►
From YouTube: DevoWorm (2021, Meeting 5): Advances in DevoLearn, VGG/Style Transfer, Systems Development
Description
Attendees: Advances in DevoLearn and DL Lightning Talk, Deep Learning and Art (Style Transfer using ResNet), Systems Development in Model Systems (C. elegans and Organoids). Susan Crawford-Young, Mayukh Deb, R Tharun Gowda, Bradly Alicea, Krishna Katyal, Mainak Deb, Jesse Parent, Assaf Wodeslavsky, and Shruti Raj Vansh Singh
A
B
A
Hi
looks
like
a
surface
here,
I'm
going
to
listen
in
the
background,
as
I
will
be
working
good
hello,
my
knock
hello.
How
are
you
my
yoke
that
is
live.
A
I
don't
know
if
he's
on
well
anyways
welcome
to
the
meeting,
so
I
I
guess
surety
was
supposed
to
be
here
and
she
was
going
to
give
a
talk.
I
don't
know
if
she's
gonna
make
it
for
today,
but
we'll
see
I
was
on.
I
think
machine
learning
art
so
we'll
see
if
if
she
comes
by
and
shows
up
so
this
week,
so
we
already
started
our
google
summer
of
code
applications
for
this
year
and
we're
getting
some
good
feedback
we're
getting
some
interested
people.
A
Let
me
go
through
some
of
this
and
then
we'll
talk
a
little
bit
more
with
mike
about
it.
So
let
me
share
my
screen.
A
Okay,
so
the
first
thing
to
announce
is
that
diva
learns
0.2.3
is
now
out
and
it's
ready
to
download.
We've
been
re,
like
I
said
in
some
of
the
previous
meetings.
We've
been
releasing
versions
of
this,
so
this
is
the
diva
learns
the
core
and
evil
learn
software,
and
I'm
going
to
talk
a
little
bit
about
it
in
the
meeting
or
yeah.
I'm
going
to
talk
about
it
a
little
bit
later
in
the
meeting
when
I
go
over
the
flash
talk,
I'm
going
to
be
giving
for
the
education
conference
this
week.
A
So
this
is
the
diva
learn.
2
0.2.3
was
released
february
6th.
So
congratulations
to
mayak
for
getting
that
out.
So
we
have
on
the
github
repo.
We
have
22
stars,
13
forks
and
one
open
issue
pull
request.
A
So
this
is
just
always
refreshing.
The
ideas
is
that
we
have
a
lot
of
activity
going
on,
and
so,
if
you've
not
seen
the
notebook
for
the
d
will
learn
package.
A
You
know
it's
it's
a
program
that
allows
you
to
segment
cells
and
analyze
the
data
that
is
outputted
from
the
segmentation
process,
so
you
can
plot
them
in
a
three-dimensional
plot.
You
can
get
a
distance
matrix
and
then
you
can
try
different
versions
of
you
know:
different
techniques
for
segmenting
things,
and
so
this
is
an
importer
in
python
and
you
can
run
it
and
the
instructions
are
here
and
we
also
have
packaged
with
this.
Some
pre-trained,
gans
and
minok
has
contributed
some
additional
training
to
this
model.
A
So
it's
a
pre-trained
machine
learning
model
that
allows
one
to
identify.
It
makes
it
easier
to
identify
cells
in
an
embryo.
So,
let's
see
then
the
pull
requests.
So
we've
had
a
number
of
pull
requests
in
the
past
few
weeks
we've
had,
let's
see,
we've
had
quite
a
few.
Actually,
let's
see,
let
me
get
up
to
the
point
where
we
were
so
artharon
gonda
arganwa
he's
been
contributing
to
this
repo
he's
made
about
six
pull
requests.
So
thank
you
for
that.
A
That's,
I
think
what
part
of
what
went
into
0.2.3.
So
there
were
a
number
of
support
modules
and
things
like
that
that
went
into
this.
So
this
is
something
that
is,
of
course,
one
of
the
summer
of
code
projects.
So,
if
you're
looking
forward
to
applying
the
summer
of
code,
please
let
us
know
through
the
channels
through
we
have
the
neurostars
channel,
which
is
the
I
think
I
mentioned
this
in
the
email,
which
is
the
sort
of
the
main
incf
portal.
A
So
if
you
go
to
neurostars.org
and
you
put
in
a
build
an
account
there,
you
can
access
the
projects.
I
think
it's
project
series
three.
So
it's
project
3.1
project,
3.2,
project,
3.3
and
all
of
those
are
you
know
we
have
the
devo
learn
project,
which
is
what
I
just
showed
you.
We
have
a
project
involving
segmentation
of
diatom
images,
and
then
we
have
another
project:
digital
microspheres,
which
is
it's
it's
a.
A
We
have
data
from
a
susan's
flipping
microscope
and
that
involves
other
organisms
such
as
axolotls,
so
that's
so
applied
to
those
by.
I
think
the
deadline
is
coming
up
at
the
end
of
march.
You'll
have
to
check
with
the
deadlines
to
be
sure,
contact,
mayoq
or
myself
or
ojawal
singh,
and
especially,
if
you're
interested
in
project
3.2.
You
want
to
contact
those
while
sing
and
we
can
help
you
with
the
application
process.
A
Okay,
so
my
x
says
my
knock,
actually
upgraded
the
gan
so
yeah
it
might
not
upgraded.
The
gan
he's
been
talking
to
me
about
doing
some
training
with
larger
data
sets,
but
I
don't
know
if
that's
been
implemented.
A
So,
oh
there
he
is
what.
A
So
do
we
have,
let's
see
we
have
a
new
member
here,
arthur
and
goda?
Could
you
introduce
yourself
been
making
some
good
commits
on
github.
C
Yes,
yes,
so
hi,
my
name
is
so
I
am
a
second
year
supporter
in
indian
technology
pharmacy.
So
I
like
the
project
so
thinking
of
contributing
connectivity
into
it
and
also
being
a
part
of
the
software.
C
A
Yeah,
that's
good
welcome,
so
we
tell.
B
A
Yeah,
that's
good
and
my.
D
A
Yeah,
my
knock
is,
is
doing
the
pull
requests
for
that
and
so
yeah
if
anyone's
any
questions
about
an
application
application
wise,
let
us
know-
and
we
can
help
you
through
the
process
it
largely
just
you
know
I
mean
there's
going
to
be
a
lot
of
scheduling
where
you
have
to
figure
out
the
schedule
for
things
and
how
they
are
going
to
be
implemented,
and
you
know
having
some
now
background.
Knowledge
of
some
of
the
things
in
the
group
are
good
too.
A
A
So
that's
good.
We
have
any
announcements
before
we
move
on.
A
So
yeah
congrats
on
the
the
progress
on
the
default,
we're
inside
it
looks
like
that's
going
to
be
probably
a
pretty
nice
thing
this
summer
to
get
into.
So
that's
I'm
going
to
talk
about
that
in
a
little
bit.
Okay,
jesse's,
mostly
listening
in
today,
welcome
that's
it.
For
now,
I'm
going
to
come
back
to
diva
learn
in
a
little
bit
in
this
flash
talk,
I'm
going
to
talk
about
submissions,
our
submissions
list
that
we
talked
about
a
couple
weeks
ago.
So
let
me
share
my
screen.
A
Okay,
and
we
have
a
number
of
things
on
our
list.
