►
Description
GSoC Community Period (week 3), getting ready for projects. Updates on D-GNNs and Digital Microspheres. Discussion on how to capture your insights and roadmap projections during GSoC (Moon shots, Mars shots, Intergalactic shots). Review of tools and education for biological and cell modeling using cellular automata and other techniques. Attendees: Susan
Crawford-Young, Jesse Parent, Harikrishna Pillai, Karan Lohaan, Richard Gordon, Jiahang Li, and Bradly Alicea
A
Welcome
to
the
meeting
see
ya
hung
is
here
and
dick
and
susan
hi.
A
C
C
Yeah
yeah,
I'm
fine.
Yes,
I'm
sorry
that
I
didn't
attend
the
meeting
the
last
week
because
it
was
conflict
with
some
of
my
family
stuff.
Yes,
but
in
actually
about
the
other
projects
we
are,
I
mean
waterloo
and
I
were
still
in
the
process
to
do
some
general
plans
of
this
project.
C
For
example,
we
have
some
discussions
before
about
what
kind
of
tours
or
what
kind
of
frameworks
we
will
use
in
this
project.
Yes,
and-
and
I
I
think
today-
I've
I
can
bring
some
updates
to
you
about
the
projects,
and
I
mean
another
thing.
Another
stuff
is
the
research
project
I've
mentioned
before.
Yes,
it
is
about
the
research
projects
we
will.
C
We
will
submit
to
such
as
some
conference
or
some
journals,
yes,
yeah,
yeah,
so
yeah.
I
would
like
to
show
these
updates
in
this
meeting
yeah
today,
yeah
be
good
yeah.
A
All
right
yeah,
why
don't
you
go
ahead
and
do
that.
C
C
Yeah
yeah
yeah
yeah.
I
I've
I've
built
such
a
list.
Yes,
and
the
first
thing
is
about
the
gist
project.
Yes,
because
in
last
week
I've
tried
to
survey
some
data
sets
you
provided
in
slack.
Yes,
and
I
see
that
there
is
one
data
set
this
one.
It
is
a
csv
fire
which
includes
the
positions
of
cells
in
c
elegans
right,
yes,
but
I
I
didn't
really
understand
some,
the
meaning
of
some
data
such
as
this
one,
the
mean
division,
division
time,
what
I'm
not
sure?
A
A
So
this
is
for
independent
cells
in
the
lineage
tree.
So,
as
we've
discussed
the
lineage
tree,
you
know
you
go
from,
one
cell
divides
into
two
cells
and
c
elegans,
and
that
keeps
happening
throughout
the
tree.
So
each
one
of
these
cells,
the
labels,
are
a
specific
cell
type.
So
if
you
go
over
to
the
left,
you
have
aba,
which
is
cell
type
and
then
abal
and
aba
la
are,
you
know,
left
branch
to
the
left
into
the
anterior
portion
respectively.
A
So
those
are
the
names
of
the
cells
and
they
correspond
to
where
they
fall
in
the
lineage
tree,
but
also
in
anatomy,
and
then
the
these
measurements
here
x,
y
and
z.
These
are
measurements
from
cell
tracking
data.
So
when
you
get
a
cell
tracking
image,
usually
it's
a
two-dimensional
image
and
then
those
images
are
in
a
stack
and
then
you
have
so
you
have
different.
You
make
different
cuts
through
the
anatomy
from
the
top
to
the
bottom
of
the
worm.
A
Now
this
these
numbers
are
sort
of
you
know
averaged
over
that,
so
that
they
can
find.
You
know
you
can
find
the
location
of
the
cell
in
the
frame.
Now
the
problem
is,
is
that
for
each
cell
and
each
point
in
development,
these
these
coordinates
are
off.
I
mean
you
can't
just
map
them
out
raw
like
this,
so
you
have
to
do
some
sort
of
normalization,
and
so
actually
I
think
these
are
normalized.
A
These
are
on
this
scale
where
they've
all
been
normalized
and
they
you
know.
The
zero
point
on
the
x-axis
is
the
point
midpoint
between
the
head
and
the
tail.
A
The
zero
point
for
the
y
is
between
the
midpoint
between
the
left
and
the
right,
and
the
z
is
the
midpoint
between
the
top
and
the
bottom,
and
then
these
are
basically
you
know
deviations
from
that
zero
point.
So
if
you
see
a
large
negative
value,
that's
going
to
be,
I
think,
for
the
x
coordinates:
it's
it's
reversed,
so
that
the
negative
is
anterior
and
the
positive
is
posterior.
A
So
the
a
b
is
anterior.
That's
those
are
all
negative.
The
y
is
so
I
think
the
negative
is
left
and
the
positive
is
right.
So
you
can
see
here.
Aba
al
is
a
b
anterior
left
and
that's
negative
for
the
y
axis.
So
that's
off
to
the
left
and
then
z
is,
I
believe
the
negative
is
what
they
call
ventral,
which
is
to
the
bottom
and
positive
is
dorsal,
which
is
to
the
top,
and
so
those
that's
how
you
measure
that
you
can
plot
these
out.
A
As
a
a
point
cloud
and
you'll
see
this.
These
values
you'll
see
where
the
the
cell
falls
in
this.
It's
a
spherical
type
thing
it's
kind
of
oblong
because
that's
the
shape
of
your
worm,
but
they
basically
they're
all
sort
of
these
averages
across
different
worms
and
across
different
stack.
C
Photos:
okay,
okay,
so
do
you
have
the
original
that
I
said
told
this
csv
file.
A
Yeah,
so
this
originally
comes
from
a
data.
I
have
was
one
of
the
links,
I'll
straighten
it
out
after
the
meeting
in
terms
of
where
it
comes
from,
but-
and
you
know
I'll,
look
I'll
show
you
our
own
boarding
guide.
I
think
our
onboarding
guide
might
have
some
information
about
that
as
well.
