►
From YouTube: DevoWorm, Meeting #10: How to GSoC Proposal, Spatial Transcriptomics II, Principles of Emergence
Description
How to write a successful GSoC proposal (join the conversation on Neurostars and Slack), examples of spatial transcriptomics in mouse and C. elegans, general principles of emergence on technical networks and in bird flocks. Attendees: Susan Crawford-Young, Karan Lohaan, Harikrishna Pillai, Ishan Shanware, Gopinath Balamurugan, and Bradly Alicea.
C
C
So
yeah,
that's
great,
did
you
have
any
questions
about
you've
been
interested
in
gsoc
applications,
the
application
period.
C
A
B
C
I
don't
know
what
the
timing
is
on
the
I
always
get
that
confused
like
the
applications
open
and
then
they
close,
and
but
you
know
you
should
be
working
on
it.
You
know
all
along
and
like
I
said
I
can
help
people
with
their
applications.
If
they
give
me,
they
send
them
to
me
in
a
reasonable
state
and
then
they
get
you
know
I'll
give
them
feedback.
C
Sure
why
don't
I
go
over
that
process?
I'll
put
it
in
the
meeting
and
then
see
if
I'll
point
people
to
it
later.
So
let
me
share
my
screen
here,
all
right,
and
so
let's
see
my
board
here
so
every
year
you
know
we
have
applications
and
it
really
changes
here.
We
have
the
longer
projects
and
the
shorter
projects.
So
this
is
a
shorter
project,
so
it's
going
to
last
through
the
summer
from
I
think
like
may
to
late
august.
C
You
know
going
through
that
review
of
the
model,
so
it
would
be
like
a
review
proof
of
concept
and
it
would
like
you
know
if
you
can
get
your
hands
on
some
data.
I
know
I
shared
some
data
with
people,
but
I
can
share
more.
You
know
if
you
can
do
like
a
preliminary
analysis
that
would
be
great
to
show
that
it
works
in
past
years.
People
have
they've
done.
You
know
a
short
analysis
to
show
how
their
technique
works
on
real
data.
So
you
know
you
might
do
like
a
little.
C
C
C
And
that
is
the
you
know
this.
This
is
the
schedule,
so
this
is
almost
the
most
important
part
of
the
proposal,
and
this
is
where
you
lay
out
how
you're
going
to
do
things,
how
you're
going
to
get
them
done
and
what
time
frame.
So
you
start
off
with
week
one
and
you
go
down
a
week.
I
don't
know
what
what
the
you
know
like
12
or
13.
C
I
don't
know
how
many
weeks
they
break
it
into,
but
for
every
week
you
want
to
have
like
a
list
of
what
you're
going
to
do
and
then
you
know
you
don't
want
to
start
off.
Like
saying
I
want
to
learn
about
graph
neural
networks,
say
that's
not
really
going
to
work.
What
you
want
to
do
is
you
want
to
say
you
know,
I'm
going
to
do
that
before
the
period
begins,
then
I'm
going
to
start
with
some
reasonable
goals.
So
I
want
to
maybe
start
off
with
a
little.
You
know.
C
Maybe
I've
shown
in
this
part
here
that
I
can
I've
done
a
little
bit
of
preliminary
work.
Then
I'm
gonna
start
ramping
up
to
a
bigger
data
set
in
like
week.
Three
and
then
going
through
to
week
four
week,
five,
I'm
gonna
implement
these
parts,
I'm
gonna
get
it
working,
and
then
you
know
by
the
end,
you
should
have
the
full
project
at
least
some
version
of
it
running.
C
As
you
know,
as
something
that
you
can
execute
as
a
program
because
at
the
end
of
the
program
they
want
you
to
submit
an
executable
file
that
runs
so
they're
going
to
at
the
end
of
the
program
to
evaluate
you,
they're
going
to
take
the
program
open
it
up
and
run
it,
and
if
it
runs
and
you
you
know,
then
you're,
okay,
you're
passed.
C
So
that's
the
way
they
want
it,
and
so
we
have
to
spend
from
about
week
two
or
three
to
the
end
of
the
program,
really
kind
of
getting
everything
in
place
to
get
to
that
end
goal.
And
so
you
know
that
could
be.
There
might
be
some
obstacles
in
your
way.
You
might
have
things
that
don't
work
out
the
way
you
want
them
to
so
you
want
to
have
in
this
schedule.
I
always
stress
this.
C
I
remember
one
year
we
had
you
know
someone
had
an
issue
with
their
ability
to
find
a
powerful
enough
computer
to
do
what
they
wanted
to
do
and
while
as
they
searched
for
that,
they
were
able
to
do
some
other
things
to
put
them
in
place.
C
Sometimes
people
propose,
like
you,
know,
starting
off
with
just
getting
everything
in
place
and
then
like
in
the
second
part
of
the
program.
Optimizing
it
or
you
might
walk
through
it.
Where
you
know
you
get
data,
you
analyze
some
data,
you
go
back
and
get
more
data
analyze.
That
depends
on
the
algorithm
you're
trying
to
use
and,
like
you
know,
what
gets
you
to
that
end
goal
fastest
or
what
gets
you
that
end
goal,
maybe
without
so
many
obstacles
but
keep
in
mind.
You'll
still
have
obstacles,
and
you
want
to
plan
for
that.
C
C
C
We
have
a
movement
group,
so
people
who
do
see
elegance
movement
and
that
actually
has
a
lot
of
machine
learning
in
it
and
they're
doing
a
lot
of
stuff
with
vision,
computer
vision-
and
so
you
know
that's
somewhere,
you
might
reach
out
and
ask
if
they
have
some
interesting
data
that
you
might
use.
You
know
maybe
not
for
the
main
project,
but
if
it's
something
you
know
that
might
help
you
with
your
validation.
It
might
be
worthwhile
reaching
out
there's
a
movement
database
which
has
been
sporadically
online.
C
So
that's
something
that
you
can
use.
There
are
other
groups
in
the
open
norm,
foundation.
There's
c302,
where
they
do
simulations
of
neurons,
they
do
other
types
of
simulations,
there's
gepetto
and
some
of
the
other
programs.
These
are
things
that,
during
the
community
period
we
would
go
over
and
you
know
you
would
be
able
to
reach
out
to
these
people
any
time,
but
that's
the
time
when
we
want
to
get
like
you're
familiar
with
the
larger
community.
C
So
I
think
that's
a
valuable
thing.
You
know
you're
going
to
put
that
in
your
schedule,
it'll
say
like
community
period
and
then
from
week
three
on
you'll
be
doing
your
project
in
steps
you'll
be
proposing.
You
know
how
you're
going
to
implement
algorithms,
how
you're
going
to
put
together.
You
know
some
of
your
resources
to
make
sure
you
have
everything
in
place.
C
I
know.
In
past
years
people
have
used
the
collab
notebooks
as
a
way
to
get
around
some
of
their
computational
limitations,
because
the
collab
notebooks
allow
you
to
they
give
you
free
computational
resources
through,
I
think
google's
cloud
hub.
So
that's
that's
a
resource.
C
I
don't
think
a
lot
of
people
think
about,
but
I've
I've
had
one
or
two
students
have
had
problems
with
that.
