►
From YouTube: DevoWorm #27: GSoC project updates, Symmetries, GNNs and the Topological Analysis of Development
Description
GSoC Coding period (week 8): Updates on Digital Microspheres and D-GNNs. Developmental Symmetries and Asymmetries. An overview of NeuroMechFly (virtual Drosophila). Discussions of applying GNNs/topological models to spatiotemporal representations of development and organisms across the tree of life. Attendees: Richard Gordon, Harikrishna Pillai, Jesse Parent, Karan Lohaan, Sushmanth Mereddy, Longhui Jiang, Anant Kumar, Jiahang Li, and Bradly Alicea
A
B
A
Welcome
to
the
meeting-
and
I
don't
know
where
the
rest
of
our
g-sock
students
are,
but
just
krishna
wanted
to
give
an
update
as
to
where
he
was.
That
would
be
good.
B
Yeah
so
I'll
give
my
update
okay,
so
this
week
I
couldn't
do
anything
much
extra,
but
in
this
week
first
of
all,
what
I
did
was
I
tried
to
make
a
readme
file.
I
documented
the
things
in
stages
which
I
did
till
now
and
also
I
worked
a
little
on
the
ui
part
of
the
application
which
I
was,
but
I
couldn't
do
much
because
my
semester
got
over
and
I
was
stamping
back
to
my
home
so
most
of
the
time
traveling
and
packing
so
yeah.
B
B
That's
good
and
I'll
also
update
the
repository
today
in
the
github.
A
Yes,
let's
take
a
look
at
that.
Let's
see.
A
Okay,
so
this
is
yeah.
This
is
the
g-suck
repository,
and
this
is
the
one
here.
A
A
Yeah,
okay,
let
me
stop
sharing
well,
that
looks
good,
so
your
plans
are
to
do
some
work
on
project
image
projection
next
week
and
start.
B
A
Yeah,
okay,
I
don't
know
yeah,
so
that's
all.
I
think
that's
the
only
update
we
have
this
week
so
yeah
I
have,
and
I
don't
know
if
it's
for
this
project
or
not,
but
we
can
extend
the
projects.
Actually,
I
think
it's
for
all
projects.
We
can
extend
the
projects
into
the
fall
if
people
need
to
so,
if
you're
needing
to
get
have
an
extension.
I
just
need
to
know
so
I
can
let
them
know.
A
A
So
if
people
are
running
short
on
time
or
you
know,
I
know
that
we've
like
in
the
summer
it
gets
choppy
and
you
don't
get
everything
you
want
done
if
it's
just
a
matter
of
you're,
not
getting
everything
you
want
done,
though,
don't
ask
for
an
extension.
Just
if
you
don't
think
you
can
finish
by
the
august
deadline.
A
I
was
actually
a
bit
confused
about
that.
I
thought
that
only
some
of
the
projects
ran
long,
but
I
think
all
of
them
can
run
22
weeks
if,
if
need
be
so
okay,
so
I've
got
a
couple
things
today.
I
have
some-
maybe
I'll,
show
my
screen
again
a
little
bit
here.
A
All
right,
so
first
thing
I
was
going
to
talk
about
is
I
found
this
interesting
resource,
it's
a
neuromechanical
model
of
drosophila,
so
it
says
neuromac
fly
and
I
think
it's
run
out
of
epfl
in
europe,
and
so
this
is
a
project
that,
where
they
take
drosophila,
which
is,
of
course,
another
model
organism
and
they're,
using
evolutionary
algorithms
and
other
types
of
of
computational
models
to
model
the
neuromechanics
of
the
fruit
fly.
A
So
neuromechanics
are,
when
you
know,
there's
some
motor
activity
and
it's
controlled
by
the
brain,
and
they
look
at
not
just
the
brain,
but
also
the
mechanics
of
the
limbs.
So
there's
physics
and
there's
neurobiology
involved,
there's
also
some
modeling
of
the
loop
between
the
sensory
motor
loop
between
incoming
sensation
and
motor
control.
So
this
is
a
computational
model.
With
this
they
have
a
fair
amount
of
code
here
in
this
repository
they
have
it.
A
This
is
actually
based
on
some
experiments
that
they've
been
doing
so
they're
collecting
their
own
data
at
the
at
their
lab
and
then
they're
actually
doing
the
modeling.
So
this
is
kind
of
a
nice
resource
if
people
are
interested.
So
this
is
the
virtual
fly
here.
A
If
you
know
just
so,
you
get
an
idea
of
what
it
looks
like
and
it
looks
like
a
lot
of
robotics
models
that
people
put
into
virtual
environments,
and
so
this
is
kind
of
a
nice
and
they
also
have
pose
estimates
from
the
deepfly
3d
database
and
there's
a
repository
along
with
that.
So
there's
a
lot.
There
are
a
lot
of
resources
here.
A
If
you're
interested
in
movement,
we
also
have
the
movement
database
for
openworm
for
c
elegans,
so
you
know
we
could
do
cross
species
comparisons
there
if
people
are
interested
so
there's
a
lot
of
stuff
there
potentially
to
work
with
another
thing
that
I
have
here
is
this
a
special
issue
of
royal
society
interface
focus
and
I
found
out
about
the
about
this
this
weekend.
A
A
I
don't
know:
okay
here
we
go
so
adam
saffron
is
one
of
the
editors
michael
levin
a
deal
rousey.
I
don't.
I
don't
know
any
of
these
other
people,
but
I
do
know
adam
and
michael.
A
So
that
would
be
something
that
you
know
be
interesting
to
submit
something
to
from
the
group
there's
taking
submissions
currently
for
this,
and
what
this
is
is
it's
a
special
issue
of
interface
focus,
exploring
the
role
of
symmetry
and
complex
adaptive
systems
across
scales,
so
this
is
emerging
dynamics
of
biophysics
to
the
shaping
of
underlying
adaptations
through
ontogenetic
and
phylogenetic
processes.