For
you
like
direct
links,
so
then
the
mean
division
time
is
so.
The
idea
is
that
each
of
these
cells
have
a
time
from
zero,
zero
being
the
single
cell.
A
A
And
then
you
know
every
time
this
cell
divides,
so
you
can
see
like
abal
is
55.
Abaoa
is
74.
so
you're
following
that
down
the
tree
and
you're
getting
these
divisions
so
aba
and
abal.
A
That's
one
division
step
that
takes
about
13
minutes
for
that
division
to
happen
from
aba
to
aba
labar,
and
so
that's
the
that's
the
way
that
works,
and
so
these
should
form
a
tree
if
you
plot
these
out,
you'll
have
a
time
from
that
when
that
single
cell
divides
and
it
goes
out
from
there,
and
so
you
know
you
have
about
how
many
minutes
from
single
cell
to
hatch,
I
think
about
750
day
100
minutes.
A
And
then
okay,
yeah,
okay
and
then
the
volume,
I
think,
is
really
the
volume
of
the
cell.
It's
an
approximation
of
the
volume.
I
wouldn't
worry
about
that
because
that's
something
we
were
using
for
something
else,
but
this
is,
I
think
these
these
first
four
are
these
first
five
fields
are
the
ones
you
want
to
have
or
the
ones
you
want
to
work
with.
C
Yeah
yeah,
okay,
yeah,
yes,
yeah
yeah,
because
I
think
our
project
is
to
build
some
kind
of
graph
structure
with
the
information
of
these
positions
positions
of
these
yeah.
So
I
think
this
data
is
quite
useful.
I
mean.
A
C
Yeah
and
yeah
and
yeah
yeah.
A
Well,
also,
the
time
is
going
to
be
useful
too,
and
the
labels
are
actually
useful
because
you
can
take
the
labels
and
they're.
You
know
a
b
is
a
sublineage
of
the
the
one
cell,
so
this
is
like
a
lineage
where
you
have
the
biggest
sublineage.
If
you
read
that
paper,
I
gave
you
on
source
the
sulston
paper
on
the
embryonic
seminar.
B
A
C
C
Yeah
yeah,
yes
yeah
yeah,
because
in
my
proposal,
and
also
the
discussion
with
water
rule,
yes,
we've
separated
the
project
into
four
stages
and
the
and
it
should
be
three
stages.
The
first
one
is
to
convert
the
video
to
the
process
data
so
that
we
can
build
such
a
graph
on
this
process.
Data
and
the
process.
C
Data
should
include
some
position:
information
of
this
sales,
but
because,
in
last
several
weeks
I
and
whatever
we've
tried
to
utilize
the
development
on
these
videos,
but
we
found
some
bugs,
which
are
a
little
bit
hard
to
fix
in
that
balloon.
So
actually,
at
the
moment
we're
thinking
about.
If
we,
if
we
have
the
position
data
of
the
sales,
maybe
we
can
directly
build
the
graph
structure
on
these
positions.
A
Yeah
yeah,
I
think
well,
yes,
for
one
thing
with
evil
learn
you
have
to
use
a
certain
type
of
data
like
you
have
to
use
the
I
know
we
had.
You
have
to
use
a
higher
resolution
data
to
get
good
results
and
so
yeah
I
would
definitely
try
you
know.
I
definitely
use
these
annotations
and
then
keep
trying
with
the
images
and
then
I
I've
actually
been
talking
to.
A
Who
is
the
first,
the
sort
of
the
originator
of
the
core
divorm
diva
learn
platform,
and
he
was
telling
me
he
wants
to
start
doing
more
with
it,
and
so
that's
good,
but
also
I
I
I
encourage
you
if
you
find
bugs
to
report
them
as
issues
in
the
devo
diva
learn.
Are
you
familiar
with
the
issue
board
on
divalearn
in
the
get
up.
C
Yeah,
okay,
okay,
yeah.
I
will
report
this
box
in
the
issues
yeah,
okay,
yeah
yeah,
but
by
the
way
I
remember
I
I
don't
remember
where
some
contributor
of
the
dividend
has
sorry
for
the
moment:
okay,
uh-huh!
C
A
A
I'm
not
sure
I
think
this
may
have
been
like
just
a
general
example,
but
I
I
I
yeah
we'll
go
through.
I
know
there
were
some
data
sets
that
were
used
in
those
projects,
but
we'll
have
to
go
through.
It's
actually
in.
I
think
the
readme
one
of
the
readmes
in
the
main
repository
where
those
are
are
made
available,
but
there
are
a
number
of
data
sets
that
you
can
use
to
validate
the
sort
of
thing
you
know
you
can
use
your
own.
A
If
you
find
a
data
set
that
you
think
is
good
enough.
I
mean
we've
had
a
couple
just
with
images,
and
you
can
just
put
those
in
but
yeah
we'll
we'll
work
on
that
a
little
bit
more
to
make
sure
that
we
have.
A
C
Okay,
yeah
because
it
seems
like
these
data
sets
are
useful
and
the
dividend
can
work
on
these
data
sets
and
everyone
can
output.
Some
I
mean
useful
information
and
some
useful
results
based
on
this
data
set.
So
I
think
if
we
could
try
this
data
set
first,
maybe
it
would
help
us
better
understand
some
latin
logic
and
some
implementations
of
dividend.
I
think
yes,
but
it's
okay.
If,
if
I
I
can
also
try
some
other
data
sets.
Yes,
it's
okay,
yeah.
A
C
A
And
see
which
ones
work
best,
because
there's
going
to
be
a
lot
of
it
depends
on
you
know
the
nature
of
the
data
set
if
it's
like,
if
the
edges
are
clear
or
you
know
the
the
sampling
rate
and
then
that
will
affect
performance,
I
yeah.