In
the
past,
like
getting
a
good,
you
know
they
have
their
own
computer
and
google
gives
you
money
for
a
computer,
but,
like
you
can
also
you
know,
with
a
lot
of
the
machine
learning
stuff
a
lot
of
the
deep
learning
projects.
C
C
C
You
know
you
want
to
make
sure
it's
not
just
a
loose
idea.
You
want
to
kind
of
write
it
up
as
a
short,
not
a
pre-proposal,
but
just
kind
of
an
outline
with
some
things
in
it.
You
know,
maybe
you
know,
try
to
run
your
algorithm
beforehand
and
see
if
it's
a
viable
idea
and
then
I
can
tell
I
can
give
you
feedback
on
it.
I
can
tell
you
whether
this
is
good.
What
what
you
need
to
change
yeah.
B
C
Okay,
yeah-
and
this
goes
for
both
proposals,
so
if
people
are
doing
the
axolotl
project
or
they're
doing
the
craft
neural
networks
project,
it's
the
same
thing,
you
know
to
kind
of
go
through
and
show
up
sort
of
a
proof
of
concept.
C
You
know,
then
have
a
schedule
and
then
at
the
end,
I
think
you
also
have
to
talk
about
yourself
a
little
bit,
but
that's
that's
something
yeah!
It's
like
a
cv.
I
think
you
have
to
give
like
your.
You
have
to
upload
your
credentials
or
something
but
like
there's
a
small
space
for
that
as
well.
C
D
There
we
go.
Oh
my
I'm,
I'm
fine,
dr
sharif
just
dumped
my
project
on
my
head.
D
No,
he
just
it's
the
way
he
works,
so
I
have
to
have
my
whole
thing
done
by
may,
27th
yeah.
C
D
So
he
says
he
says
just
just
hack
it.
He
wants
me
to
to
grab
whatever
I
can
in
the
way
of
pre-programmed
stuff
and
then
just
throw
it,
throw
it
into
the
program
and
get
it
to
work.
Okay,
yeah
right,
I
do
have
a
new
stand
for
the
microscope
and
worse
comes
to
worst.
D
I
could
always
get
a
peppercorn
or
something
like
that
and
image
it
instead
of
the
salamander
egg,
because
I
do
have
some
salamander
egg
stuff
and
I
could
what
I
did
before
was
put
markings
on
a
peppercorn
and
then
so
they're
different
colors
on
different
sides,
that's
harder
to
do
than
you
think
yeah,
but
I
could
try
something
like
that
too,
as
so,
you
could
do
like
do
both
because
the
peppercorns
around
about
three
millimeters
so
buying
a
small
peppercorn.
C
That
you
did
the
you
did
the
biosystems
paper
where
you
had
the
different
samples:
okay,.
B
Which?
One
in
which
are
yours?.
D
Okay,
do
you
have
a
copy
of
it.
D
Okay,
all
right
good,
well,
there's
a
peppercorn
in
there
and
I
could
take
try
to
get
some
more
images
of
of
it
or
something
similar.
D
Yeah
with
the
peppercorn,
it
wouldn't
be
time
series,
but
at
least
it
would
be.
You
could
get
a
whole
image,
then
we're
all
all
ten
images
of
of
the
sides
of
it
and
I'm
still
hoping
that
my
salamanders
might
decide
to
lay
eggs
sometime.
D
I
might
ask
her
to
do
something
strange
with
her
salamanders
after
that,
but
yeah.
C
D
D
C
So
yeah,
you
guys
missed
that.
I
just
went
over
some
of
the
gsoc
how
to
write
a
proposal
and
why
don't
I
go
over
that
really
quickly
again,
so
I
I
was
explaining
that
the
way
I
usually
like
to
have
the
or
the
way
proposals
have
been
written
in
the
past
that
have
been
successful.
C
Is
that
you've
had
proposals
where
you
have
like
a
page
of
introduction
to
your
problem.
So
you
know
if
you're,
using
a
technique
or
if
you're,
like
you,
know
the
project.
What
is
it
about
the
project
that?
Why
is
it
important-
and
this
is
just
to
show
that
the
student
understands
the?
Why?
Why
they're
doing
this,
then
there's
about
a
page
and
a
half
or
two
or
maybe
even
more
than
that
of
a
proof
of
concept
or
review.
C
C
C
You
can
get
an
idea
of
what
they're
doing
there
are
other
groups
like,
such
as
the
movement
validation
group,
where
they're
doing
they're
looking
at
images
of
c
elegans,
adult
c
elegans
that
are
crawling
around
in
a
dish
and
there's
trying
to
segment
those
worms
and
trying
to
figure
out
the
movement.
A
C
So
that's
something
that
you
know
might
be
useful
to
you
in
a
project
sort
of
as
a
validation,
step
or
you
know,
might
be
interesting
for
future
work.
C
So
that's
that's
the
community
period
and
then,
from
week
three
on
you
have
your
work
period,
which
is
where
you
actually
implement
your
work,
and
you
want
to
be
careful
because
sometimes
you'll
have
things
that
you'll
run
into
obstacles,
and
so
you
want
to
have
some
sort
of
plan
to
get
around
those
obstacles
to
do
other
things,
while
you're
trying
to
solve
those
obstacles
and
so
forth.
C
C
I
think
it's
at
the
end
of
august
you're
gonna
submit
your
project
and
you're
gonna
have
to
submit
an
executable
and
that
executable
has
to
run
in
order
for
you
to
pass.
So
that's
that's.
Your
end
goal
is
an
executable
program,
and
so
this
entire
period
has
to
be
geared
towards
making
that
hap
work
so
that
that's
that's
the
the
flow
and,
like
I
said
before,
I
am
happy
to
review
people's
proposals
as
they
write
them
up.
C
You
know,
get
it
into
decent
shape,
have
like
a
nice
flow
of
your
proposal,
and
then
I
can
give
you
feedback
and
to
make
sure
that
you're
on
the
right
track.
Any
questions
about
that.
E
No,
that
that
makes
sense
actually
yeah.
There
should
be
a
proper.
You
know,
format
of
any
proposal,
so
I'll,
just
staff.
My
work
up
by
this
end
of
this
near
the
weekend
and
I'll
send
a
draft
proposal
up
to
you.
C
Yeah
yeah
that
sounds
good,
so
yeah,
I
don't
know
what
the
timeline
is.
I
think
the
proposals
open.
I
don't
know
when
you'll
have
to
look
on
the.
E
Yeah
about
for
april
14th,
so
there's
about
a
month
from
now.
Okay,
so
that's
the
deadline
for
the
proposals.
Okay,.
A
C
Then,
of
course,
you
have
to
register
with
the
website
the
google
summer
of
code
website,
then
we're
also
on
neurostars,
which
is
this
site
neurostars.org.
Some
of
you
have
been
here.
C
Some
people
haven't,
but
because
there
are
multiple
ways
to
get
involved
in,
you
know
to
sort
of
get
into
the
project,
but
if
you're
not
familiar
with
the
neurostars
forum,
this
is
where
we
post
the
projects
we're
under
incf
and
we're
project
series
22.,
so
gnns
is
22.1
and
axolotl
is
22.2,
so
we
have
a
description
of
the
project,
some
links
to
some
resources
here
and
then
some
skills
and
requirements,
and
then
we've
had
some
questions
about
the
project
in
the
in
the
chat.