A
So
this
is
basically
what
we
do
in
the
group.
We
talk
about
evolution
and
development
and
we
talk
about
some
of
these
other
biophysical
mechanisms,
and
so
you
know
it's
just
a
matter
of
finding
like
a
topic
and
writing
up
an
abstract
on
it.
I
have
some
ideas
on
how
to
this
might
work
and
we
have
some
things
in
the
group
that
are,
you
know
probably
good
for
this,
but.
B
A
Yeah,
yes,
it
might
be
like
a
mashup
between
some
of
the
things
we're
doing
in
the
developmental
ai
group
and
the
other
in
the
lab,
and
then
the
stuff
we're
doing
with.
You
know
other
stuff
we're
doing
in
this
group.
So
it
might
be
an
interesting
way
to
go.
A
This
is
gage
fields
and
equivariance,
so
they
have
these
like
different
topics
that
they
focus
on
for
symmetry
in
physics
and
then
there's
symmetry
breaking
in
development,
which
is
an
analogous
to
that
where
there's
a
sort
of
an
asymmet
introduction
of
an
asymmetry
and
development.
So
when
there's
like
a
you
know,
there's
the
division
between
the
anterior
and
posterior
poles
in
a
two
cell
embryo,
that's
symmetry
breaking,
because
you
get
two
different
parts
of
the
organism
that
develop.
Evid
now
have
a
develop
different
developmental
history.
A
A
I
think
that
was
something
that
that
might
be
something
we
could
address
asymmetries
and
cognition.
A
So
there
are
psychological
arrows
of
time
and
brain
organization
or
hemispheric
differences,
so
the
brain
itself
has
a
lot
of
asymmetries
in
like
processing
and
things
like
that.
A
Symmetries
as
inductive
biases,
identification
of
invariant
features
and
objective
functions
for
natural
and
artificial
learning
systems.
So
this
is
this
has
a
lot
to
do
with
different
types
of
processing
and
you
know
symmetry
and
asymmetries,
and
I
think
they
want
to
move
towards
like
simulating
this
stuff.
So
it's
like
you
know
they
want
to
identify
a
lot
of
these
things.
Maybe
computationally,
I
think,
that's
probably
where
a
lot
of
the
papers
are
going.
A
The
diverse
rules
of
symmetry
is
in
cognition
and
behavior.
These
are
generalized
models
of
inference
and
coupled
systems
mirroring
and
coordination
dynamics
models
or
effect
across
brains
and
organisms.
So
this
is
again
kind
of
like
this
idea
of
complex
systems
and
active
inference,
something
they
call
active
inference
and
looking
at
that
in
cognition
and
behavior.
A
So
I
think,
there's
a
list
of
papers
here
if
I
can
find
it
because
this
is
the
first
page
of
design.
C
C
So
we
have
the
possibility
of
measuring
strain
and
if
we
resurrect
oh
wayne
broadland's
program
for
for
figuring
out
stresses,
we
might
be
able
to
get
that.
So,
in
other
words,
there
could
be
a
mechanics
of
symmetry
breaking.
A
C
C
I
respect,
and
that
would
also
clear
up:
that's
not
the
nonsense
about
morphogenetic
fields.
A
C
A
It's
it's
open,
but
it's
probably
as
soon
as
it
looks
like
they
were
getting
some
submissions
but
as
far
as
I
know,
it's
still
open.
So
I
don't.
I
can
find
out
what
the
I
don't
think.
There's
a
part.
A
A
C
A
C
Yeah,
it's
a
program,
he's
retired
now
not
doing
any
science.
Okay,
so
getting
a
hold
of
the
program
and
getting
it
resurrected
might
take
some
work,
yeah,
okay,
but
what
it
did
that
I
I
had
a
student
back
when
oh
boy
1978.
C
his
name
was
murray
steen
and
he
did
a
paper
on
simply
bubbles
between
glass
plates
and
we
found
that
we
could
do
the
null
case.
Looking
at
the
curvature
of
the
bubbles,
we
could
figure
out
that
they
had
settled
down
to
zero
forces
between
them.
C
Okay,
yeah
and
wayne
generalized
that
to
a
an
inverse
finite
element
method
to
give
the
forces
between
all
cells.
C
B
A
Right
so
some
of
the
featured
submissions
they
have
so
far
here:
physics
of
creation,
symmetry
breaking
and
active
inference
and
unfolding
state
spaces,
free
energies
and
inference
and
living
systems.
So
there's
some
submissions
on
what
they
call
the
free
energy
principle
and
active
inference.
A
Then
they
have
an
elementary
model
for
the
emergence
of
symmetry
concepts
by
agent
collectives.
So
there's
some
stuff
on
emergence
and
agents
and
like
groups
of
organisms
that
behave
collectively,
then
there
is
principia
qualia.
Is
that.
A
A
It's
let's
see
me:
let's
see
talks
about
the
role
of
contingent
structure
in
history
and
explaining
the
organization
of
biological
systems.
It
looks
like
it
kind
of
goes
into
doesn't
really
get
into
the
origins
of
life.
It
just
looks
like
it's
kind
of
nibbling
around
the
edge
of
that,
but
not
necessarily.
B
C
A
C
C
Okay,
do
me
a
favor,
send
me
a
url
for
the
abstract.
A
Yeah
yeah,
I
will
okay,
so.
A
C
Because
it
might
have
been
part
of
origin
yeah,
you
know
my
nick
devon
and
I
and
natalie
are
working
on
this
book
on
origin
of
life.
Okay.
So
if
there's
some
ideas
that
we
might
use
them.
A
So
there's
the
prince
of
equalia,
which
is
about
consciousness
deformation
and
a
symmetry
theory
of
valence.