I
would
try
a
couple
and
see
which
ones
give
you
the
best
result,
especially
for
something
like
this,
where
you're
trying
to
basically
approximate
a
cell
centroid,
you
know
it's
it's
it's
useful,
I
think,
to
have
like
some
a
couple
of
trials
to
see
which
work
best.
C
Okay,
okay,
yeah
yeah;
okay,
I
see
us
yeah,
yeah,
okay,
yeah,
I
see
and
also
I
got
some
other
updates
yeah
and
during
the
previous
discussions.
With
what
rule
we've
discussed
about
some
common
use,
graph
dividend
frameworks,
yeah
and
hello,
yeah.
C
Oh
yeah,
okay,
yes,
we
we've
discussed
about
some
frameworks
such
as
dgr
and
pyg,
and
a
network
x,
because
these
frameworks
are
common
use
in
in
the
field
of
such
as
the
graph
analysis
and
the
graphical
learning
yes,
and
because
actually
I'm
much
more
familiar
with
this
year,
because
I
was
once
working
for
digital
team
yeah.
C
So
I
think
maybe
we
would
choose
dgr
as
the
framework
that
I
mean
in
our
projects,
because
dgr
has
provided
some
api
that
can
convert
the
graph
structure
into
the
some
other
formats,
such
as
net
natural
x,
natural
x
and
and
the
network
x
also
provides
some
other
api
and
we
can
use
this
api
to
analyze
the
natural
properties,
yes
such
as
the
the
the
how
many
clicks
in
graph
and
degree,
distribution
and
some
other
kind
of
stuff,
and
also
the
year
has
provided
some
useful
of
strengths
classes
for
for
us
for
the
users
to
build
the
graph
structure.
C
C
Okay,
yeah
and
yeah
yeah,
and
also
I
have
another
question
about
that.
It's
about
because
I
see
that
I've
surveyed
some
paper
regarding
the
serial
segmentation
cell
tracking
and
I
will
introduce
some
of
them
later,
but
I
thought
that
there
are
a
lot
of
methods
which
can
deal
with
these
problems.
C
So
I'm
not
pretty
sure
if
we
need
to
offer
a
general
framework
that
can
that
can
employ
any
kind
of
models
on
any
kind
of
data
sets
or
all
the
pro.
Oh.
If
we
need
to
provide
some
examples
that
just
employ
some
existing
models
to
solve
some
problems
of
some
existing
data
sets.
I'm
not
sure
if
I,
if
I
I
I'm
clear
right
now,
yeah.
A
I
would
just
provide
some
examples
for
g-suck
and
then,
after
that,
we
can
talk
about
a
framework,
because
you'll
have
to
learn
like
what
you
know
how
this
works
and
how
well
everything
works
before
you
can
really
get
into
a
framework,
a
formal
framework,
but
yeah
I
mean
I
would
just
use
a
couple
of
examples
just
for
the
coding
period
and
get
those
down,
and
you
know
figure
out
how
this
works.
C
A
Yeah,
well
I
mean
you
know,
that's
that's
the
culmination
of
you
know,
maybe
a
year
or
two
of
work
before
they
even
release
it,
and
then
you
get
these
community
releases
that
are,
you
know,
built
upon
that
framework,
but
the
underpinnings
of
it
are
very
extensive,
so
yeah.
C
Yeah
yeah
yeah,
yeah.
Okay,
I
see
I
see
yeah,
I
understand
points
yes,
yeah,
yeah
and
yeah.
These
are
some
questions
regarding
the
project
and
and
maybe
in
this
week
we
will
focus
on
building
the
whole
project,
such
as
we
would
write
down
some
codes,
some
abstract
classes,
that
can
I
mean,
convert
the
video
to
process
data
or
convert
the
process
data
to
the
graph.
C
Yes,
we
will
start
coding,
I
mean
in
this
week
yeah
yeah,
yes,
yeah,
and
actually
I
have
some
updates
about.
Regarding
the
paper
I
mean
regarding
the
research
project:
yes,
okay,
just
like
yeah
what
I've
mentioned
several
weeks
ago.
Yes,
I
I'm
very
interested
in
the
sale
tracking
problem
at
the
moment.
Yes
just
yeah,
because
there
are
still
few
works
regarding
how
to
utilize
the
graph
new
networks
to
track
sales.
C
C
Yeah
yeah,
oh
yes,
yeah
and,
and
the
first
one
is
about
the
geometric
unm.
Yes,
and
the
first
one
paper
is
the
geometry
graph
convolutional
networks,
which
has
been
introduced
in
the
last
week.
Yes
and
and
this
paper,
as
I
mean
this,
these
three
items
in
this
part
about
the
geometry,
graph,
neural
networks
or
geometry,
deep
learning,
but
I
think
they
are
coming
from
the
different.
C
I
mean
respect
of
the
geometric
depending
the
first
one
is
to
map
graph
data
into
the
landing
space,
such
as
the
hyperbolic
space
and
to
aggregate
information
based
on
the
hyperbolic
space
and
the
second
one
paper,
which
is
the
hyperbolic
graph.
New
networks,
is
to
think
of
the
graph
structure
of
graph
data,
as
in
the
hyperbolic
space
and
then
map
data
into
the
ignition
tangent
space
to
employ
the
aggregation
functions
on
that
and
actually
yeah.
I
think
there
are.
C
There
are
a
different
respect
for
the
geometry
journal
and
the
the
second
item
is
about
the
manifold
lending,
which
is
maps.
The
high
dimensional
data
that
have
some
redundant
information
into
the
low
dimensional
manifold,
such
as
hyperbolic
manifold,
hyperbolic
space.
Yes,
and
sometimes
the
manifold
is
the
lattice
structure
of
data,
and
I
think
I
mean
the
shape
and
and
geometry
and
topology.
They
are
very
related.
To
the
I
mean
the
embryos
of
c
elegans
and
embryos
of
these
organisms.