C
A
C
Is
the
digital
microsphere,
which
is
what
the
axolotl
project
is
and
here
these
papers
that
susan
was
talking
about
this
paper?
I
think
here
the
2018
paper.
I
think
susan
had
a
2020
paper
or
2021
paper.
C
So
maybe
I
can
get
that
up
there
and
it's
just
the
the
citation
so,
but
we
can
you
know
this
is
the
place
to
be.
If
you
want
to
learn
more
about
the
project,
I
don't
think
I've
had
any
comments
on
it.
C
So
we'll
try
to
go
through
narrow
stars
and
we'll
try
to
go
through
the
slack
for
people.
C
C
C
C
So
why
don't?
I
share
my
screen.
I
have
some
things
more
things
on
spatial
transcription
and
and
protein
atlases
and
things
like
that.
So
one
of
the
things
I
found
this
week
was
and
last
week
I
think
I
talked
about
this,
but
I
talked
about
it
in
terms
of
just
like
heat
maps,
which
are
these
maps
where
you
have
like
a
list
of
sites
or
lilik
sites
and
then
a
list
of
like
samples.
C
So
you
have
like
a
tissue
sample,
you
get
the
dna
out
of
it
or
the
the
rna
or
the
protein,
and
you
do
some
sort
of
chemical
analysis
and
then
you
plot
out,
like
you
know,
expression,
level
or
some.
You
know
some
sequence
and
then
you
can
build.
You
can
build
a
heat
map
from
like
intensity
of
of
expression.
You
can
build
a
heat
map
out
of
like
quantitative
expression,
and
then
you
can
also
have
a
sequence
analysis
that
aligns
it
to
certain
sites.
C
C
We
we
have
spatial
transcription,
transcriptional
analysis,
which
is
where
you
try
to
link
gene
expression
or
some
transcription
pro
transcriptional
profile
to
certain
locations
in
an
organism
or
in
a
in
an
embryo.
So
in
this
paper
this
is
a
4d
single
cell
protein
atlas
of
transcription
factors,
delineate
spatial
temp,
spatiotemporal
patterning
during
embryogenesis.
C
So
this
is,
I
think,
in
c
elegans,
where
they
do
this
so
yeah.
This
is
done
in
c
elegans
embryogenesis,
they
use
protein
fusion
fluorescent
reporters
and
four
dimensional
live
imaging,
so
in
this
case
they're
using
fluorescent
imaging.
So
actually,
what
I
explained
to
you
before
is
one
way
to
do
this.
C
Another
way
to
do
this
is
to
take
a
microscope
and
use
fluorescent
proteins,
which
you
have
these
fluorescent
elements
that
are
usually
some
protein
from
like
a
a
marine
invertebrate
that
they
use
and
they're
different,
like
they
fluoresce
in
different
frequencies
of
light.
So
gfp
is
green
fluorescent
protein.
It
fluoresces
is
green
when
it's
hit
by
ultraviolet
white
there's
red
fluorescent
protein.
C
It'll
be
in
one
place
and
not
another
place
or
it'll,
be
at
a
certain
intensity
in
one
place
and
higher
intensity
in
another
place,
and
then
it's
a
matter
of
image
processing
to
sort
of
figure
out
what
the
intensities
are.
So
you,
you
know
this
is
a
nice
method,
for
you
don't
have
to
destroy
the
sample.
You
have
to
well
you
kind
of
have
to
destroy
the
sample,
but
you
don't
have
to
you
get
like
spatial
information.
C
C
This
is
an
atlas
they're
building
so
they're
building
this
atlas
at
single
cell
resolution,
so
they're
able
to
look
at
each
cell
and
how
these
proteins
are
expressed
in
each
cell
and
then
from
there.
They
can
build
an
atlas
of
each
cell
and
its
properties
within
the
cell.
C
So,
let's
see
if
I
can
find
some
images,
so
here's
a
developmental
protein
expression,
atlas.
C
This
will
give
me
a
bigger
picture
here.
Okay,
so
this
is
where
they
had.
They
identified
609
transcription
factors
in
the
c
elegans
genome
290
transcription
factors
are
selected
as
targets,
so
all
the
transcription
factors
that
they
identified
here
in
this
first
part
of
the
screen.
They
screened
it
for
these
targets
and
then
they
found
291
what
they
call
protein
fusion
reporters
and
they
bound
them
to
266
of
these
transcription
factors.
C
So
this
is
their
they
they
usually
do
a
pipeline.
They
usually
do
like
an
initial
survey
of
what
they
want
to
use
as
their
sort
of
target.
Then
they
do
some
screens
to
see
if
they're
viable
and
then
they
build
the
construct
and
they
put
it
into
the
cells
and
then
they're
able
to
get
these
kind
of
images
where
you
have
this
entire
embryo.
This
is
a
c
elegans
embryo,
starting
from
very
few
cells
and
then
increasing
in
the
number
of
cells
to
this
stage.
C
Where
you
have
this
folded,
this
is
sort
of
like
where
they
call
this
the
comma
stage,
where
the
this
ball
of
cells
starts
to
fold,
and
then
you
get
this
where
there's
folding
over,
and
this
is
going
to
form
the
worm.
I
don't
know
which
end
is
anterior
posterior.
I
think
this
is
the
anterior
under
the
head,
and
this
is
the
posterior
under
the
tail
and
it'll
bend
around
a
couple
times
by
the
time
the
egg
hatches.
C
C
Some
of
these
green
signals-
and
so
you
get
this-
you
know
their
problems
with
things
like
auto
fluorescence
and
noise,
but
it's
a
nice,
a
nice
method
for
sort
of
getting
at
this
in
in
what
we
call,
maybe
call
real
time
or
real
space,
and
so
this
is
how
they
do
this.
They
they
can
trace
lineages
using
this
method.
C
And
then
you
know
this
shows
you
this
tracing
the
protein
expression
out
in
a
cell
lineage.
So
when
you
do
this
kind
of
thing,
where
you
label
it
twice
for
different
purposes,
you
can
actually
build
something
like
this,
which
is
a
lineage
tree.
You
have
these.
You
have
the
green
branches,
which
are
where
that
protein
is
being
expressed.
You
have
the
branches
that
don't
have
a
color
which
are
branches
that
exist,
but
they
don't
have
that
ex
expressed
protein
in
it,
and
you
can
see
then,
at
the
end
in
c
elegans.
C
A
C
These
cells
are
going
to
become
pharynx
and
so
forth,
and
so
we
don't
have
to
guess
you
know
we
don't
have
to
do
any
sort
of
going
back
to
the
adult
and
checking
because
they're
all
basically
going
to
become
that
kind
of
cell
in
like
say
like
humans
or
mice,
there's
a
lot
of
signaling
that
determines
what
the
fate
of
a
cell
is
going
to
be.
So
it's
different
in
that
way.
C
This
is
where
they're
expressing
different
things.
You
have
these
different
colors
that
they've
put
on
here
and
then
for
these
different
factors,
transcription
factors,
and
then
this
is
a
lineage
tree
where
you
can
see
those
these
things
here
being
expressed.
So
you
can
go
between
the
spatial
map
and
this
lineage
tree,
so
that's
how
one
one
example
of
how
they
do
spatial
transcription
another
way
is
to
actually
get
some
of
these
examples
here.