I
think
this
is
probably
incomplete
because
we
haven't
updated
it
in
a
while.
Actually,
as
it
looks
like
so
we
have
a
number
of
things
that
are
upcoming
possible
submissions
that
we
can
make
to
different
venues
to
different
conferences,
and
this
is
where
we
put
them
in
this
in
this
file.
A
This
is
just
a
way
to
keep
track
of
deadlines,
and
then
you
know,
if
there's
a
deadline
coming
up,
we'll
put
something
in
the
slack
channel
and
then
people
can
contribute
to
them
as
they
see
fit
if
they're
interested
in
contributing.
If
you're
interested
in
contributing
you
might
jump
in
the
slack
channel,
you
know
if
someone
puts
an
abstract
in
there
or
a
paper
link,
you
can
join
in
and
make
comments
and
there's
a
I
have
a
pretty
liberal
authorship
policy.
A
So
people,
you
know
you'll
get
authorship
if
you
make
a
decent
sized
contribution,
and
so
that's
that's
a
nice
thing.
So
this
evolution
2021
last
time
we
updated
it.
It
says
diva
warm
group
and
I
know
that
krishna
actually
presented
on
this
evolution
idea
he
wants
to
do.
I
can't
remember
the
name
of
the
the
actual
name
of
it
like
this
is.
A
I'll
just
put
krishna's
idea
because
I
have
in
the
other
one,
but
I
can't
remember
so.
The
krishna
has
an
idea
that
he
wants
to
turn
into
a
abstract
he's
working
on
some
slides,
but
we
have
to
turn
this
into
an
abstract
for
submission
sake
and
then
there's
another
one
that
we
talked
about,
and
this
was
okay.
I
want
this.
A
And
there's
a
thing:
I've
been
working
on
this
week.
It's
this
thing
on
euler.
I
think
it
was
euler
paths
for
life
or
oil
or
networks.
So
it's
like
we're
using
using
these
minimal
systems
where
it's
like
a
bunch
of
cells,
a
bunch
of
shapes
that
are
kind
of
contained
in
a
single
mass
and
the
spaces
between
the
cells
form
a
network
and
that
network
can
be
traversed
according
to
an
euler
path,
which
means
you
can
only
cross
one
edge
once,
but
you
have
to
cross
all
the
edges.
A
It's
basically
like
the
traveling
salesman
problem,
but
it's
applied
to
a
network
graph,
the
edges
of
a
network
graph,
so
it's
not
crossing
bridges
or
crossing
between
islands
and
bridges.
It's
actually
crossing
the
edges
of,
and
so
this
is.
I
gave
a
presentation
on
this
a
while
back
and
I'll
probably
give
a
presentation
on
it,
an
update
on
it
soon.
A
Once
I
get
the
abstract
done,
but
that's
going
to
be
another
submission
for
revolution
and
I'll
put
actually
what
I'll
do
is
I'll
put
some
information
in
the
slack
channel
and
I
will
update
everyone
on
it,
as
as
we
will
with
krishna's
idea,
maybe
a
little
bit
before
the
deadline,
and
so
the
deadline
for
that
is
march
1st.
So
that's
coming
up
pretty
soon.
It's
always
good
to
keep
yourself
a
couple
weeks
ahead
of
these
deadlines,
so
that
you
have
time
to
get
your
materials
in,
but
this
is
march
1st.
A
A
You
know
this
is
an
online.
It's
a
networks
conference.
So
it's
complex
networks.
I
I
think
a
lot
of
you
have
heard
of
what
those
are,
but
you
know
they
usually
have
a
lot
of
different
types
of
research.
There,
a
lot
of
graph
theory,
a
lot
of
different
domains
like
social
systems,
biological
systems,
physics,
you
know,
network
started
network,
complex
network
theory
started
in
physics,
but
they
have
all
sorts
of
people
there.
It's
very
interdisciplinary.
A
So
if
you're
interested
in
attending
that
that's
open
for
participation,
but
they're,
also
looking
for
submissions
and
they're
actually
looking
for
papers,
so
this
is
going
to
be
a
full
paper
instead
of
just
an
abstract.
So
let
me
make
a
note
of
that,
and
then
this
is
the
abstract
here.
A
All
right,
so
that's
that's
due.
I
can't
remember
the
deadline
on
that.
Let's
see
if
we
can
find
the
deadline,
so
that
would
be
march
26th.
So
that's
well
coming
up
in
a
fair
amount
of
time,
we'll
see
if
we
can
make
that.
I
think
it's
an
eight-page
paper.
A
So
it's
you
know
it's
not
really
super
long,
but
it's
probably
doable.
So
that's
that's
that
one
okay!
So
let's
go
back
up
here.
Divor
evil
learn
flash
talk,
osf
virtual
conference
for
online
education,
so
this
is
the
thing
that
I'm
going
to
be
presenting
at
tomorrow.
This
is
this.
Actually,
this
is
the
9th.
A
A
This
is
a
virtual
unconference
and
open
scholarship
practices
and
education.
Research.
Let
me
see
if
I
can
find
the
schedule
here
should
be
down
a
little
bit.
You
register
here.
This
is
the
schedule
I
think
see.
If
it
works
yeah
there,
it
is
okay,
so
this
is
the
schedule.
So
today
there
are
a
couple
talks
on
pre-prints
there's
a
talk
on
hackathons
at
1pm
eastern
time.
A
There's
see
what
else
they
call
an
unconference,
because
it's
just
kind
of
a
very
informal
conference.
They
have
different
sort
of
specialty
talks
and
different
things.
Then
they
have
lightning
talks,
which
is
what
I'll
be
doing
tomorrow
and
those
lightning
talks
are,
you
know,
pretty
diverse.
They
cover
a
bunch
of
different
topics
in
education
and
open
online
education.
So
you
know
everything
from
multiverse
analyses
in
the
classroom
to
open
art
histories
to
copyright
on
open
materials
and
then
tomorrow,
there's
more
workshops
there's
an
actual
hackathon.
A
So
this
is
like
a
working
session.
So
that's
that's
nice
to
have
at
a
conference
because
you
can
follow
up
on
it
later
then.
The
lightning
talks,
so
divalern
is
in
the
one
o'clock
spot
and
we
have
a
bunch
of
people
here.
So
the
lightning
talk
should
be
about
five
to
ten
minutes
actually
about
five
to
eight
minutes.
But
it's
you
know
it's
fast,
paced
and
so
we'll
see
when
I
open
the
talk,
what
it
looks
like
and
so
yeah
that's
I
mean
that
might
be
of
interest
to
some
people.
A
So
let's
go
back
to
the
submissions
here.
So
that's.
What
I'll
do
is
I'll
color
code
this,
since
it's
sort
of
on
its
way
out
and
presented,
so
that's
that's
completed
more
or
less
and
then,
of
course,
we
can
follow
up
on
this
later.
What
the
product
of
this
was-
and
we
can
maybe
recycle
it
for
other
things.
A
So
now
we
have
this
abstract.
I
think
we've
been
a
couple
of
us
have
been
working
on
it.
I
think
mayuk
and
ujwal
and
minak
and
krishna
and
myself
on
this
growth
form
and
theory
of
deep
learning-
and
this
is,
I
didn't,
actually
submit
this
to
the
scion
meeting.