C
So
I
would,
I
think,
maybe
it'd
be
better
if
we
can
start
from
the
geometry
and
topological
deep
blending.
I
mean
start
from
these
points
and
we
can
we
can
find
if
there
there
are
any
insights
of
combining
the
geometry
journal
and.
C
Organisms
and
the
developmental
biological
analysis,
so
I
mean
I
will
share
some
links
after
the
meeting
in
the
strike,
but
I
I'm
wondering
if
there
is
any
classic
paper
regarding
the
topology
and
geometry
of
organisms,
especially
the
sea,
elegance.
Yes,
I
I'm
not
really
familiar
with
this
field.
Yeah.
A
Yeah
there
are
a
couple
there's
some
work
on.
I've
seen
some
work
on
like
the
geometry
and
topology
of
embryos,
and
I
I
have
a
couple
actually
that
have
come
out
recently.
A
There
have
been
like
some
mathematicians
have
been
playing
around
with
c
elegans
embryos,
fortunately,
because
they're
very
easy,
they're
tractable
in
the
way
that
I've
described
here-
and
you
know,
they've
done
things
like
looked
at
the
sort
of
the
shape,
they've
sort
of
analyzed
the
shape
they've
been
our
new
shape
parameters,
they've
introduced.
Other
types
of
you
know,
approaches
to
looking
at
the
embryo
is
like
looking
for
holes
in
the
emerging
structure
of
the
connectome,
and
things
like
that.
A
I
can
share
some
of
those
papers
with
you,
but
yeah.
I
think
that's
a
a
wide
open
area.
In
some
ways
I
mean
I
don't
think
I
mean.
We've
talked
in
the
group
about
topological
data
analysis,
which
is
a
whole
new
other
field
from
graph
theory
and
networks,
but
you
know
they're
they're
doing.
B
A
Where
they're,
using
like
topology
to
sort
of
characterize
how
the
cells
are
organized
and
the
shapes
that
sort
of
emerge
from
that,
so
I
I'll
get
some
papers
on
that.
I
think
it'll
be
interesting,
an
interesting
direction.
I
don't
want
you
to
necessarily
go
down
that
alley,
but
just
to
see
what
they're
doing,
I
think
is
that.
C
Yeah
yeah
yeah
yeah
yeah
yeah,
yeah
yeah.
I
I
also
see
that
I'm
not
sure
the
type
one
of
the
papers
name,
we
titled
as
the
topological
graph
neural
natural.
I
I
I
didn't
mean
I
didn't
remember
its
name
correctly,
but
I
remember
that
she
has
used
some
kind
of
topological
data
analysis
in
the
june,
I'm
not
pretty
sure
yes
yeah
and
actually
I'm
yeah,
I'm
I'm
still,
I'm
still
learning
something
from
the
topology
and
geometry,
because
I'm
yes,
I'm
also.
I
I
I'm
not
really
familiar
with
this
part
as
well.
Yeah.
A
C
Yeah
yeah
yeah,
so
yeah,
yeah
and
there's
another
part
of
regarding
regarding
the
steel
tracking.
Yes,
I've
seen
read
some
papers
and
it
is
these
are
not
old
papers
in
this
field,
but
I've
tried
to
read
some
papers.
Yes
and
the
first
one
is
called.
I
will
share
these
links
in
slack.
Yes,
and
this
paper
serial
segmentation
and
tracking
in
phase
contrast,
images
using
graph
cards
with
symmetric
boundary
codes-
and
this
is
a
this-
is
a
paper
using
frame
to
frame
association.
C
I
mean
it's
a
segments,
the
stairs
in
a
frame
in
the
image,
and
it
try
to
propagate
some
prior
knowledge
of
this
segmentation
into
to
the
to
next
frame.
Next
several
frames
and
based
on
this
segmentation,
they
will
segment
the
sales
on
the
new
images.
Yes,
and
so
by
this
way,
they
can
segment
the
series
frame
by
frame
image
by
image.
C
Yes,
so
yes,
I
mean
so
I
mean
this
assumption
will
give
will
provide
some
basis
for
using
this
frame
to
frame
association,
algorithms,
yes,
there's
another
paper
called
fast
and
accurate
reconstruction
of
cell
cell
lineage
from
large
scale
microscopy
data,
and
it
is,
I,
I
think,
the
the
essential
notion
of
this.
This
paper
is
very
similar
to
the
first
one,
because
it
is
also
segmenting
the
sales
of
one
image
and
propagates
the
segmentation
results
to
the
next
image.
C
Yes,
I'm
not
sure
if
I
understand
correctly
yes
and
yeah
and
there's
another
paper,
which
is
a
little
bit
different
from
the
first
two
papers,
it
is
called
the
reliable
sale
tracking
by
global
data
association,
because
the
main
notion
of
this
paper
is
to
generate
some
short
tragedies
of
sales
first
and
then
combine
these
tragedies
into
the
wholesale
lineage.
So
it
seems
that
it's
a
global
association,
algorithms.
A
C
Yeah
and
yes,
these
papers
and
these
three
papers,
they
do
not
use
any
kind
of
technique
of
deep
learning
or
the
cnn
or
gm.
They
are
trying
to
utilize
some
conventional
techniques
such
as
some
linear
programming
and
some
other
other
techniques.
Yes,
and
I
think
in
the
following
several
weeks,
I
will
try
to
survey
some
cell
tracking
papers
related
to
the
deep
learning
or
machine
learning
kind
of
that
kind
of
stuff.
Yeah.
A
Yeah,
if
you
go
to
any
like
developmental
biology
talk
now,
they
like
well,
not
anyone
but
like
a
lot
of
them,
will
feature
automated
cell
tracking
that
they
have
like
they'll,
have
a
software
that
they
use
to
track
cells.