These
are
different:
next-gen
sequencing
techniques
and
some
other
kinds
of
techniques
that
they're
doing
here.
C
C
C
So
this
is
the
data
set
they're
presenting
here
and
see
by
positioning
histological
sections,
which
are
sections
of
a
of
a
tissue
on
a
raid,
reverse
transcription
primers
with
unique
positional
barcodes.
It's
just
the
method
they
use.
This
study
demonstrated
high
quality
rna
sequencing
data
with
maintained,
two-dimensional
position,
information
for
the
mouse
brain
and
human
breast
cancer.
So
they're
able
to
look
at
like
different
sources
of
tissues.
C
C
So
that's
human
and
mouse
and
they're
looking
at
a
subset
of
genes
in
the
mouse
and
human,
and
then
you
know
that
so
that's
how
they
do
this,
and
so
this
is
an
example
of
the
muscle
factory
bulb.
So
this
is
a
juvenile
muscle
factory
bulb.
You
can
see
that
they
have
this.
This
is
a
section
of
the
tissue,
so
they've
cut
through
the
tissue
at
a
certain
like
orientation
and
they've
taken
a
slice
of
it,
and
then
they
prepare
the
slice
and
they
do
this
analysis.
C
C
So
you
what
you
can
do
with
this
is
actually
kind
of
interesting.
You
can
run
something
they
call
spatial
statistics
on
it,
which
is
where
you
take
a
bunch
of
points,
and
you
figure
out
the
distance
from
say
like
some
focal
point
and
you
can
find
trends
in
the
data
you
can
find
like
the
significance
of
like
spatial
distributions
using
the
spatial
statistics
techniques
and
I'm
not
going
to
get
into
those
here.
But
those
are
some
techniques
that
people
often
aren't
familiar
with,
because
they're
not
used
everywhere.
C
They're
usually
used
in
like
geography
and
in
earth
science.
But
but
those
are
those
methods
exist.
Another
thing
you
can
do
is,
of
course,
signal
processing
or
some
other
type
of
frequency
analysis,
and
so
that's
something
that
you
know
is
also
useful.
Tells
you
different
things
about
like
the
periodicity
of
some
of
these
expression
patterns
and
other
things,
so
this
is
an
example
of
a
mouse.
C
A
olfactory
bulb
like
a
slice
through
it,
and
this
shows
you
what
you
can
do
with
this
kind
of
data,
and
this
is
an
intensity
map,
so
this
is
normalized
expression
level.
Each
of
these
points
show,
like
you,
can
see
the
color
so
at
the
edges.
Roughly
of
so
like
this,
this
bulb
is
split
into
two
parts.
At.
C
C
C
To
overcome
this
representational
problem,
we
propose
punk
array
maps,
a
method
that
harness
the
power
of
hyperbolic
geometry
into
the
realm
of
single
cell
data
analysis.
So
hyperbolic
geometry
is
another
way
to
project
things:
euclidean
geometries,
where
we
had
that
x
y
map.
So
it
was
a
flat
map.
It
had
a
flat
surface.
There
was
no
reference
for
curvature
at
all,
but
this
actually
gives
you
the
ability
to
incorporate
curvature
in
your
model,
and
so
I
often
understood
as
a
continuous
extension
of
trees.
C
Hyperbola
geometry
enables
the
embedding
of
complex
hierarchical
data
in
only
two
dimensions,
while
preserving
the
pairwise
distances
between
the
points
in
the
hierarchy.
So
it
allows
you
to
look
at
like
curvatures
and
preserve
the
distance
between
points,
because
when
you
have
a
curvature,
sometimes
you
know
you,
don't
the
the
points
don't
translate.
You
can
have.
You
know
distance
between
two
xy
points
and
there's
a
there's,
a
euclidean
distance
and
that's
easy
to
understand.
C
C
This
enables
the
use
of
our
embeddings
in
a
wide
variety
of
downstream
data
analysis
tasks,
so
they're
doing
a
lot
of
things
afterwards
that
are
going
to
be.
This
is
going
to
be
useful
in
so
this
is
a
single
type
of
embedding.
So
we
talk
about
embeddings
and
graph
neural
networks.
C
So
let
me
go
to
see
if
there
are
any
images
here
that
we
can
really
dig
into
here.
This
is
a
good
example
here
of
how
plunkery
maps
discover
hierarchies
and
branching
processes,
so
this
is
what
they
call
waddington's:
landscape,
epigenetic
landscape.
This
is
a
single
cell
or
a
single
thing
in
development,
and
it's
going
through
these
transition
points
where
you
start
off
with
a
single
cell
and
you
know,
divides
into
two
cells
and
then
those
two
cells
divide
into
four
cells
and
they
take
this
trajectory
their
developmental
space.
C
So
you
end
up
with
four
cells
that
have
sort
of
different
developmental
trajectories
from
a
single
cell,
and
so
this
is
just
kind
of
what
they're
getting
at
here.
These
trajectories,
then,
can
be
mapped
from
this
conceptual
landscape
to
a
hyperbolic
space
which
is
actually
metric
space
that
you
can
measure
distances
on
or
measure
distances
between,
like
expression
patterns,
because
these
these
these
channels
down
this
landscape,
are
only
hypothetical.
C
C
This
is
an
example
of
euclidean
space
where
you
have
these
pairwise
distances
between
these
cells,
and
it
shows
that
you
know
you
have
this
space.
That's
linear!
It's
not
going
to
preserve
distances
for
multiple.
You
know
like
for
spatial
relationships
and
distances
between,
say
like
if
you
have
a
cell
cell
comparison
map,
and
you
have
differences
in
expression,
and
you
have
some
distance,
a
distance
metric
of
of
difference
versus
a
spatial
distance
metric.
C
Those
two
things
don't
map
to
weld
one
another,
but
you
can
preserve
these
pairwise
distances
by
using
a
punk
array
map,
and
so
it's
a
circle
here
and
you
have
these
cells
that
are
at
these
locations
in
the
circle,
and
so
this
is
the
way
they
do
this.
I
want.
I
know,
there's
another
image
in
here.
That
shows
you
kind
of
what
this
looks
like
for
different
samples.
C
So
this
is
where
they
have
embedding
an
embedding
quality
metric
that
they
use,
and
so
this
is
an
example
of
sort
of
fit,
I
guess,
and
so
for
different
species.
Different
samples,
these
these
different
embeddings
punker,
a
maps
are
in
blue
versus
some
of
the
other
ones
that
they
use
in
genomics.
They
use
a
lot
of
these
type
of
embeddings
to
represent
the
data
and
they
just
try
to
find
the
best
one,
so
they
have
umap
t-sne.
C
These
are
dimensionality
reduction
techniques,
you're,
probably
familiar
with
pca,
of
course,
which
is
principal
component
analysis
and
some
other
ones
that
they
use.
That,
I
think,
are
more
distinct
with
respect
to
genomics,
but
they
basically
are
comparing
these,
and
so
you
see
the
punk
array.
Maps
usually
consistently
end
up
on
top,
so
they
end
up
at
the
top.
This
good
end
of
the
y-axis.