So
valence
is
a
and
a
concept
of
emotion
in
the
brain,
where
I
guess,
they're
looking
at
consciousness
and
and
emotional
states,
the
collective
mind
and
experimental
analysis
of
imitation
and
self-organization
in
humans
and
then
a
geometry
and
functional
calculus
for
bayesian
mechanics,
which
is
more
about
sort
of
kind
of
gets
into
some
bayesian
methods
of
dynamical
systems.
B
A
They're
different
topics,
but
I
think
the
there
are
a
lot
of
it's
very
heavy
on
the
origins
of
cognition.
It
seems
like
I
was
looking
at
some
of
these
papers,
so
I'm
looking
over
these.
These
are
the
abstracts
and
there's
stuff
on
the
free
energy
principle,
there's
stuff
on
some
consciousness.
There's
stuff.
B
A
A
A
So
I
see
koran's
here
welcome.
B
Slack
because
I
I
I
feel
like
I
know,
knew
about
that
or
forgot
about
it,
but
I
really
want
to
submit
something
to
that.
A
Yeah
yeah,
we
can
talk
about
it
a
little
bit
more
and
maybe
get
some
see
where
we
are
with
that.
You
know
like
I.
I
don't
know
how
long
it's
open.
It
should
be
open,
probably
for
a
while
it
looks
like
they
don't
really
have
enough
submissions
that
on
issues
so
they'll
probably
be
accepting
for
a
while.
Yet.
A
Yeah,
hello,
kamar
or
hello,
quran.
B
Yeah
so
far
I've
just
been.
You
know
improving
the
draft
of
the
blog
that
I'm
writing
the
description
part
with
some
extra
images.
You
know,
so
I'm
clicking
images
of
the
things
that
I'm
doing
then
the
changing
for
changing
the
like
last
week,
I
was
trying
to
you,
know,
use
a
different
method
for
feature
detection
part.
You
know
detecting
features
across
the
two
images
so
that
I
can
approximate
the
angle
of
rotation.
So
earlier
I
was
using
surf,
so
I
was.
B
You
know
using
all
the
heat,
even
though
it
still
so
I'm
thinking
of
you
know
using
some
more
specific.
You
know
image
augmentation
techniques
to
improve
the
number
of
features
it
can
detect.
It.
A
A
B
The
thing
is
I,
based
on
I'm
using
two
adjacent
images.
B
So
I,
with
whenever
I
consider
two
images
at
a
time
I
I
give
them
an
approximate,
you
know
access
about
which
you
know
they're
rotating
and
based
on
that
access.
I
checked
the
feature
and
how
much
it
has
moved
the
displacement
of
that
feature.
So
let's
say
you
know
this.
There
was
a
crevice,
you
know
a
high
higher
pixel
intensity
or
a
darker
line.
You
know
on
that
embryo,
so
the
amount
that
it
has
moved.
You
know
with
respect
to
the
approximate
axis.
A
C
Oh
okay,
okay
got
it!
I
I
just
if
you
look
at
the
chat,
I
had
trouble
with
pronouncing
this
name.
I'm
working
okay
and
I
have
a
paper
submitted
on
a
new
algorithm
for
rotating
images.
C
C
You
may
know
that
if
you
take
a
two-dimensional
image
and
rotate
it
okay,
again
and
again
and
again,
it
gets
worse
and
worse
with
this
algorithm,
it
doesn't.
C
A
That's
good,
so
that's
see
what
are
you
planning
on
doing
for
next
week,
quran.
B
For
this
coming
week,
I'll
I'll
probably
have
a
good
projection
image.
You
know
model
that
I
might
be
able
to
show
you
know
so
I'll.
Just
improve
the
projection
image
generation
part
so
that
it
can
be.
You
know,
projected
onto
the
model
itself,
so
I'll
be
refining
the
current
techniques
in
the
deck
outline.
For
that.
A
Yeah,
okay,
that's
good!
Thank
you
for
the
update
and
again,
as
I
told
hurry
krishna.
If,
if
you
need
a,
I
don't
know
if
you
need
an
extension
for
the
project,
but
I
found
out
that
we
can
extend
them
up
to
22
weeks.
If
you
let
me
know
if
there
are
any
problems
that
you
run
into
that,
you
think
you
need
not
if
you
want
to
add
in
all
sorts
of
different
things
but
like
if
you
have.
A
If
something
comes
up
and
you
need
to
delay
your
submission
so,
okay,
so
we
have
a
new
member
here:
anan
kumar,
hello,.
B
Yeah,
I'm
a
student,
I'm
student
at
amrita
right
now,
so
I
found
this
org
by
twitter.
I
stumbled
upon
it.
I
got
interested
so
I
joined
so
I
joined
like
the
slack
and
all
that.
B
So
like
right
now,
like
I'm
working
on
like
ml
models
and
stuff
right
now,.
B
A
We
have
a
lot
of
people
doing
that
sort
of
thing
in
the
group,
so
yeah
yeah
well
welcome.
You
know
you're,
welcome
to
attend
meetings
and
join
the
slack
and
everything,
and
if
you
find
something
interesting,
you
know
let
someone
know,
and
then
you
know
we
can
maybe
loop
you
into
some
sort
of
project.
If
you
want-
or
you
know,.
B
D
Yes,
I
have
some
ideas
and
I
have
some
other
updates.
So
let
me
share
my
screen.
D
All
right
so
chris
could
use
my
screen
now.
A
D
D
Many
about
the
data
sets
many
about
those
microscopy
videos
and
because
it
is
a
little
bit
difficult
to
pre-process
this
video
data
set
and
to
convert
them
into
the
graph
structure,
so
we
are
still
working
on
them.
Yes,
it
is
a
little
bit
difficult,
so
you
will
see
that
the
progress
is
a
little
bit
slow.
Yes,
we
are
pushing
pushing
it
yeah.
So
besides,
besides
this
this
process,
besides
the
data
sets,
I
need
to
make
sure
you
could
see
my
screen.