They'll,
get
some
microscopy
data
and
usually
they're
looking
for
things
like
cell
migrations
and
that's
quite
informative
in
development,
because
the
cells
will
migrate,
and
you
want
to
know,
especially
in
organisms
like
in
zebrafish
and
in
mouse
embryo
and
things
like
that.
A
The
cells
migrate
and
you
want
to
find
out
if
they're
migrating
together,
you
can
get
the
tracks
and
analyze
them.
We've
done
some
work
with,
like
positional
tracking
some
sort
of
crude
work
on
it
in
c
elegans,
where
we've
taken
like
second
order
tracks
like
plotting
the
position,
changes
in
the
position
and
but
yeah.
That's
something
too.
A
So
they
don't
all
use
deep
learning
for
this.
Some
of
the
you
know
some
of
the
automated
programs
will
use
something
like
linear
programming
and
then
they'll
compute
a
track,
and
it's
like
a
lease
cost
thing,
but
you
can.
You
know
there
are
a
lot
of
ways
to
do
this
and,
depending
on
what
you
need
to
do-
and
I
don't
again-
I
don't
want
you
to
go
down
this
rabbit
hole
very
deeply
because
it's
not
you
know,
whatever
you
think
is
something
that's
suitable
for
the
project.
A
You
know
try
to
work
on
that
and
then
you
know
we
could
work
on
that.
We
could
work
out
from
there.
I
mean
just
have
like
something
that
would
look.
You
know,
maybe
something
relevant
to
the
project.
Get
like
a
first
draft
going
of
this
is
how
we
might
approach
the
problem
of
tracking
cells,
and
then
you
know
we
could
work
on
it
from
there
after
the
pro
after
the
gsoc
coding
period,
because
there's
probably
a
lot
there
that
we
could
add
on
to.
A
But
you
know
the-
and
this
is
true,
I
think-
of
anyone's
project
in
the
group.
You
want
to
pick
like
the
sort
of
the
obvious
thing
that
you
can
work
on
to
put
it
into
your
software,
and
then
you
know
you
don't
want
to
take
like
six
weeks
to
do
this.
You
want
to
do
something
in
maybe
one
week
or
two
weeks
or
you
know
refine
it
a
little
bit.
A
If
you
have
time
at
the
end
and
then
after
the
gsoc
project,
there
will
be
opportunities
for
expanding
that
work
and
you
can
even
lay
out
a
pr.
You
know
a
a
plan
for
it
or
like
a
road
map
for
it.
If
that's
something
you
don't
you
can't
get
to,
and
you
have
an
idea
of
how
this
should
work,
you
know
laying
out
a
roadmap
instead
of
like
actually
putting
it
in
code.
If
you
don't
have
time
might
be
useful
as
well,
because
then
it'll
guide
people
in
the
future.
What
needs
to
be
done.
C
Okay,
okay,
yeah
yeah,
yeah,
yeah.
Sorry,
oh
yeah!
I
see
a
point
yeah
but,
oh
sorry,
I
got
a
phone
call.
I'm
sorry.
C
Yeah
yeah
yeah,
I
I
got
your
point,
but
I
yes,
I
I
think
yes,
these
these
topics
can
be
yeah
can
be
leave
up
after
the
gsa
project.
Yes,
but
I
mean
I
mean
if
we,
if
how
about,
if
we
can,
if
we
can,
regardless
this
these
topics
as
an
another
research
project
for
the
learning
of
graph.
A
Yeah
yeah
definitely
yeah.
We
could
do
that.
I
mean
that's
something
that
we
can
do.
Yeah
yeah,
I
think
there's
a
lot
to
work
on
yeah,
yeah
yeah.
Definitely.
C
Yes,
yes,
yeah,
so
yeah
yeah,
yeah,
yeah
yeah,
that's
what
I
mean.
Yes,
I
mean
yes,
we
got
a
gso
project,
yes,
and-
and
we
got
some
topics
regarding
such
as
geometry
and
or
cell
tracking,
and
and
we
can
also
put
these
topics
in
in
the
gsa
project
after
the
coding
period.
Yes-
and
we
can
also
try
to
put
these
topics
into
another
research
project
for
the
learning
of
graph
conference.
A
Yeah
you
know
this
is
this:
is
a
175
hour
project,
so
you'll
have
like
you'll,
be
roughly
working
on
it
for
20
weeks
during
each
week
of
the
coding
period,
so
you'll
have
other
you'll
have
more
time.
You
know
it's
not
like
you're
working
80
hours
a
week
on
it,
but
if
yeah
you'll
have
more
time
to
do
other
things,
just
I
want
to
make
sure
everyone
gets
their.
You
know
hits
their
benchmarks
for
the
week
like
or
hits
their
targets
for
the
week
so
that
we
can
move
through
and
yeah
yeah.
C
Yeah
yeah:
okay,
okay,
okay,
okay,
I
got
your
points
yes
yeah,
I
think.
Maybe
in
the
following
weeks,
I
will
try
to
put
some
targets
for
that
week
and
the
stuff
I've
done
in
the
previous
weeks
and
in
a
dog
in
a
document
so
that
I
can
give
a
better.
C
So
I
can
give
a
better
updates
for
you
in
this
meeting.
I
think
yeah.
A
I
like
this
notion
I,
like
this
notion
format.
I
know
jesse
likes
notion
to
just
kind
of
outline
things
now
I
wanted
to
show
next
I
wanted
to
show-
or
I
don't
know
we
can
have
some
updates
from
harry
krishnan
quran,
but
I
wanted
to
show
where
maybe
we
can
track
issues
at
different
stages.
I
do
so
in
github,
so
I'll
I'll
talk
about
that.
A
So
quran
or
hari
krishna
do
you
have
updates
for
us.
E
Yeah
hi
hi
bradley
hi
regarding
the
digital
microscope
project
I
was
like,
like
I
told
you
right,
I
had
gotten
to
the
contours
part
and
I
was
generating
more
outlines
for
that
embryo
image
data
set.