C
C
We
see
that,
like
we
have
some
things
that
we
can
identify:
different
cell
tissue
types
of
body,
wall,
muscle,
coliomyosite,
germline
z1
through
z4,
which
are,
I
believe,
germ
cells,
ciliated,
amphit,
neuron,
ciliated,
non-amphit
neurons.
So
you
have
all
these
different
cells
or
tissue
types
that
are
put
onto
this
embedding
and
they
have
this
these
these
distances
between
each
other.
So
you
can
see
that
there's
a
distance
between
ciliated,
non-amphineuron
and
body
wall
muscle,
and
you
can
see
that
the
distance
is
here.
C
This
is,
I
guess,
an
expression
data,
so
this
is
a
little
bit
different
than
you
know
the
the
type
of
thing
we're
looking
at
with
lineage
trees
and
it's
a
different
way
of
thinking
about
it
because
you
know,
like
you
know,
if
you
are
familiar
with
t-sne
and
umap,
you
know
that
those
are
very
hard
to
interpret.
I
think
we've
talked
about
this
before
in
meetings
where
you
have
this
multi-dimensional
data
set,
and
you
boil
it
down
to
like
two
or
three
dimensions,
and
then
you
build
this
map
and
it
looks
really
pretty.
C
This
is
actually
kind
of
looks
like
a
umap
representation,
but
you
can't
really
interpret
it,
and
so
it
is
with
this
method
as
well,
it's
kind
of
hard
to
interpret,
but
we
can
see
that
actually
we
know
the
distances
are.
But
beyond
that
we
don't
really
have
any
good
interpretive
theory
of
why
these
things
are.
You
know
far
apart
or
close
together,
except
to
say
that
those
things
cluster
together
in
this
analysis,
so
we
know
that
they're
similar
other
than
that.
C
We
don't
really
know
that
much,
and
so
this
shows
you
again
with
embryo
age,
where
these
cells
are
coming
up
and
where
what
they
are
so
here's
the
germline
and
then
you
see
other
cells
coming
up
later
in
development
to
650
minutes,
and
so
these
cells
from
this
map
are,
you
know,
differentiated
in
time,
dermaline
cells
being
first
and
then
going
on
to
these
cells,
which
are
the
last
ones
to
differentiate.
C
In
the
same
embedding,
so
this
is
again
pseudo
time.
I
guess
is
some
sort
of
like
time
measurement,
it's
not
like
minutes,
but
it's
maybe
order,
and
so
you
know
one
can
do
these
kind
of
maps
for
different
types
of
you
know
different
criterion.
So
cell
types,
so
you
know
if
we
classify
the
cell
c
elegans,
we
know
what
the.
If
we
have
a
lineage
tree.
C
We
know
what
the
adult
cell's
going
to
be,
so
we
can
actually
classify
things
in
development
as
the
adult
cell
pseudo
time
is
like
when
these
things
appear
and
then
the
edge
of
the
embryo
is
like
the
age
of
the
whole
embryo
versus
when
that
cell
appears,
and
so
that
you
can
do
a
lot
of
things
like
that
and
build
these
kind
of
maps
again.
They're
very
beautiful
and
they're
somewhat
informative,
although
not
completely
informative,
and
so
that's
that's
kind
of
the
idea
there.
Okay
ishaan
had
to
go.
C
Thank
you
so
yeah
any
questions
about
that.
I
know
that
was
a
lot.
I
know
quran
had
started
this
conversation
about
adding
in
transcriptomic
data.
F
Yeah
yeah
yeah,
especially
gene
expression,
related
data,
even
the
previous
people,
I
think,
had
a
lot
of
insight
related
to
possible
features
that
could
be
added.
I'm
thinking,
maybe
you
know
how
about
adding
another
like
a
bit
more
general
model.
You
know
for
even
c
elegans.
I
don't
know
if,
if
that
will
be
useful
for
c
elegans
as
well,
and
I
think
I'll
have
to
look
more
into
it,
but
I
think
it
could
solve
both
of
these
issues.
C
D
Do
you
know
what
the
mechanical
properties
of
the
nucleus
are
in
in
most
cells?
Is
it
softer
than
the
surrounding
cytoplasm,
or
is
it
harder
or
more
elastic,
or
do
you
have
any
clues
as
to
that.
C
I
don't
I
don't
know
if
there
are
any
studies
that
have
characterized
that
I
think
like,
though,
because
there's
a
lot
of
dna
in
the
nucleus.
It
might
be
harder,
but
I
don't
know
because
there's
a
lot
of
stuff
in
the
in
the
cytoplasm
too
floating
around.
So
it's
hard
to
know,
I
mean
the
cytoplasm
is
likely
to
be
an
isotropic
like
you
know,
there's
cytoplasm,
which
is
the
the
fluid,
but
it's
like
there's
stuff
in
there,
and
so
any
measurements
gonna
have
to
take
that
into
account.
C
D
Yeah
it's
packed,
it
might
match
it
kind
of.
I
was
thinking
because,
because
it's
all
one
cell,
it
probably
has
some
sort
of
matching,
like
its
elasticity,
will
probably
match
the
elasticity
of
the
cell
anyways.
If
you
don't
know
of
any
studies
on
that,
no.
C
Well,
I'm
not
familiar,
I
mean
there
might
be
some
biochemistry
like
in
the
biochemistry
literature
there
might
be
something
because
I
know
that
sounds
like
a
very
biochemistry
oriented
question
like
what
is
the.
C
Density,
it's
like
in
a
textbook,
you
see,
like
the
you,
have
the
cytoplasm
and
it
looks
almost
empty
which
is
wrong,
but
and
then
the
the
nucleus,
which
is
the
shell
and
it
has
like
the
dna
in
it,
but
there's
a
lot
of
space
in
there
and
then
you
know
the
cytoplasm
is
a
liquid.
But
it's
like
you
know,
there's
it's
it's
it's.
I
guess
a
thick
density.
I
don't
know
what
the
I
don't
know
what
the
viscosity
is
of
it,
but.
A
C
You
have
a
lot
of
things
in
there,
so
it's
like
gonna
be
very
hard
to
I
mean
you
know
it.
I
would
imagine
yeah
yeah
imagine
it
would
be
highly
an
isotropic
and
it
would
be
yeah
actually
kind
of
hard
to
measure
because
you
have
to
get
a
single
cell
and
you
had
to
have
to
get
like
isolate
the
nucleus
and
like.
D
G
C
Yeah
so
back
to
koran's
point
yeah,
I
think
like
for
c
elegans,
you
know
that's
a
good
the
good
system
to
work
in,
because
there's
not
only
it's
easy
to
work
with
in
terms
of
lineage
tree,
but
there's
also
a
lot
of
data
available
like
secondary
data,
and
then
people
have
done
studies
like
the
one.
I
showed
you
the
first
one
I
showed
you
where
they
did.
You
know
they
were
able
to
do
everything
in
one
shot.
C
C
You
have
like
each
cell
has
an
identity
and
it
comes
up
at
about
the
same
time
and
so
that's
a
place,
maybe
where
you
can
do
a
lot
of
that
sort
of
thing
like
combining
the
data
sets
and
things
if
you're
trying
to
do
that
and
it'd
be
very
much
harder
to
do
an
axolotl
because
we
have
like
you
know
there,
there's
variability
in
how
the
cells
are
differentiating
in
time.