Okay,
yes,.
A
D
Besides
those
data
sets,
actually,
I
see
the
bottle
and
longhui
are
working
on
this
micro
microscopy
video.
So
I
I'm
working
on
some
other
parts
which
are
which,
which
is
the
model
design
and
the
motivations
of
our
work
on
something
else,
so
I
will
briefly
introduce
some
updates
of
our
last
few
weeks.
Yes,
yes,.
B
D
Yeah,
the
I
I
mix,
I
make
a
slice,
so
that's
you
could
see
my
you
could
understand
my
my
presentation.
Clearly,
yes
yeah,
so
the
topic,
the
topic
of
the
day
is
the
graph
new
network
and
serial
developmental
analysis,
and
I
am
focusing
on
the
search,
hacking,
yeah
and
probably
you
remember
that
in
the
last
few
weeks
I've
discussed
some
other
ideas,
two
durian
design,
which
is
which
like
this
yes
and
today,
I
want
to
talk
about.
D
First
of
all,
I
want
to
talk
about
the
motivations
of
our
work
and
then
I
want
to
talk
about
the
model
design
itself,
because
I
found
I
found
that
in
the
last
few
weeks
I
thought
that
this
model
design
is
a
little
bit
unreasonable
and
it
can
really
work
well,
yeah.
D
For
example,
why
do
we
need
to?
I
mean
the
questions
here?
You
can
see
these
those
questions.
Some
of
them
have
no,
I
mean
satisfying
answers,
so
I
would
appreciate,
if
any
one
of
you
could
discuss
these
questions
with
me.
Yes,
that
would
be
better
and-
and
so
first
of
all,
it's
about
our
motivation.
Why
do
we
need
to
model
data
as
a
graph
and
employ
graphical,
although
there
is
only
one
paper
which
has
thoroughly
discussed
this?
This
question
this
motivation.
D
I
think
we
need
some
more
discussions,
because,
although
this-
although
that
paper
has
been
accepted
by
a
conference-
but
I
think
we
still
lack
sufficient
motivation
or
sufficient
answer
for
these
questions,
because
there's
only
one
paper
discussing
the
gerund
and
the
sale
tracking
or
set
developmental
analysis.
D
D
And
actually
I
think
this
question
is
very
important,
because
I
think
our
main
points
is
first,
we
want
to
use
adrian
to
solve
some
problems
and
the
second
we
want
to
consider
the
topology
of
data.
D
The
target
is
the
same
so
so
this
factor
will
make
modern
learning
more
difficult.
Actually,
yes,.
A
D
Is
the
first
point
yes
and
the,
and
the
second
point
is
why
we
need
to
consider
the
topology
or,
more
specifically,
what
we
need
to
consider.
The
local
topology
is
actually
when
we
consider
the
topology
we
will.
We
actually
do
some
calculations
based
on
the
3d
positions,
so
it
seems
like
a
topology
can
be
regarded
as
a
as
a
map
or
function
which
map
positions
to
some
some
special
information
or
some
special
outputs.
D
So
we
can
consider
the
local
topology
as
a
measure
of
the
local
shape,
such
as
the
void
or
the
circle,
and
my
argument
is
that
is,
it
is
more
important
to
say,
and
I
would
I
would
try
to
use
some
examples
to
explain
this
these
points.
D
So
please,
with
a
moment,
could
you
see
my
screen
now?
Yes,.
A
D
Okay,
okay,
yeah.
So
my
point
is
probably
you
remember
that
in
the
last
few
weeks
I
my
argue
is
that
if
we
have
a
points
crowd,
sorry
sorry.
D
If
we
have,
I
have
a
points
card
and
this
points
code
actually
are
those
cells
in
a
certain
frame
and
given
a
point,
we
can
explicitly
con
consi
calculate
some
topological
information
of
this
points
cloud
such
as
the
persistent
diagram
and
some
other
kind
of
tools.
This.
However,
you
will
find
that.
However,
I
remember
my
argue
was
consider
these
points
as
l
is
this
here
as
a
and
we
consider
the
topology
and
its
its
neighborhood,
such
as
this
neighborhood
and
my
other
words.
D
I
mean
I
I'm
not
sure
if
you,
if,
if
I
I
explain
this
clear
clearly
because
it
because
the
sale
tracking
task
is
about
the
cell
developmental
process,
so
we
need
to
consider
the
term
dimension
more
than
just
the
space
dimension,
but
here
we
only
consider
the
topology
of
this
coins
cloud.
So
I
think
it
is
not
very
it's
not
really
sufficient.
D
So
I
remember
I've
read
some
papers
which
discussed
the
topology
of
cell
linear
tree,
assume
that
we
have
been
in
a
tree
like
this
and.
A
A
D
D
Oh
sorry,
I'm
not
sure
sorry,
so
you
could
see
that
you,
you
cannot
see
the
those
red
drawings
no.
A
D
Oh,
what
said
this
mode,
I
could
not
show
I
could
not
screen
drawing
actually.
A
D
So
so,
with
a
moment
I
I
will,
I
will
first
just
finish
the
drawing
and
I
will
explain
it.
Okay
for
the
moment.
Okay.
D
Okay,
it's
I'm
not
sure
if,
if
it
is
kadir.
B
D
Okay.
Okay!
Yes,
yes,
my
point
is
that
the
left
protest
is
the
points
crowd
of
our
student
frame
and
the
right
plot
is.
This
is
a
democrat
linear
trade?
Yes,
and
this
this
this
we
can
call
it
cross
or
x,
they
are
actually,
they
represent
the
cell,
the
cra
and
the
cell
b,
and
this
circle
represents
its
neighborhood
along
the
cell
linear
tree.
Yes,
yes,
so
is
it
clear.
A
D
Yeah
yeah
yeah,
yes
yeah.
