
or
not,
because
when
I
started
cyber
infrastructure,
you
know
we
said
well,
we
did
hardware,
software
algorithms
and
we
joined
them
all
together,
and
that
was
the
study
we
did
for
hpc,
but
now
they
are
defining
cyber
infrastructure
to
exclude
surrogates
effectively
because
they
exclude
them
because
surrogates
involve
changing
the
application
code.
B
They
are,
however,
in
my
opinion,
quite
generic,
so
they're
like
an
applied
math.
It
was
like
multi-grid
or
something
it's
like
a
technique,
that's
a
broadly
applicable
technique,
but
can't
be
sort
of
managed
by
a
supercomputer
center.
It
has
to
be
learned
by
the
user
and
put
in
and
supported
by
the
supercomputer
center
and
surrogates
are
the
same.
There's
important
support
software
for
surrogates,
but
you
can't
actually
do
a
surrogate
unless
you
involve
the
application
people
to
understand
the
nature
of
the
differential
equations
and
what
have
you
for
the
simulations
anyway?
B
B
Should
I
say
I
have
written,
probably
four
proposals
on
surrogates
and
this
one
from
doe,
which
involved
benchmarking.
Surrogates
is
the
only
one
that
was
funded.
The
previous
ones,
which
are
nsf
were
all
turned
down
for
no
trivially
good
reason.
I
mean
they
would
just
turn
down
because
they
had
better
proposals,
of
course,
but
I
think
a
lot
of
the
problem
is
this
field
is
not
well
and
well
recognized
as
a
field
in
the.
In
my
opinion,
the
promise
is
greater
than
p.
B
A
B
B
C
F
A
I
think
we've
all
been
getting
messages
from
zoom,
saying
everything's
going
to
change
in
october,
at
least
for
for
me.
I
have
no
idea
what
that's
going
to
mean,
but
in
our
case
there
will
be
a
waiting
room
on
this
meeting.
So
don't
don't
feel
don't
feel
put
off
if
you
see
a
waiting
room.
The
next
in
our
next
meeting-
and
we
have
next
next
meeting-
will
be
jim
bosni
from
the
trustworthy
data
working
group
and
he'll.
A
I
think
he's
been
at
this
session
before,
and
this
will
be,
I
think,
a
maybe
nine
month
update
on
what
they're
doing
and
and
how
it's
going
in
the
so
going
back
to
one
other
topic.
Let's
just
do
a
quick
vignette
from
each
of
the
hubs
and
I'll
just
point
at
people
and
and
if
you
can
talk
for
for
maybe
a
minute
or
so
about
how
things
are
going
and
we'll
we'll
go
from
west
to
east.
So
I'll
start
with
christine.
D
We
are
partnering
with
the
border
solutions
alliance,
which
has
five
universities
on
the
u.s
side,
and
so
this
actually
includes
a
little
bit
of
the
south
hub
as
well,
and
we
had
our
partner
matching
event.
Yesterday,
people
are
choosing
teams,
and
this
is
like
you
know.
Sometimes
you
know
when
you're
working
on
a
hub
project
or
like
jeffrey
fox's
presentation
today
and
you're,
like
I'm
so
lucky
to
be
on
this
project.
This
is
one
of
those
things.
D
Yesterday
we
had
people
on
from
tijuana
from
juarez
who
are
partnering
with
people
on
the
u.s
side
to
try
and
look
at
cross-border
stuff,
also
trying
out
different
approaches
about.
Looking
at
you
know,
are
there
high-risk
areas
in
a
certain
city,
because
that's
where
all
the
visitors
are
just
some
really
fun
projects?
A
E
G
So
you'll
hear
more
about
both
of
those
projects
coming
up
and
feel
free
to
to
join
that
group.
If
that's
of
interest
to
you
as
well,
it's
been
a
busy
summer
here
as
well,
and
so
we,
like
some
of
the
other
hubs,
have
kicked
off
our
seed
funding
programs
this
year
and
we
have
a
great
group
of
projects
that
will
be
coming
online
later
this
year.
G
I
think
those
are
the
main
topics.