So
I'm
not
really
sure
where
this
is
going
to
be
submitted.
A
I
didn't
think
it
was
a
very
good
venue
for
it,
but
but
this
is
a
it's
a.
I
think
it's
probably
at
this
point
like
an
extended
abstract
on
looking
at
deep
learning
networks
and
then
looking
at
how
maybe
we
could
use
them
for
developmental
biology
or
in
the
sense
of
like
simulating
developmental
biology,
not
necessarily
analysis.
So
this
would
be
a
little
bit
different
from
what
we
do
with
divalern
the
program,
and
you
know
it's
it's
a
speculative
piece,
but
it's
also
can
maybe
lead
to
some
technical
advances
as
well.
A
So
this
is
something
we're
still
kind
of
thinking
about
where
to
submit
it.
I
mean
that
you
got
to
find
the
right
venue
and
we
could
also
submit
it
as
a
paper,
but
I
think
that's
going
to
take
a
little
bit
more
work.
So
I
think
maybe,
if
good
first
step
is
like
a
extended
abstract
type
thing.
A
We'll
keep
working
on
that
we've
been
worked
out
in
a
while,
but
it's
available.
So
I
might
actually
put
this
on
the
list
of
open
papers
that
we
might
be
able
to
work
on,
and
that's
just
me
putting
a
link
to
the
draft
on
on
the
github
in
the
in
the
github
repo,
where
we
have
the
open
papers,
then
we
have
google
summer
code,
various
projects,
so
that's
completed,
that's
up
on
neurostars
so
and
we
have
to
update,
of
course,
what
our
deadlines
are
on
this.
A
So
now
the
deadline
isn't
no
longer
february
1st.
I
think
that
I
don't
know
when
the
actual
applications
open
but
the
applications-
I
don't
think
open
until
the
end
of
the
month,
and
then
I
think
the
end
of
march
is
when
they
actually
open
the
application
period,
and
then
that
goes
to
march
30th.
A
If
they
don't
get
sucked
for
the
projects,
they
usually
are
able
to
do
something
in
the
group,
and
it's
always
you
know
it's
always
a
positive
thing
to
be
involved
in
the
next
thing
is:
is
this
bacillary
non-neuronal
cognition
paper?
So
this
is
something
that
we
put
out.
A
D
Hi
bradley,
I'm
so
sorry
I
was
a
little
late
today
I
was
stuck
in
traffic
last
night.
Oh.
A
It's
fine:
are
you
prepared
to
give
a
talk
today,
yeah.
D
A
So
this
so,
if
you're
interested
in
being
involved
in
this,
it
can
get,
you
know
maybe
I'll,
put
some
more
information
into
the
slack
channel
soon
and
we
can
start
to
move
on
this
paper.
So
this
paper
we
gave
a
talk
at
neuro
match
conference
on
non-neuronal
cognition,
with
a
focus
on
basil
area.
So
we've
already
got
some
of
the
work
done.
We
just
need
to
get
this
together
into
a
paper
and
maybe
get
some
feedback
on.
You
know
what
we
should
cover.
We
shouldn't
cover.
You
know
some
there.
A
There
are
a
lot
of
things
to
do
so
I'll,
be
updating
people
on
that.
The
cielo
international
c
elegans
conference
is
coming
up
march.
25Th,
that's
the
deadline
for
the
abstract
submissions
and
then
the
conference
is
this
summer
and
it's
going
to
be
on
it's
going
to
be
virtual
online,
and
this
is
the
big
c
elegance
conference
that
everyone
goes
to
every
other
year.
So
every
other
year
there's
a
cl
wiggins
conference.
It's
it's
biologically
oriented
like
towards
people
doing
c,
elegans
biology.
So
you
know
the
computational
topics
are
a
bit
sparse.
A
Maybe,
but
it
doesn't
mean
that
they
won't.
You
know,
maybe
you
won't
get
accepted
into
a
session,
so
you
have
to
submit
an
abstract
there's
a
oh
there's,
a
fee
for
abstract
submission.
I
don't
know
why,
but
well
we
may
or
may
not
do
this.
I
don't
know
but
anyways
we'll
well,
if
you're
interested
in
in
doing
this,
we'll
talk
about
it.
I
haven't.
I
know
a
couple
years
ago
the
open
worm
group,
the
larger
open
worm
group
they
submitted,
or
actually
they
had
a
workshop
on
machine
learning.
A
So
I
think
it
was
in.
I
can't
remember
what
year
it
was.
I
think
it
was
2017
or
maybe
2019.
They
had
a
session
on
machine
learning.
They
had
a
session
on
machine
learning
at
eurips
as
well,
but
they
have
have
engaged
with
this
community.
So
this
is
something
that
it's
a
little
pricey
to
attend.
I
believe,
but
it's
a
place
to
make
some
connections
if
you're
interested
in
c
elegans
research,
the
the
diva
learned
paper
we're
still
working
on
that
a
that
deadline's
passed.
I
was
thinking.
A
Maybe
we
could
submit
that
to
a
conference
or
maybe
to
a
as
a
preprint,
maybe
as
a
journal
article,
but
I
don't
know
so.
This
was
the
one
that
was
the
when
we
were
going
to
submit
to
the
journal
of
open
source
software,
but
that
didn't
happen.
They
weren't
interested
in
it.
D
A
A
That's
the
thing
about
this
list,
too.
Is
we
can
put
in
different
things
that
we're
working
on
and
try
to
find
match
them
to
venues
and
or
put
venues
on
and
match
them
to
papers.
So
keep
that
in
mind.
Finally,
this
this
complement
thing
I
mentioned.
A
So
that's
that's
and
then
again
we
can
add
to
this
list
as
we
go
along.
So
please,
if
you
have
something
to
add
to
the
list,
please
put
it
in
the
slack
channel
and
I'll
try
to
add
it
in
when
we
can
so
we
have
some.
Let's
see,
we
have
some
things
in
the
chat.
A
So
oh,
yes
about
the
periodicity
paper,
so
I
think
jesse
and
and
ma
er
oswal
have
committed
to
the
periodicity
paper
revisions.
So
we
have
and
I'll
talk
about
this
more.
When
we
talk
about
the
papers,
we
have
some.
We
have
the
periodicity
paper,
it's
been
reviewed,
it's
pretty.
The
reviews
are
pretty
positive,
so
we
have
to
respond
to
the
reviews.
A
By
march
1st,
we
have
some
one
big
thing:
we're
looking
for
animations
of
embryogenesis
in
zebrafish
or
c
elegans,
and
so
we
have
maybe
have
some
candidates
for
that,
but
other
than
that,
there's
a
lot
of
editing.
That
needs
to
be
done,
and
things
like
that.
So
we'll
go
over
that's
maybe
next
week,
and
then
we
also
have
this
other
project
in
and
I'm
trying
to
organize
or
organize
this
project
on
neural
organoids,
and
so
I
haven't
mentioned
neural
organoids
here
yet,
but
it's
an
initiative
we're
trying
to
start
up
so
neural
organoids.
A
Are
these
they're
not
really
embryos?
They're?
You
know
they're
they're,
like
basically
neural
stem
cells
that
you
grow
in
in
big
groups
of
cells.
A
You
can
grow
them
in
a
3d
medium
or
you
can
grow
them
on
a
scaffold
and
they,
basically
you
grow
them
as
a
bunch
of
these
neural
stem
cells
and
they
sometimes
they
differentiate
into
different
layers.