I
I
will
make
him.
I
will
make
him
more
clear
and
ask
out
yeah
yeah,
just
just
so.
Actually
my
point
is
that
we
can
so
we
can.
We
should
consider
the
topology
of
the
cell
linear
tree
more
than
just
a
topology
of
the
point
scout
of
a
certain
frame.
I
mean
I
mean
look
at
this
celina
tree,
actually,
the
cla
and
the
clb
they
we
can
think
that
they
are
located
in
different
frames
right,
yes,
yeah.
They
are
located
in
the
different
frames.
D
We
can
think
of
that,
and
but
the
cra
and
serbia
they
might
have
similar
navals
neighborhood
in
their
own
points
cloud.
Is
it
clear.
D
Yeah
yeah,
I
mean
I
don't
know
how
to
do
drawing
here
I
mean
oh
yes,
I
can
do
this
wait
a
moment.
Okay,
I
mean.
D
I
mean
just
assume
that
this
rectangle
represents
a
certain
frame,
and
we
suppose
that
this
frame
has
a
point
scout
like
that
like
like
this,
and
this
is
node
a
and
first
we
consider
the
neighborhood
in
the
space
dimension,
which
is:
is
this
the
neighborhood
of
a
and
another
neighborhood
along
the
linear
tree?
Is
this
circle
right.
D
D
There
are
two
different
neighborhoods
and
the
first
one
is
in
the
space
dimension
and
another
one
is
along
the
linear
tree
dimension,
which
is
a
temporal
dimension.
That
means
why
do
we
need
to
consider
the
temporal
dimension?
The
topology,
the
topology
of
the
temporal
dimension
is.
Is
that
because
we
need
to
consider
the
time
dimension
of
the
whole
things,
because
we
are
talking
about
the
cell
tracking?
D
That
means
our
final
target
is
to
is
to
find
the
absolute
position
of
a
in
this
cell
linear
tree
right,
although
most
methods
are
considering
considering
about
how
to
associate
cells
across
different
frames,
but
their
final
target
is
to
to
construct
such
a
cell
linear
tree
and
to
find
the
position,
the
absolute
position
of
a
of
cell,
a
of
cell
b
and
other
cells
in
this
cell
in
the
tree
right,
yeah
yeah.
D
Yes,
so
we
need
to
consider
the
temporal
dimension,
so
we
need
to
consider
the
topology
of
this
cell
linear
tree,
although
probably
my
explanations
are
still
unclear,
but
my
my
point
is
we
I
mean
my
point
is
where
we
need
to
consider
the
topology
of
this
points
cloud
in
a
certain
frame,
and
we
also
need
to
consider
the
topology
on
along
this
this,
the
only
linear
tree.
That
is
because
we
have
assume
that
we
have
two
nodes:
a
two
cells,
a
and
b.
D
They
are
located
in
different
frames,
but
probably
cla
has
a
similar
neighborhood
as
cell
b
and
the
neighborhood
here
is
divided
in
the
points
cloud
instead
of
the
cellular
tree.
I
call
this
the
space
neighborhood
and
this
base
neighborhood
and
probably
the
a
and
the
cell
b,
have
similar
space
neighborhood
and
we
cannot
differentiate
a
and
b
based
on
their
space
neighborhood
because
they
have
similar
space
neighborhood.
D
A
A
To
do
yeah
it's
it's
basically
like
connecting
the
the
lineage
tree,
the
time
information
with
the
spatial
information
and
sort
of
making
the
connection.
So
you
have
this
neighborhood
of
points
and
then
they
evolve
in
time
and
they
may
move
around
and
you
get
like
this.
Basically
single
set
or
the
single.
A
I
guess
it
will
be
some
topological
interpretation
of
that
for
a
and
for
b
and
and
so
on,
so
you
can
track
them
much
more
easily,
because
that's
one
thing:
we've
had
a
problem
with
is
going
over
time.
Like
the
positions
are,
we
know
the
positions
and
discrete
points
in
space.
We
can
average
them,
but
that's
not
great.
If
you
want
to
look
at
the
time
evolution
because
the
cells
do
you
know
they
divide,
you
have
daughter
cells,
those
migrate
to
some
extent
they
you
know.
A
So
the
position
isn't
always
clear,
like
you
can
normalize
for
every
time
point,
but
that
still
leaves
you
with
like
this
weird
weird
set
of
relationships,
so
you
yeah.
I
think
that
this
is.
This
is
a
little
bit
better
way
of
doing
it.
You
can,
I
think,
it's
more
continuous
in
terms
of
the
mathematics
and
in
terms
of
the
tracking
of
the
cells,
okay,
yeah
bradley,
mr.
C
Interject
here
motivation,
one
possibility
is
to
track
bacillaria
and
track
them
long
enough,
so
that
one
sees
some
vision.
B
B
C
And
this
could
be
relevant
for
diatoms
in
general
or
for
any
linear.
Well,
cyanobacteria
do
this
to
for
any
linear
colony,
the
the
colony
would
be.
The
geometry
of
the
economy
would
be
simple
because
it's
linear,
but
some
colonies
develop
specialized
cells
in
specialized
positions,
and
so
one
might
be
able
to
work
out
not
only
tracking
the
cells
but
building
up
a
tree
of
when
different
kinds
of
cells
occur.
A
C
C
Okay,
yeah
and
one
other
thing
opportunity.
Just
this
morning,
we
got
a
proposal
for
a
book
on
chain,
diatoms.
B
B
C
Could
be,
this
could
could
be
a
a
contribution
to
the
time
skill
would
be
adequate.
Yeah.
A
Yeah,
that
would
be
good,
so
yeah.
This
is
great
work
jiahung.
So
this
is.
This
is
sort
of
like
trying
to
solve
a
problem
that
you've
encountered
and
then,
of
course
it
has
applications
to
other
things
so
yeah
we
have
other.