I'm
sure
that
florence
will
probably
talk
about
the
covet
information
commons
project
so
I'll
just
say
in
advance
of
that
that
we're
leveraging
that
information
to
do
some
of
the
science
writing
that
christine
talked
about
in
terms
of
outreach
to
the
the
awardees
from
those
nsf
awards
to
try
to
get
them
partnered
up
with
our
priority
area.
Folks,
as
well,
so
trying
to
do
some
community
building
through
the
use
of
that
resource
that
the
hubs
have
collaborated
on.
C
So
also
in
the
in
the
midwest
and
just
I,
I
pasted
a
link
in
the
chat
for
people
who
are
interested
in
agricultural
data
coming
up
here
september.
24
25
is
unlocking
the
data
revolution
meeting
and
you're
all
welcome
to
join
that
the
link
to
registration
again
is
included
in
the
chat,
and
this
is
far
and
a
number
of
other
people
are
participating
in
this
event
and
giving
talks
from
universities
all
over
the
nation.
But
I
think
there's
a.
C
Quasi
is
not
participating
in
this,
although
they
have
sort
of
a
role
now,
because
we've
been
talking
with
them
for
a
while,
and
so
I'm
looking
at
the
agenda
because
we
get
these
things
mixed
up.
We
just
had
the
water
resources
event
where
we're
working
through
with
quasi
their
role
on
some
of
the
things
that
they're
presenting,
but
I
don't
believe
they're
explicitly
a
part
of
this.
E
E
We
launched
our
seed
fund,
we're
doing
two
rounds
or
two
tranches
as
if
we're
acting
like
a
bc,
and
so
the
first
round
just
closed
and
the
next
one
will
close
october
1st
and
some
of
the
I
love
the
proposals
we
get
from
these
researchers,
they're,
amazing
and,
as
part
of
it,
we
also
announced
a
northeast
student
data
core
founding
committee
and
the
idea
there.
If
you
go
to
our
website,
I
can
I'll
post.
Something
in
the
chat
here
is
to
create
a
student
data
core
to
work
with
professors
and
industry.
E
Ibm
wants
to
help
us
put
the
content
and
pedagogy
together
to
teach
data
science
to
underserved
institutions
and
students
throughout
the
northeast,
so
students
teaching
students
creating
a
community
of
data
science,
as
well
as
an
outreach
committee
to
find
institutions
that
either
want
to
be
learners
or
teachers
in
this
community
and
find
student
internship
opportunities.
So
we're
really
excited
about
that.
So
we've
launched
a
fa
we're
asking
for
participants
in
the
founding
committee
so
that
the
community
can
help
us
design
it,
which
we
really
want
to
do.
E
Regarding
the
code
info
commons,
it's
the
gift
that
keeps
on
giving
it
keeps
on
doing
stuff.
We
asked
for
pis
to
present
in
our
launch
and
we
begged
peter
rose
and
ilya
from
ucsd
and
then
41
other
pis
offered
to
present.
So
now
we're
having
a
series
of
pi,
lightning
talks
and
our
september
16th
is
the
next
one,
then
october
16th,
so
that
went
on
in
our
newsletter.
E
Today
we
have
api
is
doing
five
minute:
lightning
talks,
plus
q,
a
we're
developing
a
student
hackathon
competition
leveraging
the
covet
info
commons,
which
is
really
exciting.
We
were
talking
to
our
student
on
our
staff,
our
part-time
student
today
she's
putting
the
whole
thing
together
and
I
think
we're
going
to
have
her
do
a
little
demo
video
to
announce
it.
I'm
really
excited
about
that
students
talking
to
students
and
we
announced
a
call
for
participation
in
a
connected
healthcare
cybersecurity
workshop
we're
going
to
do
in
collaboration
with
an
ieee
ul
working
group.
E
I
lead
on
clinical
iot
data
and
device
interoperability
with
tips,
and
we
announced
our
call
for
participation
for
the
urban
tour
communities
working
group
as
well,
so
it's
been
busy
and
then
we
have
pi
stories.
Like
john,
does
he
thought
up
the
idea
and
we're
doing
it
too.