Of
of
you
know,
different
layers
that
you
can
that
are
actually
are
functional,
but
they're.
You
know
you
have
to
use.
You
have
to
use
certain
media,
you
have
to
use
certain
substrates
to
grow
them
on,
but
you
can
get
some
systems
that
sort
of
approximate
a
developing
nervous
system,
and
so
it's
a
it's
a
fascinating
area.
A
But
the
thing
is
it's
very
open
because
it's
a
very
new
area,
so
there
are
a
bunch
of
groups
working
on
this
in
the
world,
maybe
about
20
or
30,
and
we're
trying
to
organize
some
literature
around
like
what
they've
been
doing.
We're
trying
to
find
a
niche
to
do
computational
analysis,
a
sample
of
an
organoid.
A
Because
I
mean
I
can
I
can
share
pictures
of
it.
I
don't
know
if
I
have
a
paper
right
now,
but
I
can
actually
get
some
pictures
of
it.
Oh
biological
sample.
Well,
I
think
the
they're,
you
know
they're
groups
that
it's
specialized,
they
have
a
specialized
lab
set
up
for
growing
organoids
and
I
think,
like
you
know,
if
you
just
I
don't
know,
if
you
inquire
to
a
lab,
they
may
or
may
not
be
cooperative
and
sending
you
things.
I
think
it's
more
that
you
have
to
grow
them
from
scratch.
A
Then
you
can
actually
get
samples.
I
don't
know
how
the
it
actually
works.
I
don't
know
how
the
like,
I
know
the
protocols
are
pretty
specific
and
yeah,
so
I
can
actually
share
some
pictures.
Maybe
next
week
I'll,
maybe
I'll
put
some
together
because
we
have
a
you
know
we
can
just
take
them
out
of
some
of
the
papers
or
I'm
sure
they'll
well
I'll
get
into
it
yeah.
A
A
Maybe
I'll
talk
about
it
next
week
and
we'll
do
a
little
bit
more
on
that
and
then
finally,
can
we
put
the
link
to
the
submission
submission
sheet
in
the
slack
yeah
I'll
put
it
in
the
sla?
Actually,
I
can
put
this
list
in
the
chat
and
then
you
can
ask
for
permissions.
A
So
that's
the
google
doc
for
that.
So
why
don't
we
move
to
surety
and
why
doesn't
she
give
her
a
talk
and
then
I'll
return
to
the
flash
talk
after
that.
A
D
Usa
transfer
can
be
used
to
generate
an
output
image
that
is
based
on
the
stuff
that
is
shown
in
what
we
call
the
content
image,
but
the
styling
and
the
feel
of
the
image
is
more
from
the
silence
it
can
be
called
like
a
filter,
but
how
is
it
different
from
a
filter?
Let
me
tell
you
that
first.
A
What
we
do
is
basically,
we
either
manipulate
the.
A
D
All
these
various
effects
that
you
need
that
you
can
see
with
no
filters
and
all
this,
but
in
your
generation
technique
which
I
have
used,
it
basically
exploits
the
way
in
which
layers
add
different
lengths
of
the
cn
and.
D
A
In
the
last
week's
session,
which
we
talked
about,
manik
was
talking
about
resnet
50,.
D
D
Of
players
there
are
more
features.
The
model
can
learn
it
can,
you
know,
try
to
extract
all
the
small
details
in
the
image
which
we
have,
so
that
is
what
it
is
doing
trying
to
this
this
model,
particularly
what
is
doing
it.
D
D
Put
together,
they
gave
us
an
output
image.
So
now
all
this
I've
told
you.
So
how
could
we
know
that
if
the
generated
image
has
the
content
of
the
image,
for
example
your
face
and
the
style
of
the
silence
in
an
example
of
another
painting,
for
all
this,
we
have
lost
function.
But
since
we
have
two
images
to
be.
D
D
A
This
is
how
it
would
be
done
like
this
first
particular.
This
is
different.
D
From
the
filters
which
we
have
used
till
now,
that
is
because,
in
that
what
we
do
was
just
fix
the
manipulation
we
would
increase
the
brightness
or
decrease
the
brightness
normalize
it
or
add
some
filter
on
top
of
it,
but
we
never
transferred
this
side
and
I
thought
prisma
came.
I
guess
two
four
to
four
years
before
which
did
use
the
exact.
D
D
A
Well,
thank
you.
Let
me
see
what
we
have
in
terms
of
questions.
Oh
susan.
She
liked
it.
She
was
put
a
laughing
and
then
my
hook
and
my
knock
both
clapped
so
there
you
go.
Thank
you
so
much
so.
Could
you
share
your
screen
again?
I
wanted
to
go.
Look
at
the
slides
a
little
bit
more.
A
Yeah
so
go
back
to
the
maybe
some
of
the
like
susan's
image,
and
then
we
can
go
forward.
A
D
It's
not
the
mask.
The
masks
are
what
we
initially
find
in
the
filters.
A
D
Was
to
increase
the
amount
of
you
know,
content
that
had
to
be
there.
We
have
a
content.
A
D
D
D
A
A
So
if
you,
if
you
train
how
long?
If
you
go
out
like
about
well,
I
mean
first
of
all
at
some
point:
it
converges.
If
you
over
train
the
model,
say.
C
A
It's
very
good,
very
good,
and
so
what
did
you
use
for
this?
What
what
I.
D
Used
the
vg
and
s60
network
for
this
because
it
had
early
16
years
and
what
we
had
to
do
was
take
the
initial
convolution
there
for
the
side
image
and
the.
D
D
This
training
around
two
to
three
hours
and
susan's
worth
like
one
and
a
half
hours
so
yeah.
He
took
a
little
time
and
you
know
changing.
A
Yeah
yeah
that
looks
good.
Thank
you.
So,
yes,
thank
you
very
much
rudy.
If
people
are
interested
in
giving
a
talk,
oh
my
knock,
would
you
like
to
say.
B
A
Excellent
yeah,
if
you
wanted
to
push
it
to
the
we
have
a
data
science,
demos,
repo
on
evil,
learn,
and
if
you
want
to
push
it
there,
it
would
be
good.
We
have
like
a
bunch
of
demos
there
for
all
sorts
of
different
topics,
so
that
would
actually
be
a
good
place
to
share
it.
A
Very
good,
and
so
if
anyone
wants
to
give
a
talk
on
something
it
doesn't
have
to
be
machine
learning,
it
can
be
some
other
area
that
we
do
in
the
group
here.
Please
let
me
know
you
know
that,
so
this
is
a
good
example.
It's
very
good!
Why
not?
Would
you
like
to
say
something.
B
Yeah,
so
actually
you
are
speaking
of
toxic.
So,
like
I
was
working
on
the
new
update,
I
was
actually
working
on
a
new
update
for
divorce,
so
I
think
I.
B
A
A
So
let
me
go
into
the
flash
talk
now
and
I
I'm
going
to
give
a
it's
going
to
be
a
pretty
high
level
talk
on,
so
I've
been
taught,
we've
been
talking
about,
diva
learn
as
the
platf
as
like
a
standalone
program,
but
we
also
it's
also
a
platform
and
I'll
show
you
what
that
looks
like.
So
this
is
the
if
you
can
see
my
screen
okay,
so
this
is
the
talk.