Just
enough
that,
like
it's,
it
would
be
hard
to
do
that
to
build
something
that
was
really
consistent,
so
yeah,
okay,
okay,
yeah.
B
C
I
go
I'm
going
to
go
over
some
papers
in
a
bit,
but
I
wanted
to
first
go
over
the.
We
have
the
major
task
board
and
I
bring
this
out
every
so
often
just
to
kind
of
give
an
idea
of
where
we
are
on
it.
A
lot
of
action
items
that
we
need
to
follow
up
on.
I
think
some
of
these
have
been
done.
I
think
this
one
is
done
this
one
is,
you
know
like.
C
Sometimes
we
come
up
with
things
in
the
meeting
and
then
they
drop,
and
then
I
come
up
with
something
else,
and
then
you
know
like
sometimes
they
get
addressed.
Sometimes
they
don't
so
this
is
actually
on
our
diva
worm,
github
its
group
meetings
projects.
So
this
is
where
we
keep
a
lot
of
this
accounting.
C
We
have
things
like
the
1d
ising
model.
I
don't
know
if
that
ever
went
anywhere.
I
think
people
are
still
kind
of
thinking
about
it.
We
have
the
this.
This
hyper
graphs
submission
was
something
that
I
showed
last
week
and
that's
the
submission
is
finished,
but
now
there's
some
more
work
to
do
on
that
apply
to
gsoc
as
our
own
organization.
So,
unfortunately
we
didn't
get
selected
as
an
organization
for
that,
but
we're
still
participating
through
incf.
C
We
actually
applied
as
our
own
organization
this
year,
but
that
didn't
happen.
That's
fine!
It
was
we're
just
trying
to
maybe
get
more
slots.
Some
of
these
things
are
really
old.
I
would
go
I'm
going
to
go
over
this
later,
but
I
know
that
like
jesse's
sometimes
works
on
this,
I
work
on
this.
Sometimes
so,
just
just
so
people
know
there
are
a
lot
of
kind
of
open
issues.
C
So,
if
you're
looking
to
contribute
to
the
community
here
we
have
a
lot
of
open
issues,
a
lot
of
things
to
follow
up
on
just
mention
some
issue.
You
know
by
number-
and
you
know,
if
you're
interested
and
can
kind
of
give
you
an
idea
of
where
we
are
on
it.
That's
the
best
way
to
kind
of
keep
track
of
all
these
moving
parts
that
we've
had
so
now
I'm
going
to
go
to
papers
here,
and
so,
let's
see
what
do
we
want
to
talk
about
this
week?.
D
Today,
before
you
start,
I'm
participating
in
the
aps,
physics
march
meeting
okay
and
there
there
are
quite
a
few
people
who
do
networks
of
various
sorts
in
that
meeting.
But
there's
soft
matter
physics
in
in
the
meeting
all
over
the
place
with
cytoskeleton.
B
D
C
Yeah
yeah,
that
would
be
great,
and
then
we
have
that
special
issue
that
we're
working
on.
Maybe
we
could
find
some
people
there
who
might
want
to
contribute
to
that.
C
C
Okay,
yeah
well
gopinath
welcome.
A
C
So
let
me
go
back
to
this.
What
is
this
one?
Okay,
yeah?
Why
don't
we
talk
about
this?
So
this
is
a
kind
of
call
life
as
emergent
phenomena,
and
this
is
about
how
life
is
sort
of.
C
We
think
we
talk
about
like
how
things
are
emergent
when
we
talk
about
whole
organisms,
as
opposed
to
like
gene
expression,
we're
talking
about
like
biophysics
measurements
of
cells.
We
talk.
C
I
see-
maybe
I'm
not
sharing
my
screen,
I'm
not.
Okay.
C
There,
we
go
okay,
so
when
we
talk
about
when
we
saw
that
embryo
with
all
the
cells
and
how
it's
the
cells
are
dividing
and
forming
shapes,
this
is
something
that
people
have
argued
is
what
they
call
an
emergent
phenomenon.
C
That
happens
where
things
sort
of
spontaneously
organize-
and
you
know
it
sounds
like
magic,
but
people
have
debated
this
for
many
years
and
so
people
have
tried
to
study
in
different
ways
and
it's
kind
of
hard
to
study,
because
you
know
how
do
you
study,
you
know
a
bunch
of
interactions
simultaneously
and
how
do
you
get
to
the
structure
of
those
things?
So
sometimes
people
use
tools
like
network
theory,
which
is
something
we've
talked
about.
People
use
different
types
of
you
know
scaling
laws
to
explain
like
you
know.
C
C
Studies
from
a
large
scale
void
simulation
and
web
data.
So
in
our
group
we
talk
a
lot
about,
like
maybe
the
emergence
of
life
or
the
emergence
of
embryos
or
the
emergence
of
biochemicals,
and
so
you
know
there
are
different
rules
there,
but
a
lot
of
people
think
that
emergent
phenomena
have
rules
that
transcend
any
one
system.
So
if
you
look
at
multiple
systems
and
compare
them,
you'll
see
commonalities
in
them
and
what
they
do
in
this
paper
is.
C
They
have
a
large
scale,
void,
simulation
and
web
data,
so
they're
using
two
separate
things
and
they
have
nothing
to
do
with
biology,
but
this
is
something
that
you
know
is
also
classified
as
emergent
phenomena.
So
this
is
a
japanese
group,
and
so
a
large
group
with
a
special
structure
can
become
the
mother
of
emergence.
C
We
discussed
this
hypothesis
in
relation
to
large-scale
void
simulations
and
web
data
in
the
boyd
swarm
simulations.
The
nucleation
organization
and
collapse
dynamics
were
found
to
be
more
diverse
in
larger
flocks
than
in
smaller
flocks.
When
they
talk
about
voids,
they
mean
this
type
of
simulation,
and
I
don't
know
if
people
are
familiar
with
this
computer
science.
C
People
might
be
familiar
with
this,
but
it's
a
simulation
of
basically
bird
flocks
and
they
did
this
back
in
the
80s,
where
they
had
these
sort
of
digital
birds
that
each
bird
has
a
set
of
rules
on
board
that
sort
of
govern
their
interactions.
C
So
it's
like
each
bird
will
have
a
set
of
rules
that
say
don't
get
too
close
to
your
neighbor.
Don't
get
too
far
away
from
your
neighbor
when
your
neighbor
turns
turn
with
the
neighbor
and
it's
basically
like
well,
you
know
we
talked
about
cellular
automata
where
you
have
neighbors,
and
the
state
of
the
individual
depends
on
the
state
of
the
neighbors,
so
they
basically
use
a
rule
to
calculate
a
rule
based
on
the
state
of
their
neighbors.
C
Now,
if
you
do
this
in
parallel,
if
you
find
that,
like
you
know,
you
can
have
these
kind
of
cascading
effects
where
the
neighbor
influences
another
neighbor
influences
another
neighbor
and
you
get
these
kind
of
patterns
where
there's
collective
behavior
there's
a
collective
movement
of
a
bunch
of
agents,
there's
activation
across
a
grid.
That
is,
you,
know,
organized
and
looks
like
something.
That's
like
a
blob.
That's
moving,
and
you
know
there
are
a
lot
of
like
lifelike
aspects
to
it.