We
have
other
data
beyond
c
elegans
that
we
could
work
on
with
this
too.
So
this
isn't
just
something
we
have
with
c
elegans.
A
This
is
something
we
could
work
on
with
other
things,
but
for
for
the
prod
for
the
gsoc
project,
you
know
I'd
want
to
keep
it
to
c
elegans,
but
I
like
the
the
sort
of
conceptual
stuff
that
you've
kind
of
drawn
out
yeah
and
then
I
think,
like
a
lot
of
this,
the
topological
views
are
important.
I
think
that's
important
to
keep
developing.
A
Sometimes
you
want
to
have,
like
you
know,
they're
signaling
molecules
that
are
exchanged
between
cells
that
are
close
together
or
that
have
like
certain
functions.
So
we
want
to
know
what
those
networks
might
look
like,
and
so
we
can
build
networks
of
this
organism.
You
know
that
show
these
different
properties
that
show
some
of
these
features
like
you're,
showing
with
you
know.
If
we
have
the
topological
model,
we
can
show
some
of
those
sort
of
things
that
you
wouldn't
necessarily
see
from
experimental
data.
A
Like
you
know,
you
know
pathways
of
flow
of
you
know
physical
forces
or
small
molecules,
or
things
like
that.
So
there
are
a
lot
of
ways.
Potentially
these
could
be
useful,
but
you
know
we
don't
we
don't
worry
about
that
right
now.
We
just
kind
of
we
want
to
build
something
that
will
you
know,
allow
us
to
get
there
and
then
finally,
is
a
reasonable
consider.
The
cells
are
in
euclidean
space.
A
It
could
be.
You
know
you
you
don't
it
doesn't
necessarily
have
to
be
in
euclidean
space.
You
could
have
some
sort
of
other
mapping
and
you
know
of
some
other.
There
are
other
types
of
projections
that
you
can
use
so
we're
doing:
projections
with
data
on
the
axolotl
and
the
other
group
they're,
taking
two-dimensional
images
and
they're
projecting
out
to
a
three-dimensional
image.
A
I
know
the
embryo
is
originally
three
dimensions
but
you're
making
sort
of
a
projection
outward
and
so
to
do
that
they're
doing
some
things
with
topology
as
well,
just
not
calling
it
that
and
so
yeah.
I
think
that
you
can
use
other
types
of
space
like
there.
You
know,
there's
euclidean
space,
other
types
of
like
projections
that
you
can
use
that
are
not
euclidean
that
are
curved,
and
so
that's
something
also
that
can
be
used
a
little
harder
to
work
with
for,
like
you
know,
looking
at
you
know
you
can
look
at
different.
A
You
can
look
at
the
connectivity
in
euclidean
space.
You
can
get
certain
information
about
like
a
shortest
path,
but
you
can
also
get
other
types
of
information.
You
can
look
at
curvature
and
things
like
that
which
occur
in
the
embryo.
So
you
know
that
that's
useful
as
well.
D
Yeah
yeah
yeah
yeah.
I
I
think
the
topology
is
is
our
first
question
and
if
we
consider
is
it
reasonable
to
to
make
the
series
in
the
euclidean
space
I
mean?
I
remember
there
are
some
papers
talking
about
the
sears
and
I
mean
the
embryo
and
some
other
kind
of
space
such
as
the
manifold
or
such
such
as
the
hyperbolic
space.
But
I
think
it
is
another
question
and
maybe
we
can
talk.
We
can
dis
discuss
it
later,
but
I
think
it
is
not
very
important
by
now.
D
Yes,
so
I
think
as
for
now,
I
actually
I
think,
for
now
my
I
will
focus
on
first
of
all
the
json
projects.
We
need
to
make
sure
that
we
will
have
have
an
effective
after
natural
model
which
can
be
employed
on
the
c
elegans
data
and
to
have
some
reasonable
output,
and
this.
My
second
focus
is
that
I
I
will
try
to
combine
the
topological.
D
I
mean
topology
to
create
a
biological
technologist
with
graphing
natural
again
to
see
if
we
have
any
new
findings.
Yes,
but
yes,
this,
I
will
focus
on
these
two
points
for
now,
but
actually
I
have
two
questions
here.
First,
I
think
I'm
not
sure
it
are
these
questions.
Oh
sorry,
I'm
not
sure
are
these
questions.
D
D
A
Is
the
topology
invariable?
I
I
don't
know
well
that's
a
hard
question
to
ask
because
we're
just
looking
at
like
an
average
of
many
organisms
in
the
data
that
we
get
for
cell
tracking,
or
sometimes
we're
looking
at
just
one,
and
so
knowing
like
what
the
variation
is
across
the
different
individuals
and
that
species,
or
you
know
that
that's
a
problem
that.
B
A
Probably
going
to
make
it,
you
know
not
invariable.
So,
like
you
know,
if
you
you
want
to
find
like
a
sort
of
a
platonic
tree
that,
like
you
know,
fits
every
data.
I
mean
they
do
this
with.
You
know
different
models
where
they,
like
sometimes
they'll,
do
it
with
the
human
body.
A
They'll
have
like
a
model
of
like
how
the
you
know,
the
the
shape
of
the
body,
the
the
specific
gravity
of
the
different
parts
of
the
body
like
different
tissues,
they'll
use,
like
some
you
know,
they'll
do
some
averaging
and
they'll
kind
of
come
up
with
an
average,
but
it's
not
perfect.
You
still
have
these.
You
know
you
still
have
a
lot
of
variation
around
that
average,
and
you
see
this
with
like
embryos.
You
know
you
have
in
c
elegans,
you
have
specific
mutants.
A
You
have
different,
you
know
in
c
elegans
we
work
with
it
because
it
is
more
or
less
invariable
as
as
embryos
go
for
for
different
organisms.