A
A
But
this
is
what
you
know
this
is
as
of
like
last
night.
We
have
a
lot
of
people
and
we're
adding
people
all
the
time
to
the
to
the
repo.
These
are
all
the
people
who
have
filed
a
pull
request
with
the
organization,
so
we
did
some.
We
have
some
core
contributors
here
in
the
group
who
attend
the
meetings
we
had.
A
People
participated
in
hectoberfest
and
all
those
people
are
listed
and
they're
people
coming
in
all
the
time
in
different
ways,
so
diva
worm
group
as
a
as
a
whole
is
devoted
to
building
the
world's
first
digital
c
elegans
embryo,
although
you
know
we're
we're
focused
on
other
organisms.
Other
model
systems
as
well
open
worm
foundation,
of
course,
is
what
we're
affiliated
with
for
the
aspect
of
building
a
sort
of
a
digital
model
of
a
c
elegans
embryo
or
a
generalized
embryo.
A
We,
you
know,
look
at
the
nematode
c
elegans
and
other
model
organisms
using
simulation
analysis
and
visualization,
and
we
deal
a
lot
with
a
lot
of
secondary
and
tertiary
data
sets,
which
are
data,
sets
that
have
been
processed
and
made
available
publicly.
So
there
is
all
sorts
of
data
we
focus
on
segmented,
microscopy
data,
public
repositories
and
literature
mining.
Among
other
things,
and
so
one
of
the
things
we're
interested
in
is
deep
learning
and
quantitative
morphology
and
so
on
the
right.
A
You
see
this
example
of
c
elegans,
embryogenesis
being
segmented
into
individual
cells
and
then
into
these
cell
centroids,
which
represent
like
not
the
nucleus
of
the
cell,
but
the
center
of
one
of
these
segmented
cells,
and
so
we
can
track
those
things
in
space
and
in
time
as
they
divide
and
are
born.
We
also
can
use
theory
building
to
explain
developmental
processes
so
explaining
how
we
go
from
a
sphere
to
this
asymmetrical
cow,
and
you
can
see
in
this
gif
that's
sort
of
a.
A
I
mean,
that's
sort
of
an
idealized
process,
but
that's
what
we're
interested
in
and
then
we're
also
looking
at
things
like
vascularis
cell
colony
morphology.
So
this
is
a
little
bit
beyond
conventional
development,
but
we're
looking
at
applying
these
techniques
to
other
systems
and
understanding
things
like
movement
and
other
other
phenomena.
A
What
was
that?
Oh
okay,
so
we
started
with
diva
or
ml,
and
this
was
something
that
happened
in
2019
in
the
fall,
and
it
was
a
course
on
machine
learning
that
had
a
specific
bent
towards
developmental
biology.
So
we
brought
together
machine
learning
and
developmental
biology.
You
know
this
was
totally
online.
This
was
amongst
our
group,
but
we
invited
people
in
to
give
talks
and
other
blog
posts,
and
things
like
that.
A
Under
this
data
science
theme
under
this
machine
learning
theme,
then
this
of
course
turned
into
our
one
of
our
projects
for
google
summer
of
code
2020,
where
we
worked
on
something
called
the
devo
zoo,
but
also
on
these
pre-trained
models.
So
we
have
the
we
upgraded,
our
our
secondary
data
portal
called
devozu,
but
we
also
started
to
build
on
these
pre-trained
models,
and
so
the
pre-trained
models
were
done
by
my
oak
deb
and
he
produced
this
divalent
0.2.0
release,
which
is
a
pre-trained
model
for
c
elegans.
A
Embryos
well
also
produced
this
upgrade
to
the
devozu
repository,
which
is
a
collection
of
data
sets
which
one
might
use
then
to
plug
into
something
like
divalern
and
so
the
way
we
do.
This
google
summer
of
code
is
a
a
core
thing
in
our
educational
pipeline.
A
So
we
have
this
one
of
the
things
we
do
is
we
prepare
students
for
the
google
summer
code
application
period.
There's
this
education
and
evaluation
pipeline
that
we
have.
We
find
a
call
for
involvement.
A
We
contribute
to
a
github
issue
or
we
have
them
contribute
to
a
github
issue
during
the
application
period,
and
then
they
can
join
weekly
meetings
and
get
involved
in
bigger
projects.
So
you
can
see
it's
sort
of
a
a
way
to
get.
You
know.
The
call
for
involvement
is
a
way
to
get
people
into
this
educational
and
research
infrastructure
and
on
the
right
you
can
see
the
divo
learn
github
organization,
where
we
have
the
diva
learn
software,
but
we
have
other
things
like
data
science,
demos
and
the
general
biological
model.
A
So
there
are
all
sorts
of
activities
under
the
sunbrella
that
we're
working
on
alongside
evil
and
eviler
and,
of
course,
being
the
software
being
the
most
developed
of
those.
So
right
now
we're
at
evil
in
0.2.3,
and
that
is
the
latest
release
of
divalern
diva
learn
the
program
is
a
standalone
program.
A
A
We're
trying
to
make
more
advances
in
the
gui
and
the
way
people
interact
with
it
and
better
benchmarks,
and
we
have
a
project
in
google
summer
code
this
summer
to
build
upon
our
success
with
this
platform,
and
this
is
of
course,
a
sort
of
a
schematic
of
the
github
source
and
the
user
environment
in
this
program,
as
it
stands
right
now,
so
we
have
a
lot
of
functionality
built
in
already,
but
we're
looking
to
expand
it
out.
A
But
then
we
also
have
this
platform,
so
we
not
only
have-
and
I
need
to
change
this-
we
not
only
have
the
standalone
pro
platform
released
as
open,
open
source
software,
but
we
have
this
umbrella
of
things.
So
we
have
two
other
things
here:
I'd
like
to
highlight
species
specific
models
and
that's
what
we
call
divo
worm
ai,
which
they're
programs
that
allow
us
to
analyze
different
different
other
model
organisms
and
in
different.
A
A
So
this
is
divorm
ai.
This
is
the
splash
page
for
it
we
have
here
a
number
of
specialized
programs
for
different
model
organisms.
We
have
links
to
the
devo
zoo,
which
is
a
place
where
we
store
data
sets
for
students
or
other
collaborators
to
engage
with
to
start
to
analyze
and
start
to
understand.
What's
going
on,
we
also
look
at
a
number
of
different
ways.
A
We
have
access
to
a
number
of
different
development
in
a
number
of
different
species,
so
we
have
c
elegans,
of
course,
which
is
the
place
we
started
with
all
of
this,
and
then
we
have
also
acquired
data
from
drosophila
from
zebrafish
and
from
spider
embryos,
and
there
are
many
more
other
types
of
model
organisms
we'd
like
to
incorporate
their
plant
models
where
you
look
at
plant
embryos,
their
ant
embryos-
and
there
are
other
types
of
embryos
that
are
specialized-
that
maybe
aren't
model
organisms,
but
we'd
like
to
understand
anyways,
and
so
this
is
what
this
platform
allows
you
to
do
is
to
you
know,
find
a
model
organism,
maybe
analyze,
something
that
isn't
like.
A
We
don't
have
a
lot
of
knowledge
about
it.
We
might
use
one
set
of
models,
but
we
might
use
a
pre-trained
model
for
something
like
c
elegans,
where
there
is
a
lot
of
outstanding
data,
and
we
can
get
a
sense
of
what
you
know.