So
I
was
going
through
two
three
missing
algorithms.
You
know
because
after
I
have
the
3d
point
cloud,
you
know
I'll
be
using
open3d
python
libraries
meshing
algorithms,
so
there
are
like
two
three
of
them
like
there's.
This
meshing
algorithm
called
the
ball
pivoting
algorithm
and
there's
another
one
called
poisson
reconstruction.
E
So
you
know
I'm
just
seeing
out
of
these
two.
You
know
which
one
would
suit
the
model
better.
So
this
thing
is
new.
Otherwise,
regarding
even
jia,
hangli's
project,
you
know,
I
think
it's
very
very
similar
to
what
developed
was
like
it.
But
what
nanak
did
you
know
in
this
previous
year?
I
think
they've
shared
it
with
you,
because
I
was
even
going
through
their
projects
also
because
we
are
planning
on
doing
some
sort
of
a
collaboration
with
this
development
project
as
well.
E
E
It
has
a
lot
of
resources
about
you
know
doing
either
of
those
things.
E
A
Good
yeah,
so
you're
coming
along,
you
think
you're
getting
ready,
because
I
think
this
coming
week
is
the
coding
period
starting
up
so.
A
Everything
is
okay
and
yeah.
E
I've
already
done
most
of
the
things
you
know
that
I
had
outlined
in
the
like
this
week.
I
was,
I
think,
I'll
just
get
this
part
done
earlier,
because
the
outlines
part
and
all
those
things
you
know
were
supposed
to
be
done
from
weeks,
one
to
three.
I
think
week,
one
week
two
and
week
three
so
I'll
just
you
know,
refine
them
and
you
know
factor
the
code
in
a
more
neater
way.
You
know
so
that
if
anybody
else
wants
to
collaborate,
you
know
they
can
see
it
much
more.
E
Clearly,
I
think
this
is
what
I'll
be
doing
this
week,
otherwise
functions
and
all
those
things
you
know
already
mapped
them
out,
so
I
don't
think
they
will
take
more
than
a
few
hours
just
to
you
know,
factor
it
properly
yeah.
I
look
into
more
things.
You
know,
if
I
can
add
within
this
for
this
particular
week,.
A
Okay,
yeah
yeah,
if
you're
working
on
when
you're
working
on
your
from
your
schedule,
don't
forget
that
you
know
it's
it's
a
work
in
progress.
If
that
was
something
that
would
show
us
that
you
could
sort
of
organize
things
and
get
things
in
a
general
sense
of
you
know.
A
I
could
do
this
in
a
certain
amount
of
time,
but
you're
constantly
going
to
be
revising
your
schedule,
so
I
would
recommend
going
back
to
your
schedules
and
making
sure
that
you
make
sure
you're
hitting
what
you
want
if
you've
completed
something
just
kind
of
move
things
up
and
then
like
you'll,
have.
A
All
right
hurry:
krishna!
Do
you
have
an
update.
D
Yeah,
hello,
hi,
so
right
now,
I'm
going
forward
like,
like
I'm
experimenting
with
experimenting
with
the
method
in
which
I'm
taking
two
images
and
trying
to
find
the
angle
with
which
it
has
changed
and
trying
to
find
the
overlap
and
finding
a
new
image
and
using
that
new
image
to
overlap
with
the
next
image
so
kind
of
like
that
and
experimenting
with
it.
Maybe
if
it
works
I'll,
go
ahead
with
it
look
for
other
ways.
That's
it.
A
B
Very
I
have
a
suggestion.
This
goes
back
to
the
1960s
okay.
People
used
to
astronomers
would
look
for
stars
that
were
moving
relative
to
other
stars
by
taking
pictures
at
two
different
times
and
flashing
them
back
and
forth
on
the
screen.
B
B
B
B
Yeah
bradley,
maybe
I
should
indicate
what
we're
trying
to
do
here
is
the
the
problem.
The
problem
with
aligning
rotating
images
can
be
boiled
down
to
spherical,
coordinates,
theta
and
phi,
and
what
you're
doing
is
a
search
in
the
theta
phi
plane,
so
to
speak?
B
Okay,
now
the
problem
with
it
is
that
there
may
be
local
minima
okay
and,
if
they're
local
minima,
then
you
know,
if
you
get
the
wrong
alignment
automatically,
then
then
you're
a
local
minimum
and
then
you
have
to
you-
have
to
use
optimization
techniques
such
as
simulated
annealing,
to
try
to
get
past
the
local
minimum.
B
Okay,
and
by
putting
by
finding
the
overlap
between
two
consecutive
images
that
are
close
to
each
other,
we
may
be
able
to
we'll
do
two
things.
One
is
find
out.
B
If
there
are
local
minima,
because
you
can
track
the
you
can
track
the
the
differences
as
you
approach
the
minimum
and
the
other,
is
you
may
do
it
better
than
most
of
these
algorithms,
I'm
not
sure
yeah,
because
by
using
using
the
human
brain
you
you,
you
take
into
account
many
things
that
are
hard
to
take
into
account
by
by
computer,
and
you
may
disregard
things
like
differences
in
shading
and
stuff
like
that?
That's
right!
B
Okay!
So
it's
a
it!
It's
an
interesting
approach.
It
will
yield
data
as
to
whether
or
not
we
could
automate
it
by
knowing
by
getting
the
structure
of
the
of
of
the
landscape.
You
know,
in
other
words,
of
the
the
two-dimensional
heights
of
the
optimization,
optimization
curve
and
local
and
local
optimization
curves.
B
B
Okay,
all
right
and
susan's
susan's
got
a
large
number
of
data
sets
of
rotating
embryos
from
the
flipping
microscope,
where
the
angles
between
them
are
small.
So
it
should
be
relatively
easy
to
do.
The
overlap.
B
A
B
Yeah,
okay,
okay:
now
I
expect
it
will
get
harder,
as
the
number
of
cells
increases.