But,
like
you
know,
you
still
have
a
lot
of
variation
in
c
elegans.
You
have
a
lot
of
variation
between
males
and
hermaphrodites
between
different
mutants
between
you
know.
Other
just
environmental
factors
that
maybe
but
give
you
know,
introduce
a
tiny
bit
of
variation,
but
it's
you
know
not
something.
My
point
being
is
that
I
wouldn't
worry
about
like.
A
If
it's
you
know
the
perfect
model,
it's
just
kind
of
like.
I
think.
Actually,
a
better
approach
is
just
think
that
it's
like
a
generative
thing
where,
if
you
generate
a
model,
you
know
it's
like
one
possible
solution
and
then,
like
maybe
there's
like
some,
you
know
you
can
generate
different
models
that
are
similar
and
that's
like
basically
what
you're
dealing
with
for
that
species
that
you
generate
different
variations
on
it.
It's
not
too
much
different.
But
that's
that's
what
that's
what
it
looks
like
so
then.
A
The
second
question
is:
is
the
topology
of
a
cell
tree
of
a
model
organism
analogous
to
most
other
organisms?
So
that's
another
thing
again.
You
know
when
we
apply
this
to
different
organisms,
it's
gonna.
Sometimes
it's
vastly
different,
like
the
bacillaria,
which
is
a
an
algae,
it's
much
much
different
than
c
elegans.
A
So
it's
going
to
have
different.
It's
going
to
have
a
much
different
topology
as
it's
a
colony
as,
but
the
point
is,
is
that
you
should
have
like
if
it's
multicellular,
maybe
it
should
have
certain
properties,
but
that's
as
far
as
the
similarities
might
go.
If
we
look
at
a
species,
that's
closely
related
to
c
elegans,
it
might
have
very
similar
features
in
relation
to
some
of
the
anatomical
parts.
But
again
the
anatomical
parts
might
be
related.
A
There
are.
There
are
commonalities
in
development
across
different
embryos,
depending
on
where
you
are
in
the
tree
of
life,
but
those
commonalities
you
know,
should
we
should
be
able
to
pick
those
up
and
that's
kind
of
the
point
it
isn't
about.
Like
replicating
other
organisms,
it's
about
finding
things
that
are
in
association.
So
if
they're,
certain
parts
of
the
anatomy,
like
you
know,
muscle
groups
or
you
know
a
connectome,
those
should
be.
You
know
you
should
be
able
to
identify
them
with
the
method.
A
D
Yeah
yeah,
I
know
I
understand
yeah,
I
I
I
ask
these
two
questions,
because
the
first
one
is
that
I
was
discussing
that
the
topology
over
cellular
tree,
but
because
the
model
will
learn
from
a
lot
of
data
and
if
we,
if
the
the
topology
of
a
cell
linear
tree,
is
fixed.
D
That
means
when
the
model
then
from
one
scene
in
the
tree,
then
the
knowledge
from
this
cellular
tree
will
be,
I
mean,
will
be
employed
on
other
kind
of
other
other
data,
other
microscopy
video
of
this
organism,
and
they
will
help
this
model
to
track
sales,
even
though
they,
even
though
this
model
doesn't
has
doesn't
have
the
certain
imagery
of
that
microscopy.
Video
and
the
second
question
is
that's
probably
their
third
industry
will
not
be
the
totally
sand,
but
if
I
mean
just
like,
I'm
not
sure
have
you
I'm
not.
D
D
So
if
we
do
not
have
the
whole
synonym,
that
means
we
cannot
obtain
the
topology
of
this
energy.
That
means
it.
So
I
would
wonder
if
we
have
the
synonym
of
a
small
motor
organism
such
as
the
c
elegans,
if
we
can
transfer
knowledge
of
this
cell
in
a
tree
to
other
more
complex
organisms
which
have
no
cellular
tree.
Yes,
so
yes,
that
is
why
I
ask
these
two
questions
I
mean.
D
A
It's
I
think
that
yeah
I
mean
they're
gonna,
be
some
things.
You're
gonna
be
able
to
transfer
in
some
things.
You
won't
so,
like
you
know
any
kind
of
cell
division
we're
gonna
be
able
to.
If
it's
like
one
mother
cell
to
two
daughter
cells,
we
should
be
able
to
do
some.
You
know
it
should
be
able
to
considering
you
know
if
you're
like
trying
to
extract
it
from
input
data.
If
the
mode
of
division
is
basically
similar,
then
you
know
it
should
be
able
to
transfer
some
knowledge
to
that.
A
If
it's
a
different,
you
know
if
the
mode
of
division
is
different
in
in
some
ways.
Like
you
know,
in
the
in
the
image
like
visually,
I
guess
you'd
put
it,
then
you
know
it's
gonna
learn
less,
but
it
should
have
some.
There
should
be
some
applicability,
so
you
should
always
be
able
to
transfer
some
of
what
we've
learned
in
cl
against
other
species.
If
we
apply
this
method,
but
it's
it
depends
on
a
lot
of
things,
there's
a
lot
of
variation
in
in
development.
A
So
I
you
know,
I
think,
it'll
work
with
other
organisms,
I'm
not
really
sure
which
ones
it'll
struggle
with.
We
could
probably
make
guesses
as
to
that,
but
yeah.
I
think
there
will
be
some
transfer
learning
it's
going
to
really
depend
on
like
the
specific
species,
but
then
you
know
if
we
want
to
use
it
on
a
specific
species.
We
can,
you
know,
re
retrain
it
give
it
more
information,
etc.
A
A
C
Yeah
three
other
points
topology
of
a
linear
lineage
tree
is,
is
it
invariable?
There
was
a
paper,
I
think
in
the
50s
or
60s
on
cloning
frogs
and
they
found.
C
Okay,
and
this
kind
of
work
has
led
to
a
whole,
a
small
there's,
a
small
literature
on
variability
and
development,
stochastic
processes.