We
can
do
a
lot
of
things,
it's
a
very
diverse
platform,
that's
what
we're
aiming
for,
and
so
we
have
a
lot
of
things
as
part
of
the
stevia
worm
or
divalern
umbrella.
A
We
have
a
lot
of
things
that
are
being
developed
by
contributors,
as
well
as
the
main
group
contributors,
meaning
like
people
from
just
from
interactions
through
github
or
through
the
open
worm.
Slack.
A
Who
join
the
group
on
a
regular
basis,
so
we
have
data
science,
demos.
We
just
saw
a
demonstration
of
that
right.
Just
previous
to
me
talking.
We
do
this
in
jupiter,
notebooks
or
collab
notebooks.
A
We
have
methodological
tutorials
and
the
likes
so
we're
trying
to
engage
people
in
terms
of
data
science,
education,
we're
trying
to
engage
people
in
more
general
education
with
respect
to
teaching
computational
students
about
developmental
biology
and
vice
versa.
And
then
we
have
this
theory
building
aspect
which
is
building
explanatory
frameworks
and
hypotheses
for
the
data
that
were
generated
so
we're
generating
data
or
we're
generating
analysis
of
data
where
sometimes
we're
even
generating
data,
and
then
we're
analyzing
it.
And
then
we
have
to
understand
that
analysis.
A
A
Krishna
is
busy
with
the
paper
easter
egg,
say:
hello,
hello,
krishna,
hello,
hi,
so
so
I
guess
the
last
and
again,
if
you
have
to
leave
at
the
top
of
the
hour,
that's
fine,
but
I
wanted
to
go
over
some
papers
to
for
the
week.
A
Let's
see
okay,
so
the
first
thing
I'm
going
to
talk
about
here
is:
I
did
find
a
zebrafish
embryo
semi
like
it's
a
it's,
a
video
from
the
fabian
lab
at
usc,
and
this
is
one
of
the
zebrafish
embryo
where
cells
are
migrating.
It's
a
time
lapse.
Video!
A
A
Paper,
so
let's
I
just
want
to
give
you
a
taste
of
that.
The
other
thing
I
wanted
to
go.
Oh
yeah.
I
wanted
to
also
say
that
there's
I
put
up
a
an
announcement
for
the
summer
code
projects
on
twitter.
Here
we
have
incf,
openworm
and
orthogonal
lab
and
diva
worm
are
all
involved
in
this
initiative.
Getting
people
involved
in
these
different
projects,
so
you
can
see
they're
linked
to
the
neurostars
descriptions,
and
hopefully
people
will
respond
to
this,
maybe
just
trying
to
advertise
it
as
many
places
as
possible.
A
Oh
okay,
I
do
have
a
paper
on
the
organoids,
so
I'm
going
to
talk
a
little
bit
about
the
organo
herb
this.
This
is
one
one
paper
on
organoids.
This
is
a
review
of
the
embryoids
organoids
and
gastroids
new
approaches
to
understanding
embryogenesis.
A
So
this
is
a
review
of
organoids
and
I'm
going
to
present
more
in
more
detail
on
this
and
maybe
next
week.
But
I
wanted
to
show
this
just
to
give
you
a
taste
of
what
it
is.
So
organoids
are
here's
a
this.
A
panel
is
an
organoid
and
then
they
compare
it
to
gastroids
and
embryoids.
So
an
organoid
is
where
you
have
mescs,
which
are
stem
cells
and
they're
dissociated
into
a
3000
cell
cluster,
and
then
you
put
it
into
this
medium
with
minimal
growth
factors,
and
this
is
a
big
thing
with
growth.
A
Factors
are
very
important
in
cultivating
organoids,
but
I
don't
think
anyone
has
the
right
formula.
They
just
try
to
try
different
things,
but
you'll
see
it's
a
big.
It's
a
big
controversy
in
the
field,
so
they
create
these
3
000
cell
clusters.
They
put
them
in
the
media,
they
put
them
in
like
either
a
bioreactor.
A
They
put
them
on
a
scaffold
which
is
usually
some
surface
where
they
might
grow
it
into
something
like
an
ear
or
a
heart
valve,
or
something
like
that,
and
then
they
let
them
grow.
And
you
end
up
by
day
five.
You
start
to
get
this
sort
of
apical
and
basal
sections
of
the
organoid,
and
then
you
get
this
something
like,
in
this
case,
they're
growing,
an
optic
cup.
So
you
start
to
get
this
optic
cup
emerge
and
then,
by
day
nine
you
see
this
thing
in
the
sustain.
This
is
stained
gfp,
so
you
can
see.
A
D
A
These
two-
it
has
two
layers
in
it,
so
it's
kind
of
hard
to
see
from
this
image,
but
that's
basically
the
idea.
You're
growing
these
these
masses
of
cells
and
you're
trying
to
differentiate
them
into
different
things
and
it's.
I
know
it
seems
a
little
bit
magical
but
there's
a
lot
of
hardcore
like
wet
lab
work
that
goes
into
this
and
again
I'll,
give
more
information
about
that
next
week.
A
So
there
are
a
lot
of
things
that
you
can
see
like.
You
can
examine
a
lot
of
self-worth,
self-organization
and
patterning
in
these
organoids,
as
in
in
the
neural
organoids,
you
get
like.
You
know,
different
layers
of
neurons
and
you
can
differentiate
them
in
that
way.
But
you
can
also
observe
a
lot
of
pattern
formation
and
self-organization.
A
A
You
can
look
at
things
like
lumen
formation,
look
at
the
mechanical
influences
and
also
look
at
geometric
confinement,
which
is
something
that
we
can
do
in
in
cell
culture.
But
you
know,
I
think
this
allows
you
to
culture,
a
lot
more
cells
at
one
time,
so
it
allows
you
to
build
these
tissues
basically
and
then,
and
again,
it's
very
early
on
in
the
science.
So
it's
hard
to
kind
of
know.
You
know
what
the
it's
kind
of
hard
to
show
really
good
results.
A
A
lot
of
the
papers
out.
There
are
basically
on
like
how
to
grow
them,
and
maybe
some
cool
things
we
can
do
with
them.
So
I'll
give
a
talk
on
that
more
next
week.
I
think
susan
sent
me
this
paper
c.
Elegans
is
a
model
organism,
and
this
is
this
is
an
article
on
what
you
know:
sort
of
the
relevance
of
c
elegans
as
a
model
organism.
A
So
it's
not
a
mammalian
model,
but
we
can
use
it
as
a
way
to
look
at
sort
of
physiological
properties
of
an
animal.
We
can
replicate
human
diseases
and
because
it
has
a
fast
life
cycle,
we
can
do
things
like
experimental
evolution
or
we
can
look
at
the
life
history
of
the
organism
without
spending.
You
know
a
lot
of
time
on
a
single
subject,
and
so
you
know
you
can
study
things
like
human
diseases,
ranging
from
parkinson's
disease
to
mitochondrial
diseases,
immune
system
diseases,
and
so
this
is
actually
a
lot
of
stuff.
A
They
talk
about
at
the
c
elegans
meeting,
and
this
is
a
very
basic
description
of
c
elegans
as
a
model
organism.