C
So
they
found
that
in
these
void
swarms
a
lot
of
these
dynamics,
they
say
nucleation
organization
and
collapse.
They
mean
the
patterns
of
the
swarm.
So
a
swarm
is
a
collective
of
agents
moving
around
in
a
coordinated
fashion.
You
know
they
can
those
swarms
can
collapse,
they
can
nucleate
into
smaller
swarms.
C
Those
are
more
diverse
and
larger
flocks.
So
if
you
have
more
voids
like
say,
a
thousand
woods
versus
100
voids
there's
a
more
diverse
set
of
phenomena
that
you
can
look
at,
as
you
may
guess,
they're,
not
highly
predictive.
In
other
words,
if
I
do
one
simulation,
I
get,
I
observe
one
type
of
say,
flocking
versus
another
simulation.
I
see
another
type
of
flocking
it
it's
very
highly
variable,
but
they
have
basically
the
same
behavior.
C
Those
behaviors
are
more
diverse,
but
they
still
resemble
that
that
same
flocking,
behavior
and
then
the
more
agents
you
have
in
a
simulation,
the
more
diverse
that
is
in
smaller
flocks,
like
maybe
a
hundred
agents,
their
their
behavior,
is
much
more
uniform
across
runs
simply
because
there
aren't
a
lot
of
options.
I
mean,
if
you
think,
about
what
I
just
described.
C
You
know
there
aren't
a
lot
of
options.
I
mean
they're
only
like
maybe
100
voids,
so
they
don't
have
a
lot
of
divergence
across
the
flock
it's
pretty
consistent,
but
in
larger
flocks.
You
have
a
lot
more
distance
say
between
one
end
of
the
flock
and
the
other.
So,
there's
a
lot
of
opportunity
for
noise
to
enter
the
equation,
or
you
know
if
there's
a
if
the
the
flop
goes
around
a
barrier,
there's
a
opportunity
for
bifurcation.
C
C
In
the
second
analysis,
which
is
large
web
data,
which
I
guess
is
like
looking
at
this
network
of
sites
on
the
web,
consisting
of
shared
photos
of
descriptive
tags,
tended
to
group
together
users
with
similar
tendencies,
so
they're
using
a
database
here
of
shared
photos
of
descriptive
tags,
they're.
Looking
at
the
I
guess,
the
you
know,
users
and
what
their
photos
are
and
what
the
descriptive
tags
are,
and
so
they're
actually
looking
at
how
that
forms
a
network,
and
so
this
is
network
science.
It's
a
little
bit
different
than
phlox.
C
This
is
where
you
have
a
network
of
users
connected
by
these
common
photos
and
common
tags,
and
so
this
this
type
of
thing
you
just
basically
have
a
bunch
of
users.
You
look
at
how
they're
using
photos
and
putting
together
tags,
and
then
you
see
how
that
network
grows
over
time.
C
C
Sometimes
you'll
hear
about
this
in
terms
of
a
rich
club
network.
Sometimes
you'll
hear
about
this
in
terms
of
a
hierarchical
network,
or
you
know
something
like
that
where
you
have
this
node,
that's
in
the
core
and
then
the
periphery
nodes
are
out
here.
This
node
has
a
high
number
of
connections
as
compared
to
these
nodes
on
the
periphery.
C
There
might
be
another
core
out
here
and
and
so
forth,
and
so
that
core
is
an
important
point
in
the
network
for
traffic
to
flow.
If
you
think
about
like
an
airline
network
where
you
have
hubs
like
chicago
or
los
angeles
or
new
york
in
the
us,
those
are
core
cores
of
the
airline
network,
as
opposed
to
periphery,
places
that
aren't
connected
to
everywhere
else.
So
this
is
the
core
periphery
structure
of
a
network.
C
In
this
case,
novelty
is
not
considered
to
arise
randomly,
rather,
it
is
generated
as
a
result
of
the
word
in
the
structured
network.
So
again
we
get
this
theme
of
having
having
to
have
a
larger
structure
in
place
to
get
a
lot
of
variation,
so
this
novelty
and
variation
is
generated
by
having
a
lot
of
nodes
or
connections
or
voids
or
whatever
we
contextualize.
These
results
in
terms
of
adjacent
possible
theory.
C
So
now
it's
a
matter
of
figuring
out
like
where
you
want
to
go
and
then,
when
you
go
to
that
place,
what
are
the
adjacent
grids
that
you
are
grid
cells
that
you
can
go
to?
So
let's
say
you
can
go
to
these
two
places:
okay
and
then,
if
you're
here
after
the
third
step,
where
can
you
go
from
there
and
it
turns
out,
you
can
go
three
places
right.
C
C
So
we've
in
four
steps,
we've
gone
quite
a
ways
across
our
our
graph
or
our
grid,
and
the
idea
is:
is
that
each
step
you
take
unlocks
a
new
set
of
possibilities,
so
the
adjacent
possible
means
it's
an
adjacent
state
and
then
that
opens
up
another
adjacent
state
and
if
you
take
a
path
through
those
adjacent
states,
you
end
up
with
new
possibilities
and
you
end
up
in
new
places
in
the
grid
or
in
the
world,
and
so
you
can
end
up
very
far
away
from
where
you
started,
based
on
just
kind
of
navigating
through
a
structure
and
so
and
then
it
depends
on
what's
available
to
you.
C
So
you
think
of
this
as
like
configurational
entropy
right,
you
go
to
one
cell
and
then
the
next
cell
and
the
next
cell,
and
when
you
get
to
that,
maybe
fourth
cells
out,
you
are
in
a
different
part
of
the
grid,
but
you'll
also
have
different
possibilities
in
front
of
you.
But
if
you've
taken
another
path,
you
might
take
four
steps
out.
You
might
have
a
whole
new
different
set
of
possibilities
open
to
you.
C
So
if
you
go
out
this
way,
there
are
new
possibilities
out
here
and
so
forth,
and
you
end
up
in
a
totally
different
place,
and
so
that's
that's
what
they
mean
by
adjacent
possible
theory.
That's
an
they're
proposing
that
that's
a
new
way
to
understand
collective
intelligence.
C
You
know
this
is
something
that
I
think
they've.
They
stress
a
lot
in
complexity.
Theory,
you
have
an
initial
condition
and
you
your
system
evolves
away
to
different
novel
states.
It's
a
similar
thing
here,
where
you
have
different
possibilities
in
front
of
you.
You
end
up
in
totally
different
places.
I
think
that's
an
interesting
way
to
see
how
these
systems
navigate
complexity,
that
emergent
systems
aren't,
like
necessarily
predictable,
they're.
C
C
Theory
and
a
lot
of
this
sort
of
you
know
these
sort
of
contingencies
and
that's
actually
can
link
us
back
to
development,
because
development
thrives
on
contingencies,
when
we
have
a
single
cell,
the
cells
divide
and
then
those
cells
divide
to
their
own
daughter
cells,
but
that
constricts
say
where
they
can
be
in
space
or
what
their
fate
can
be.
So
there
are
a
lot
of.