Probably
oh
yeah.
It
will
okay,
okay,
so
we'll
start
with
very
young
embryos.
B
Okay,
so
it's
it!
It's
sort
of
yeah
yell
them
to
the
computer,
that
which
is
the
computers
but
yeah
yeah,
we're
on
pain.
When
you
can't.
A
Yeah,
okay,
so
jia
hyung
had
to
leave.
He
said
that
the
koran
put
the
cell
tracking
challenge
link
into
the
chat
and
xiang
said.
Yes,
I
see
data
sets
in
cell
tracking
challenge
they're
very
common
used
in
experiments
a
lot
of
related
projects.
I
would
also
try
them
in
the
coding
period.
Then
he
had
to
leave
so
good.
I'm
gonna
actually
go
through
some
things.
A
Now
I
wanted
to
talk
about
some
of
the
things
we
might
start
doing
on
github
with
respect
to,
if
you're
going
through
your
project
and
you're
planning
it
out,
and
then
I
had
a
couple
more
things
to
talk
about
as
well.
A
So
so
as
for
like
planning
on
some
of
these
projects,
you're
gonna,
have
you
know
certain
sort
of
buckets
that
you
can
put
things
in
and
I'm
gonna
go.
I
don't
have
a
project
board
set
up
for
this,
but
you
know
in
our
gsoc
2022
repository
here
and
we
have
our
our
repu
or
our
directories
for
different
people
to
push
to.
A
I
think
I've
discussed
this
in
a
previous
meeting,
but
one
of
the
things
I'd
like
to
do
is
is
create
a
project
board
and
I
haven't
created
it
yet
where
we
have
three
buckets
and
that
first
bucket
is
going
to
be
like
the
things
that
maybe
we
want
to
do
in
the
short
term.
A
So
if
you
come
up
with
an
idea-
and
I
know
that
jia
hung
has
come
up
with
some
ideas
and
quran
and
hari
krishna
are
coming
up
with
ideas
and
what
tarou's
ideas
the
best
way
to
sort
through
these
is
to
put
these
in
a
project
board,
and
so
the
first
section
of
this
project
board
is
going
to
be
sort
of
like
a
to
do
so.
Let
me
do
this
on
my
board
here,
all
right,
so
the
first
column
will
be
to
do
and
that'll
be
kind
of
like
a
small
scale
deadline.
A
Maybe
this
is
stuff
that
we
can
do
within
the
coding
period.
You
know
it's
maybe
a
target,
but
we're
not
committed
to
it.
This
is
just
you
know,
maybe
like
a
something
with
a
maybe
a
couple
week:
deadline
or
a
couple
weeks
of
implementation.
A
So
these
are
things
that
you
might
not
do
in
the
coding
period,
but
a
relatively
small
scale,
and
you
know
I
said
to
jia
hong.
You
know
you
don't
want
to
go
down
too
many
rabbit
holes,
because
during
this
coding
period
you
want
to
get
through
it
rather
quickly.
You
want
to
hit
the
things
that
you
can
get
done
and
then
have
a
project
to
turn
in,
but
for
future
development.
You
want
to
have
these
to
do's
in
a
list
and
for
the
divo
learn
repository
we
have
or
the
devo
learn
project.
A
We
have
a
task
board,
that's
already
well
populated
with
things,
and
this
is
something
that
we
started
during
the
coding
periods
when
this
was
built,
but
also
things
that
afterwards
we
had
open
source
contributors,
filing
bug,
reports
and
issues
and
then
addressing
them
accordingly.
So
this
can
actually
be
quite
useful
to
revisit
things,
so
the
first
column
would
be
to
do.
A
The
second
column
would
be
nice
to
have
so
these
are
things
that
I
don't
know
necessarily
how
to
implement
them,
but
these
are
things
that
maybe
we
want
and
there's
a
medium
sort
of
deadline
for
that.
So
the
nice
to
have
stuff
isn't
really
the
to
do.
Stuff
is
stuff.
You
basically
know
how
to
do
it.
It's
very
constrained,
but
you
know
it's
something
that
maybe
you'll
address.
A
A
You
know
a
moon
shot
would
be
like
something
that
is
well
compared
to
that
scale
relatively
small
scale,
something
you
can
do
in
a
short
amount
of
time,
but
you
don't
know
if
you
can
do
it
or
not
when
you're
working
on
your
project,
nice
to
have
would
be
like
a
mars
shot.
Where
you
know
it's
a
larger
effort,
but
it's
not
something.
That's
really
long.
You
know
like
not
like
a
whole
separate
part
of
the
road
map,
and
then
future
directions
are
like.
If
you
were,
you
know,
sketching
out
a
road
map.
A
A
If
you
know
for
the
gnn
stuff,
maybe
like
some
sort
of
cell
tracking
algorithm
or
some
sort
of
cell
tracking
package
might
be
in
this
category,
whereas
nice
to
have
might
be
like
you
know,
some
little
module
that
does
some
cell
tracking
and
to
do
might
be
implementing
an
existing
module
that
allows
you
to
calculate
distances.
A
A
A
Then
I
have
a
couple
things
that
I
found
interesting,
and
this
is
something
that
morgan
has
been
turning
me
on
to
some
of
these
resources.
So
I
know
you
met
morgan
last
week,
but
you
know
he's
been
he's
always
putting
resources
in
our
one
of
our
other
slack
channels,
our
other
slack
team,
and
we
have
I
I
wanted
to
go
over
some
of
those.
A
This
one
is
a
hybrid
automata
library,
and
this
is
abbreviated
hal,
like
the
computer
in
2001-
and
this
is
something
that
is
was
developed
by
the
mathematical
oncology
team,
which
is
a
group
I
think
out
of
the
cleveland
clinic
and
they
have
this
open
source
software.
So
this
is
for
a
cellular
automata.