So
maybe,
if
you
know
a
lot
of
people
are
coming
in
as
computational
people,
they
might
want
to
read
this
article
as
a
way
to
sort
of
get
up
to
speed
on
some
of
the
potential
for
c
elegans.
A
It
has
some
nice
references
at
the
end,
and
so
this
is
a
nice
article.
So
that's
that's
good.
What
else
I
want
to
talk
about,
then
there's
this
paper
continuing
down
the
road
of
c
elegans
and
development.
We
have
this
paper
animal
systems
biology
towards
the
system's
view
of
development
in
c
elegans.
A
This
was
published
in
2009
and
it's
a
nice
paper
on
sort
of
thinking
about
c
elegans
as
a
systems
model
again
like
it's
a
it's
a
animal
model
where
we
can
study
a
lot
of
bio
medicine,
but
we
also
it's
also
a
system
and
and
there's
a
systems
level
biology
that
can
be
done
here.
A
So
the
nematode
worm
c
elegans
is
the
preeminent
model
for
understanding
animal
development
at
a
systems
level.
Embryonic
development,
in
particular
has
been
studied
intensively
in
c
elegans
and
genes,
essential
for
early
stages
of
embryogenesis
and
their
specific
phenotypes,
and
in
catalog
comprehensively.
A
Here
we
review
the
systems
level
approaches
that
have
been
used
to
study
the
early
development
in
c
elegans
and
how
these
are
deepening
our
understanding
of
complex
molecular
programs
underlying
development,
so
they
they
kind
of
go
through.
Why
I
see
elegance
as
a
model
for
development,
and
part
of
that
is
because
it's
a
very
simple
organism
in
the
sense
that
it
has
cells,
that
we
know
what
they're
going
to
become
as
they
divide,
we
know
kind
of
what
their
fate
is
going
to
be.
We
can
draw
a
fate
tree.
A
You
know
we
can
draw
a
tree
of
different
sulfates,
they
call
it
a
lineage
tree,
but
the
lineage
tree
is
deterministic
as
they
call
it.
It
also
has
adults,
have
the
same
number
of
cells,
the
well
depending
whether
it's
a
male
or
hermaphrodite,
there's
a
different
number
there,
but
any
one
c
elegans
is
going
to
have
a
certain
number
of
cells
as
an
adult.
Unless
it's
a
mutant
and
in
which
case
it
will
maybe
have
a
few
more
or
few
less
cells,
but
it's
very
pretty
much
consistent.
A
So
we
can
track
all
of
that,
and
this
is
why
we
use
it
in
in
open
worm
as
well,
because
you
know
it's
something:
pretty
predictable
as
a
system,
and
so
from
this
we
can
do
things
like
single
cell
sequencing.
We
can
do
other
types
of
studies
of
development.
We
can
do
knockouts
of
development,
so
we
can
knock
out
genes
and
we
can
we
know
kind
of
what
the
effect
is
much
more
clearly
than
maybe
in
something
like
a
mouse,
and
so
all
these
things
are
very
powerful.
A
D
A
Is
2009
so
this
is
well
before
we
started
working
on
this
or
anyone
else
really
thought
about
this
stuff.
As
a
phenome,
I
mean
people
were
working
on
c
elegans
since
the
60s,
but
this
view
is
not
that
old.
So
there's
there
have
been
a
number
of
attempts
to
try
to
build.
A
You
know
these
type
of
holistic
models
of
c
elegans
and
you
know
they've
had
very
been
a
varying
success,
so
the
good
kind
of
walks
through
development
of
c
elegans
here
cellular
phenomena
during
early
embryogenesis,
so
it
really
really
gets
deep
in
the
weeds
on
this
stuff.
So,
if
you're
interested
in
learning
about
like
very
specific
events
in
development,
this
is
a
good
paper
to
read,
so
they
actually
have
some
good
like
microscopy
here
of
the
embryo.
A
So
this
is
the
embryo
at
the
two
cell
stage,
which
is
where
you
get
the
basic
anatomical
pulls
between
the
anterior
and
posterior
ends,
and
these
are
the
two
major
cell,
lineages
p1
and
ab.
A
So
when
you,
if
you've
been
working
on
the
machine,
learning
and
you've
still
seen
those
you
know,
letters
representing
you
know
letters
of
representing
different
cell
subliminages.
That's
what
these
look
like.
These
are
two
sublineages
a
b
and
p1,
and
then
you
get
to
the
this
is
the
four
cell.
So
this
is
like
two
a
b
cells-
and
this
is
like
two
p
one
cells
and
those
forms
sub
lineages,
and
then
the
eighth
cell,
I
think,
is
where
you
have
your
basic
eight
founder
cells.
A
Those
are
like
major
events
of
differentiation
and
in
the
ab
line.
Sublineages,
you
have
a
lot
of
muscle
cells.
A
lot
of
you
know
other
types
of
cells,
and
then
you
have
in
the
p1
lineages
you
have
the
germ
line.
You
have
very
specific
muscle
cells.
You
have
other
types
of
very
specialized
cell
types
in
the
gut
and
things
like
that,
and
so
a
lot
of
these
things
are
defined
right
from
that
point
in
development,
and
so
you
get
a
lot
more
cell
proliferation.
A
You
end
up
with
this
comma
period,
which
is
where
the
you
start
to
get
a
deformation
in
this
basic
spherical
shape,
and
then
you
finally
get
to
something
that
looks
like
this
just
before
the
egg
hatches,
which
is
where
the
at
the
larval
worm
is
all
curled
up
and
ready
to
unfurl,
and
so
that's
what
the
developmental
process
looks
like
and
then
they
get
into
genes
which
are,
if
you're,
not
a
geneticist.
It's
a
little
bit
daunting
to
understand
this.
A
You
know
the
different
symbols
that
they
use
for
genes,
but
I
think
they
they
define
some
of
this
in
here
and
they
go
through
this.
They
go
through
reverse
genetic
approaches
and
they
go
through
forward
genetic
approaches.
I
don't
know
if
they
actually
go
through
that
too
much
depth,
but
they're
forward
and
reverse
genetic
approaches
which
allow
you
to
either
look
at
genes
by
sort
of
examining
their
mutations
or
knocking
genes,
they're,
knocking
them
out
functionally
and
then
looking
at
their
effects,
and
so
there
are
ways
to
do
this.
A
There
are
techniques
and
they
kind
of
go
through
these
techniques.
They
use
something
called
rnai,
which
is
a
way
to
knock
down
gene
expressions,
so
that's
also
used
and
they
go
through
that
which
is
a
little
bit.
It's
a
older
techno.
I
don't
know
how
much
they
use
it
anymore,
but
it
was
a
big
deal
about
15
years
ago.
A
They're.
All
these
tools
that
are
constantly
circulating
in
molecular
biology
so
getting
into
all
those
tools
is
harrowing.
Actually,
but
you
know,
if
you
read
this,
I
think
you
get
a
sense
of
like
some
of
the
things
that
people
are
doing
to
to
control
gene
expression,
to
study.
You
know
mutants
and
and
study
phenotypes,
and
things
like
that.
A
So
a
bird's
eye
view
of
the
embryonic
phenome
global
trends,
and
so
they
kind
of
go
through
the
trends
here
for
looking
at
some
of
these
reverse
genetic
studies
and
looking
at
one
thing
that
c
elegans
is
really
good
for
is
looking
at
mutants
specific
mutants
to
specific
genes.