C
There
are
a
lot
of
constraints
on
what
that
next
set
of
states
can
be,
but
it
takes
you
in
very
different
directions,
so,
like
one
cell,
lineage
and
development
might
evolve
down
this
path,
another
cell
one
engine
might
evolve
down
this
path
and
you
get
two,
you
know
very
different
sublineages,
but
they
all
originate
from
the
same
other
cell,
and
so
that's
one
way
to
look
at
emergence,
and
so
this
paper
kind
of
goes.
There's
a
special
issue
that
this
is
associated
with
reconceptualizing
the
origins
of
life
that
might
be
of
interest
to
people.
A
C
A
C
Are
the
boyd
simulations
swarming
behaviors?
So
you
see
you
have
your
agents
in
a
space
and
then
they
have
this
as
far
as
you
add
agents
and
they
move
around,
they
start
to
swarm
and
flock
in
these
in
these
aggregates,
and
you
can
see
the
the
very
complex
patterns
that
they
take,
the
shapes
that
they
take
on.
You
see
thicker
groups
here,
because
they're
more
agents
and
you
can
see
what
they
look
like
they're
very
highly
nonlinear,
very
highly.
They
almost
look
like
turbulence
in
a
flow,
but
that's
kind
of
the
idea
too.
C
In
turbulence.
You
see
some
of
the
similar
things
that
are
going
on
here.
Okay,
so
then
the
second
paper
in
this
series
is
a
blog
post
by
someone
called.
His
name
is
johan,
john
and
he
does
a
lot
of
philosophy
and
actually
he's
a
neuroscientist,
I
think,
by
training,
but
he
has
a
nice
blog
where
he
talks
about
a
lot
of
ideas
that
are
really
kind
of
you
know
theoretical,
and
this
this
post
is
called
what
is
emergence,
and
why
should
I?
C
Why
should
we
care
about
it
so
here
he
shows
some
shapes
that
have
sort
of
these
emergence
structures
in
them
and
how
you
can
basically
use
the
same
assembly
rules
to
different
different
types
of
shapes.
C
And
he
says:
emergence
occurs
when
there
is
a
conceptual
discontinuity
between
two
descriptions
targeting
the
same
phenomena.
This
does
not
mean
that
emergence
is
a
purely
subjective
phenomenon.
So,
like
I
talked
about
with
the
adjacent
possible,
you
coded
two
totally
separate
outcomes
from
the
same
sort
of
map.
C
There
is
an
aggregate
description
of
individual
birds
and
a
description
of
the
flock
in
its
unified
entity,
so
you
can
describe
individual
birds
and
what
they're
doing-
and
you
can
describe
the
state
of
the
flock,
because
you
saw
the
flop
can
be
like
you
know,
have
the
shape
it
can
look
like
a
comma.
It
can
move
around
like
a
blob.
It
can
have
this
sort.
It
looks
kind
of
like
turbulence,
and
you
can
also
describe
what
the
individual
birds
are
doing.
C
You
can
use
like
a
set
of
rules
to
describe
their
interactions
with
other
birds,
but
you
can't
necessarily
get
in
to
sort
of
that
middle
scale,
which
is
you
know
what
you
have
in
different
parts
of
the
flock.
What
are
going
on
there,
where
that
collective
structure,
and
so
this
description
of
the
flock
as
a
unified
entity,
invites
descriptions
in
terms
of
concepts
from
fluid
dynamics.
C
So
this
is
something
that
you
know.
We
talked
about
in
fluid
dynamics,
with
respect
to
flows
and
turbulence,
and
so
turbulence
actually
has
a
number
of
parameters
that
you
can
use
to
describe
the
flow
and
it's
you
know,
but
it
it
depends
on
you
describing
the
entire
flow
it
doesn't
just
doesn't
depend
on
you
describing
the
individual
particles
that
are
in
the
flow
and
and
forming
the
flow.
C
So
this
is
a
sort
of
thing
that
we
don't
actually
have
descriptions
of
individual
birds
or
particles
that
are
you
know
and
then
linking
it
to
the
larger
collective
to
the
larger
scale.
So
that
that's
something
we
don't
have.
We
only
describe
that
individual
particle
or
bird
in
terms
of
what
it's
doing
individually.
C
That
is
what
it's
doing
collectively,
and
then
he
gets
into
this
idea
of
phase
transitions
and
physics,
which
is
where
you
look
at
changes
over
time.
So
those
flocks
can
transition
from
one
state
to
another.
A
flow
can
transition
from
one
state
to
another
and
in
phase
transitions
you
have
different
phases
of
matter,
so
you
can
take
a
matter,
that's
a
gas
and
it
can
transition
into
a
liquid
and
it
does
this
fairly
quickly
and
so
what
they
call
that
as
a
phase
transition.
C
C
So
surface
tension,
for
example,
is
not
defined
for
gases.
Since
gases
do
not
have
surfaces,
but
when
you
have
that
phase
transition
to
liquid
suddenly
surface
tension
can
be
measured
and
the
transition
gas
to
liquid
a
qualitatively
new
attribute,
not
only
emerges,
it
becomes
a
defining
feature
of
the
post-transition
system.
C
So
this
is
again,
you
know
when
you
have
these
phase
transitions,
it
brings
new
properties
into
being,
and
so
it
kind
of
goes
through
a
lot
of
this
emergence
and
ontology,
and
you
know
kind
of
talking
about
how
it
emerges
is
controversial
and
a
lot
of
that
has
to
do
with
the
language
that
we
use
and
the
descriptors
that
we
use.
So
that's
an
important
thing
to
think
about,
and
then
of
course.
C
C
You
can
use
a
reductionist
approach
to
look
at
the
individuals,
but
it
doesn't
really
do
much
to
describe
this
collective,
and
so
that's
the
idea
that
we're
trying
to
get
here
is
that
we're
definitely
not
talking
about
a
reductionist
system
and
we,
although
we
can
reduce
these
systems
down
to
specific
individuals,
individual
constituent
parts,
it
doesn't
really
do
much
to
describe
what
we're
looking
at
so
so
the
parts
of
one
discipline
say
cells
in
the
context
of
organ
level.
Biology
are
the
holes
of
another
discipline
cells
in
the
context
of
molecular
biology.
C
So
we
have
you
know
parts
can
be
holes,
and
so
that's
another
thing
to
consider.
Depending
on
your
perspective,
if
you're
a
molecular
biologist,
the
cell
is
a
whole.
If
you're
an
organismal
biologist,
the
cells
are
parts
of
a
larger
hole.
So
your
perspective
changes
your
interest
in
what's
emerging
so
like
you
might
be
interested
in
the
emergence
of
you
know,
maybe
like
proteins
and
as
a
cell
biologist.
C
C
C
So
we
record
our
meetings
and
I
can
send
you
a
recording
of
this
so
that
you
can
get
some
of
the
information
I
talked
about
earlier
in
the
meeting,
which
was
how
to
write
a
proposal.
So
that's
good
and
then
hari
krishna
had
to
leave.
So
thank
you
for
attending
hare
krishna
and
I
also
hey.
I
think
I
see
there
are
any
other
messages
in
here
that
we
had
someone
more
okay,
yes
for
gsat,
good
okay.
So
yes,
so
thanks
for
attending
the
meeting,
we
have
any
questions
before
we
end
the
meeting.
D
I'll
maybe
report
back
to
you
on
some
of
the
interesting
things.
In
aps,
physics,.