A
A
You
can
use
the
cellular
automata,
but
they're
also
spatial
models
that
have
agent-based
properties
and
they
have
partial
differential
equation
components.
So
these
are
things
that
we've
been
talking
about.
In
the
group
this
is
a
well
validated
system
for
this
house.
Components
can
be
broadly
classified
into
agent
containers
on
and
off
lattice.
You
can
have
containers
with
agents
in
them,
finite
difference,
diffusion
fields,
so
they
have
physics
with
graphical
user
interface,
components
and
additional
tools
or
utilities
for
computation
and
data
manipulation.
A
So
this
is
okay,
so
they're,
actually
at
the
moffett
cancer
center
as
well.
So
that's
in
florida.
So
this
website
hello
world,
which
is
there's
a
lot
of
this
2001
referencing
here.
But
it
basically
shows
you,
you
know
they
have
a
nice
open
source.
A
You
know
interface
here,
where
they're
trying
to
get
people
to
use
the
software
and
contribute.
So
they
have
this
github
repository
this
github
organization,
mathonco
hal,
and
then
they
have
this
website.
Where
they're
doing
a
description
of
the
software,
they
actually
published
a
paper
in
plus
computational
biology.
So-
and
this
is
you
know,
this
is
what
you
can
do
with
these
type
of
cellular
automata
models.
You
can
use
differential
equations
to
account
for
diffusion
in
a
setting
where
you
have
a
bunch
of
cells
and
they're.
A
You
know
signaling
each
other
and
they're
forming
patterns
based
on
local
chemical
gradients.
So
you
have
these
cells
here
that
are
these
agents
in
the
agent-based
model.
You
have
these
differential
equations,
which
are
diffusible
factors
and
they're
interacting
in
this
environment.
So
this
is
for
cancer
research,
but
it
might
be
useful
for
developmental
research.
A
You
know
you
can
do
two-dimensional
three-dimensional
grids
and
you
can
do
different
processes
like
births,
birth,
death
dynamics,
so
cells
are
born
in
cells,
die
and
that's
an
interesting
set
of
dynamics,
neighbor
interactions,
so
cells,
interacting
with
each
other
and
then
of
course,
movement
which
we've
talked
about.
They
deal
with
it
as
a
matter
of
advection
that
you
know
this
is
not
cell
tracking,
but
it's
a
you're
able
to
sort
of
generate
these
cell
tracks
using
this
type
of
model.
A
Then
this
other
thing
here,
if
you're
interested
in
some
of
these
methods
and
we've
talked
about
this
other.
This
other
platform
called
copy
cell
3d
and
I've
never
really
been
able
to
get
it
up
and
working
in
this
group.
But
it's
it's
it's
an
interesting
model
and
if,
but,
if
you're
interested
in
you
know
these
computational
models
for
developmental
biology
or
you
know
computational
biology
more
generally,
this
is
a
page
that
morgan
drew
my
attention
to.
A
It
has
a
lot
of
different
components
to
it:
it's
a
summer
summer,
school
and
symposium,
but
they
have
some
of
these
sessions
on
youtube.
So
they
have
sessions
here
from
different
from
different
perspectives.
So
if
you
want
to
model
cellular
networks
using
python,
there
are
a
number
of
speakers
here.
James
glazier
was
the
person
who
invented,
I
guess
compusal
3d.
A
So
there
are
a
lot
of
different
topics
here.
It's
it's
a!
I
think
it's
a
free
workshop
if
you're
interested
over
the
summer,
but
they
have
these
youtube
videos
that
are
sort
of
from
last
year's
virtual
symposium.
That
kind
of
go
over
a
lot
of
these
methods,
so
they
have
parameter
fitting
and
zomatic
modeling
network
modeling
software
testing,
reproducible
modeling,
which
means
being
able
to
get
the
same
result
over
and
over
and
so
forth.
So
that's
something
that
you
might
want
to
check
out.
Let
me
get
a
put
that
in
the
chat.
A
And
so
that's
I
just
wanted
to
show
those
resources,
because
I
thought
they
were
interesting
and
you
might
find
them
useful.
So
dick
said
he
has
a
reference
for
how
okay
yeah,
if
okay
so
jesse
says
he
can
make
a
board
if
he
has
access.
Actually,
I
can
give
you
that's
going
to
be
on
the
divorm
repository,
so
I
can
give
you
access
to
that
jesse.
A
A
So
this
is
on
stochastic
growth
and
form
this
paper
that
dick
wrote
in
the
60s.
This
is
a
pnas
paper
so
that
check
that
out,
if
you're
interested
and
then
yeah
so
jesse
said
you
can
start
making
issues
now
too
yeah.
So
this
is
yeah
so
we'll
set
that
up
and
I'll
send
out
a
message.
A
Also,
I
wanted
to
remind
people.
I
know
that
I
put
a
message
in
the
slack
for
all
of
our
gsoc
interns,
and
that
is
that
I
would
like
to
get
a
description,
personal
description
from
each
of
you,
for
I'm
going
to
put
a
blog
post
together
on
what
we're
doing
in
this
summer.
A
So
I
want
to
get
like
just
a
you
know,
two
or
three
pair
two
or
three
sentence
description
of
yourself
where
you're
going
to
school
your
interests,
maybe
outside
of
gsoc,
maybe
within
g
sock
as
well,
and
then
send
it
to
me
in
the
slack
as
a
response
to
my
mess
dm.
A
A
Okay,
so
yeah
next
week,
we'll
do
more
gsoc
updates
this
coming
week
is
the
beginning
of
the
coding
period
and
we'll
be
working
on
these
sort
of
side,
channel
conversations
about
getting
data
and,
as
susan's
really
excited
about
doing
some
things
at
the
microscopy
data
and
we'll
be
doing
more
updates
on
some
of
the
projects
and,
let's
start
moving
forward
on
this.