So
you
can
knock
out
a
specific
gene
and
you
can
see
a
specific
phenotypic
outcome
like
you
know
there
might
be
a
metabolic
gene,
you
knock
out
and
it
has
an
effect
on
growth.
A
A
lot
of
those
things
exist
and
they
have
them
all
cataloged
and
you
can
actually
get
mutants
of
that
specific
knocked
out
genotype
and
you
can
grow
them
up
and
you
can
observe
these
effects,
and
so
it's
a
really
nice
system.
For
that.
That's
another
reason
why
it's
very
good
for
looking
at
disease,
because
you
can
knock
out
different
genes,
so
you
know
that
might
be
like
homologs
of
things
in
humans
and
get
an
understanding
how
they
function
so
and
then
they
cluster
phenotypes.
A
They
have
a
lot
of
stuff
in
this
paper
data
integration
towards
a
systems
view
of
early
embryogenesis,
so
they're
putting
together
data
here.
This
is
a
very
nice
paper.
Definitely
if
I
were
if
you're
interested
in
that
in
just
kind
of
understanding
some
of
that
depth
of
c
elegans,
what's
going
on
in
the
c
elegans
community,
I
would
read
that
paper.
A
Finally,
I
think
I'll
just
bring
your
attention
to
this
paper.
Actually,
this
is
this,
isn't
the
paper
itself,
but
this
is
a
press
release
on
this.
So
there's
this-
this
is
actually
quite
recent
december
23
2020,
novel
computational
tool
can
systematically
analyze
cell
images
and
c
elegans,
and
so
there's
a
group
out
of
city
university
of
hong
kong,
who's
developed
a
novel
computational
tool
like
reconstruct
and
visualize
three-dimensional
shapes
and
temporal
changes
in
cells,
and
so
this
is
what
are
they
using
here.
A
So
this
is
a
paper
in
nature.
Communications
called
establishment
of
a
morphological
analysis
of
the
c
elegans
embryo,
using
deep
learning
based
4d
segmentation,
so
they're
using
this
tool
called
c
shaper
and
it's
a
powerful
computational
tool
can
segment
and
analyze
cell
images
systematically
at
the
single
cell
level,
and
so
this
is
much
needed
for
the
study
of
cell
division
and
cell
and
gene
functions,
and
they
might
mention
a
bottleneck
and
analyzing
a
lot
of
the
data
that
related
to
cell
division.
A
So
there's
a
lot
of
data
out
there,
but
it's
hard
to
know
you
know
it.
You
know
it's
hard
to
analyze
it
all
in
order
to
analyze
it
deeply,
because
you
know
you
can
do
things
like
put
a
marker
on
a
cell
and
look
at
how
it
splits
apart
and
you
know,
but
but
there's
so
much
detail
in
cell
division
from
what's
going
on
inside
the
cell,
to
sort
of
its
position
in
the
environment,
to
the
walls
of
the
cell,
how
they
change.
So
there's
a
lot
of
information
there.
A
What
they're
doing
here
is
they're,
actually,
okay,
what
I
think
they're
doing
here
is
they're
actually
looking
for
they're
doing
some
specific
experiments.
They're
they're,
knocking
out
a
dna
sequence,
which
is
what
we
mentioned
in
the
last
paper.
They
sort
of
describe
how
that
works.
Then
they
compare
two
lineage
trees
that
are
generated
from
like
a
wild
type,
which
is
the
one
without
any
sort
of
changes
to
the
dna
sequences.
A
And
so
this
will
allow
you
to
infer
maybe
gene
functions
or
changes
in
how
the
cells
are
dividing
and
so
they're
able
to
generate
a
bunch
of
data,
and
this
team
is
analyzing,
it
so
they're,
analyzing,
yeah,
so
they're
using
segment
as
they're,
using
this
technology
to
segment
the
soul,
images
that
are
generated
from
these
experiments,
they're,
actually
using
time-lapse
3d
images,
which
is
what
they
refer
to
as
4d
images.
A
So
this
is
again,
you
know
this
is
like
getting
it
for
movies,
but
it's
not
they're
just
kind
of
looking
at
time
and
differences
in
time.
So
the
you
know,
the
way
in
which
you
deal
with
this
flow
of
time
and
development
is
is
kind
of
important,
and
I
don't
think
people
have
really
thought
about
that
too
much.
I
think
it's
something
people
are
just
kind
of
trying
different
things
on.
A
So
it's
hard
to
know
like
what
the
the
you
know.
This
is
something
we
talk
about
further,
but
this
fourth
dimension,
this
time
dimension
of
of
image.
Segmentation
images
is
actually
maybe
maybe
an
essential
part
of
this
image.
A
Segmentation
is
a
critical
process
in
computer
vision
that
involves
involves
dividing
a
visual
input
into
segments,
but
researchers
have
spent
hundreds
of
hours
labeling
many
cell
images
manually,
so
the
state
of
the
art
maybe
30
years
ago,
was
label
cells
manually
and
then
detract
them
using
like
very
simple,
rudimentary
algorithms,
and
so
now
we're
able
to
do
this
and
with
advanced
techniques
in
a
few
hours
relatively
speaking,
and
so
this,
the
deep
learning
based
model
dmap
net
developed
by
the
team,
plays
a
key
role
in
the
c
shape
or
system
they're
able
to
capture
multiple
discrete
distances
using
image.
A
So
c
shapers
achieve
95.95
percent
accuracy
on
identifying
cells,
which
we
know
in
advance
what
the
cell
should
be
in
in
the
embryo
and
c
elegans,
and
this
outperformed
other
methods
substantially
and
so
they're
able
to
generate
a
time
lapse.
3D
analysis
of
cell
morphology
from
the
4
to
350
cell
stages,
including
cell
shape
volume,
surface
area,
migration,
nucleus
position
and
cell
cell
contact.
A
So
that's
pretty
good.
It's
get
gets
a
lot
of
information
out
of
the
cells.
I
don't
know
best
of
our
knowledge
c.
Shaper
is
the
first
computational
system
for
segmenting
and
analyzing
the
images
of
c
elegans
embryo
systematically
at
the
single
cell
level.
I
think
there
are
people
working
on
this,
though,
like
other
people,
so
we
won't,
through
close
collaboration
with
biologists,
we've
been
able
to
develop
a
more
useful
tool,
and
so
this
is
just
kind
of
promoting
this
platform.
A
They've
also
tested
c
shaper
and
plant
cell
tissue
plant
tissue
cells
showing
promising
results.
So
this
is
I'm
not
going
to
get
into
the
paper,
but
that's
basically
the
idea
of
the
paper
and,
if
you're
interested
these
in
this,
they
cite
that
paper
in
this
article.
So
I
mean-
let
me
put
this
in
the
chat
and
I'll
put
this
on
these
papers
in
the
in
the
slack
channel
as
well.
D
A
A
Thank
you,
shorty
for
presents
presenting
on
the
on
the
r
on
your
art
project
and
please
push
that
to
our
diva
learn
repository,
and
hopefully
we
have
some
more
participation
on
the
divalern
platform
in
general
and
if
you
have
any
questions,
let
me
know
on
slack
I'll
post
some
of
the
readings
on
slack
for
people
so
otherwise
have
a
good
week
and
talk
to
you
see
you
on
slack
next
week
in
the
meeting.