So
thank
you
for
that
and
we
call
it
five
questions
with
coveted
researchers
and
I
think
we
have
about
a
dozen
researchers
that
we're
highlighting
so
far.
C
C
C
We
also
kicked
off
a
education
data,
science,
education
and
workforce
working
group
today
at
11
o'clock,
and
we
had
a
number
of
participants
there
and
two
great
talks,
and
that
happens
on
the
first
friday
of
every
month
going
forward
and
we've
got
a
calendar
of
12
speakers
for
the
coming
12
months.
So
we're
very
excited
about
that
and
I
guess
the
last
thing
is
just
a
quick
shout
out
and
a
welcome
to
martin
who's
here
from
nsf.
Thank
you
for
joining
us,
martin.
C
A
Is
maybe
echoing
something
that
somebody
said
earlier
is
this
working
group
has
been
around
for
pretty
much?
The
span
of
you
know
both
branches
of
the
hub
program
and
it's
been
fun
to
look
across
the
hubs
to
watch
them
mature
and
it
feels
like
there's.
Maybe
this
exponential
phenomenon
of
for
a
long
time
you
double
and
double
the
double,
and
you
can't
don't
see
any
difference.
Then.
All
of
a
sudden,
you
start
to
see
an
inflection
point
feeling
like
a
bit
of
an
inflection
point.
A
F
Sure,
thanks
john
I'm
just
going
to
drop
something
in
chat
and
sadly
I
can't
do
two
things
at
the
same
time.
So
I'm
going
to
do
that.
First,
okay,
so
yeah,
thanks
before
I
launch
into
the
osn
update
just
a
very
quick
note
about
this
year's
adsa
data
science.
Leadership
summit
will
be
happening
in
october
and
I'm
really
pleased
so.
F
Several
of
the
hub
leaders
are
part
of
the
program
committee
this
year,
but
I'm
also
really
pleased
just
to
note
that
there
is
a
an
entire
session
as
part
of
this
summit
this
year
on
collaborating
across
campus
to
build
infrastructure
for
data
sharing.
So
I
think
that's
a
really
great
thing
and
we're
seeing
some
inroads
in
that
space.
Around
data
sharing
cyber
infrastructure,
which
is
great
okay,
so
osn
preparing
now
a
sort
of
new
year,
kickoff
outreach
to
our
current
user
groups.
F
So
that's
getting
started
up
and
that's
exciting
and
then
the
other
new
use
case
is
cdms
and
cdms,
for
those
of
you,
that
don't
know
are
a
multi-institutional
project
on
cryogenic,
dark
matter,
search
so
cdms
and
that
allocation
will
be
held
at
illinois,
but
help
move
data
around
from
that
very
large
and
distributed
project.
So
those
are
a
couple
of
our
new
use
cases.
F
What
I
dropped
into
the
chat
is
that
this
fall
as
part
of
our
coming
together
to
look
at
lessons
learned
and
growth
and
status
of
the
osn
in
preparation
for
the
proposal
that
we
are
planning
to
put
together
this
year
for
scaling
the
osn,
we're
going
to
run
a
webinar
series,
and
that
starts
october
22nd
with
research,
drivers
and
software.
So
those
will
again
be
our
our
stakeholder
user
community.
F
So
some
of
the
use
cases-
and
then
also,
I
believe,
a
couple
of
the
software
front
ends-
are
third-party
collaborators
that
are
helping
provide
interface
to
the
data
on
osn.
You
see
the
list
there
and
the
topics
we
don't
have
the
time
set
up
yet,
but
those
are
the
dates
right
now
planned
and
then
what
will
happen
from
those
webinars
is
that
concept
papers
will
be
produced
out
of
those
each
of
those
webinars
they'll
provide
input
to
a
final
paper
around
the
osn
and
help
us
get
ready
for
the
next
proposal.
F
F
A
This
is
maybe
a
plug
for
next
month's
talks.
One
of
the
most
noticeable
drivers
for
for
policy,
particularly
security
policy,
was
an
engagement
with
the
trusted
ci
working
group,
which
their
checklist
and
the
engagement
really
helped
us
to
kind
of
shape,
best
practices
there
and
not
have
to
invent
them
from
from
scratch.