C
Okay,
so
there's
information
about
it.
I
don't
I
don't
know
if
that
applies
to
nematodes.
B
C
Okay,
second,
applying
the
cell
lineage
tree
to
other
organisms.
Well,
maybe
maybe
not,
but
the
there's
I'll,
give
you
an
interesting
example.
C
There's
a
cyanobacterium
that
develops
cyanobacteria
generally
consists
of
two
cell
types,
vegetative
and
heterosis,
and
they
develop
there's
a
colonial
cyanobacteria
that
forms
a
line
of
vegetative
cells
with
with
heterosystem
at
the
end
and
as
it
grows
it
forms
another.
It
moves
a
pair
of
heterosis
in
the
middle
and
then
it's
what's
between
them
into
two
colonies.
C
Okay,
so
that
kind
of
data
is
probably
available.
It's
it's
it's
an
example
which
might
be
fun
to
follow
as
a
cell
lineage
tree
and
the
last
question
on
the
euclidean
space.
So
I
was
trying
to
think
of
that
and
if
you
take
a
a
new
organism
that
has
differentiation
waves
in
it,
you
could
use
the
time
at
which
a
wave
impinges
on
a
cell
as
a
coordinate
and
that
might
that
might
be
a
non-euclidean
space.
I
don't
know:
okay,
okay,
it
would
be.
C
No
I'm
having
trouble
thinking
outside
of
euclidean
space,
so
that
would
be
a
time
plus
its
position.
So
it's
that's.
Four
coordinates.
C
B
D
C
C
A
A
D
I'm
I'm
sorry,
I
I
didn't
really
get
goddess.
Oh.
A
D
Okay,
next
few
weeks,
actually
because
it
is
august
now-
and
I'm
not
sure-
this
idea
will
be
work,
but
we
need
to
make
the
gso
project
so
actually
because
there
is
few
works
regarding
employing
german
graphene
network
on
the
cell
developmental
process.
So
my
my
plan
is
to
we
have.
We
have
only
one
implementation,
which
is
to
employ
june
to
track
sales,
and
that
was
the
paper
that
I've
mentioned,
which
was
accepted
by
the
ewcb
conference
and
I'm
not
sure,
is
it.
D
Okay
to
I
mean
to
re-implement
that
implementation
as
such
as
the
api
or
something
else
as
the
implementation
of
the
of
our
gso
project.
A
A
D
Okay,
I
would,
I
would
I'll
make
sure
that
we
will
modify
the
original
implementation
and
make
it
like
a
platform
or
framework,
and
you
would
and
people
would
find
that
they
would
easily
use
this
framework.
Yes,.
A
A
A
And
then
you
know
we
can
make,
we
can
actually
add
in
things
like
the
topological
aspect.
I
mean
there's
a
lot
there,
it's
just
kind
of
like
a
huge
area
and
a
huge
chunk
to
take
on
which
can
be
done
after
the
project.
But
you
know
we
have
to
have
us
a
nice
baseline
to
start
from,
so
we
can
add
on
to
it,
and
I
think
that
you
know
I
think,
if
you're
using
something,
that's
already
kind
of
been
developed
and
and
it
actually
works
and
then
you
can
modify
it.
A
That
would
be
a
good
way
to
go
because
you'll
have
like
at
least
some
core.
You
know
functions
and
things
like
that
that
can
be
used.
D
Yes,
yes,
yes,
I
know.
Yes,
I
understand,
but
actually
the
the
most
difficult
part
by
now
is
not
the
model
or
the
genuine
parts.
It's
about
the
computer
vision
part
it's
about
how
to
pre-process
those
microscopy
video,
it's
it's
it's
more
difficult
than
what
devlin
has
done.
Yes,
it's
not
the
simply
employed
rescue,
connect,
rest
nets
or
some
other
cnn
on
the
microscopy
video.
Yes,
we
need
to
make
sure
the
quality
of
the
outputs
of
this
cn
is
is
good,
is
satisfying.
Yes,
I
mean
where
this
is
very.
A
D
D
Yes,
yeah
actually
I've.
Actually,
the
implementation
of
constructing
a
graph
of
datasets
based
on
the
given
the
given
points
cloud
or
something
else
has
been
implemented
has
been
done.
Yes,
it's.
I
think
it
has
been
done
in
the
in
the
june
or
the
july.
I
I
remember
it's
it's
not
very
difficult.
Actually,
the
most
difficult
part
is
the
is
how
to
process
those
video
data.
D
Yes-
and
I
remember
the
water
tarot
and
long
way
they
are
working
on
these
parts
and
they
are
still
diving
into
the
details
of
the
implementation
of
that
paper
of
their
genome
for
their
checking
paper,
because
they've
used
a
lot
of
tours
and
computer
vision
tours
to
implement
such
parts,
and
I
felt
I
found
their
implementation
are
very
hard
to
describe.
They
are
very
difficult.
Yes,
they
are
very
hard
yeah
yeah
yeah,
so
we
would.
We
would
make
make
this
first
process
faster.
We
would
push
push
it
yeah.
A
Yeah
thanks
for
the
update
and
discussion
that
was
good.
Okay,
I
don't
know
if
we
had
anything
else
we
wanted
to
talk
about
today,
but
that's,
I
think,
we're
at
time
for
this
meeting
looks
like
jesse
had
to
leave
and
hurry
krishna
had
to
leave
and
we
have
some
things
in
the
chat.
Let
me
go
over
those
real,
quick,
okay.
Is
this
the
matter
of
like
people
saying
goodbye,
okay,
yeah?
So
thanks
for
attending
the
meeting-
and
I
don't
you
know
I'll-
send
out
some
information
on
some
things.
A
We've
talked
about
those
abstracts
so
that
dick
can
read
through
them
and
if
you
have
any
questions
and
you
can
catch
up
on
email
or
slack.