►
From YouTube: DevoWorm (2021, Meeting 31): DevoLearn, Digital Shape/Geometry, LCB III, Pathways, Ethomorphogenesis
Description
Considering the Future of DevoLearn, Digital Shape and Geometry in Liquid Droplets and Axolotl. Liquid Crystal Biology (LCB), Gene Expression, and Turing Morphogenesis. More on the historical contingent pathways of life, Shape, Size, and Muscular Behavior in Cnidaria. Attendees: Bradly Alicea, Shruti Raj Vansh Singh, Jesse Parent, Akshay Nair, Susan Crawford-Young, and Mainak Deb.
B
A
B
B
So
I'm
looking
forward
to
doing
that.
I'm
going
to
talk
about
a
number
of
things
today.
I
don't
know
if
people
other
people
are
going
to
show
up,
but
I
have
we're
going
to
go
over
the
project
a
little
bit
we're
going
to
go
over
some
other
things
that
related
to
the
project.
Maybe
next
steps
in
terms
of
what
we
want
to
do
with
the
diva
learn
project.
This
is
the
broader
project
of
this
diva
learn
platform
and
then
we'll
get
into
some
papers.
B
B
B
B
Repository-
and
this
is
basically
the
official
work
repository
link
the
pipey
package,
so
this
is
on
python
p,
which
is
pi,
which
is
the
it's
like.
You
know
a
place
where
they
they,
you
can
download
packages
for
different
python,
based
open
source
software,
there's
the
web
app
and
then
the
web
code
repository,
and
then
this
is
the
readme
which
has
a
lot
of
the
stuff.
Here,
it's
broken
down
into
the
different
phases
of
the
projects.
This
is
segmenting
nuclei,
defining
the
data
set
and
data
loader.
B
This
is
the
model
here
to
give
some
nice
detail
here:
model
performance,
upgrading,
the
cell
membrane
segmentation
model,
fixing
the
flawed
data,
the
training,
metrics
deploying
evil
or
models
on
the
web.
So
this
is
the
web,
the
web
architecture
for
that,
and
so
yeah
we're
right
now.
I
guess
it's
on.
D
B
Yes,
so
then,
okay
and
then,
but
of
course
you
need
to
handle
all
of
the
the
model
and
the
data.
So
this
is
where
this
comes
in
here
and
then
then
you
go
through.
You
can
step
through
the
web
app
and
then
interactive
plots
and
then
the
weekly
blog
posts,
which,
if
people
are
applying
for
g
suck
next
year,
this
will
be
useful
for
their
to
see
how
to
do
this.
B
So
let's
go
up
to
the
official
work
repository.
This
is
the
this
is
our
repository.
We're
in
this
is
the
pipey
package
here,
which
is
actually
a
divo
learn,
slash,
evil
learn.
So
this
is
the
pipey
package
as
it
exists
on
their
site.
This
is
where
you
can
download
the
package,
install
it
in
your
on
your
own
computer
and
run
it.
B
But
we
have
diva
learn
is
just
hit
10
10
000
downloads
via
pipey.
So
this
is
the
interface
I
was
just
showing
with
over
800
downloads
in
the
last
30
days.
So
that's
really
nice.
That's
a
good
user
base,
potentially
that
we're
building
on
this.
So
this
is
the
you
know.
You
can
look
at
the
stats
for
these
pipey
projects
and
I
know
we
looked
at
it
in
an
earlier.
B
I
think
in
february,
where
my
oak
had
brought
up
the
stats
for
it
and
you
know
it
was
starting
to
grow
then,
but
it
looks
like
it's
grown
even
more
now,
so
we've
got
where
we're
getting
a
critical
mass,
so
we're
at
10
000
downloads-
and
this
was
a
couple
days
ago.
So
congratulations
to
everyone
involved
in
that.
B
D
C
D
B
Had
a,
we
also
have
the
open,
devo
cell,
which
kind
of
does
the
same
thing.
It's
also
hosted
on
heroku
app
and
it's
it's
a
different
model,
but
it's
it's
very
similar
to
this,
so
this
is
divalearn
hero,
herokuapp.com,
and
so
we
have
choose
one
of
the
following
here,
which
is
home
and
you
can
choose
your
type
of
segmentation
or
lineage
prediction
that
you
they
want
in
your
analysis.
B
So
this
just
kind
of
goes
over
the
detail
of
it,
and
then
we
also
have
some
links
to
data
sets
here.
The
epic
data
set
the
cell
tracking
challenge.
One
of
the
things
we
didn't
do
in
in
the
gsoc
period
is
talk
about
the
devozu
platform
and
how
it
were
the
devozu
repositories
and
how
we
might
incorporate
that,
but
I
think
that's
maybe
down
the
road
a
bit,
the
devo
zoo.
Of
course
we
know
and
talked
about
this
in
you
know.
B
For
development
and
we
basically
have
got
sort
of
curated
a
number
of
data
sets
for
each
organism.
So
it's
not
just
c
elegans,
it's
drosophila,
it's
spiders.
Even
you
know
we
have
some
other
types
of
embryo
zebrafish
that
have
like
this.
You
know
we
have
some
data
that
we
you
know
people
can
download
either
microscopy
data
or
data,
something
like
the
epic
data
set
where
you
have
images
and
or
cell
tracking
data.
So
all
those
things
are
available.
B
Now
we
didn't
really
integrate.
Any
of
that.
With
I
mean,
we've
been
doing
some
of
that
integration
with
the
models
in
2020
the
gsoc
project
sort
of
we
did
some
work
on
that
then,
but
so
these
are
data
sets
that
are
probably
the
most
relevant
to
this
model,
though,
and
so
you
can
try
those
out,
but
also,
if
you're
a
neophyte,
you
might-
or
you
know,
a
beginner.
You
might
want
to
try
one
of
these
so
for
nucleus
segmentation
and
I
think
there's
so.
B
You
have
like
some
base
data
here
that
you
can
use,
so
you
can
actually
like
just
put
like
your
own
imaging
data
or
you
can
take
your
image
data
like
you
know
one.
You
know
image
by
image
and
put
it
in
a
file
on
your
laptop
or
whatever,
and
you
can
upload
these
images
and
then
you
can.
You
know,
pull
them
up
and
it'll
segment
them
right
in
this
interface,
and
so
you
have
to
limit
your
file
to
200
megabytes
per
file.
A
lot
of
well
most
microscopy.
B
Images
are
not
that
nearly
that
big,
so
you
know
you'll
be
able
to
do
quite
a
few
images
without
overloading
the
system,
but
so
this
runs
on
this
image,
which
looks
pretty
noisy
here.
But
you
get
a
nice
image
here
and
there's
also
a
threshold
slider
on
this
side,
so
you
can
increase
or
decrease
the
threshold.
B
B
B
You
get
this,
which
is
a
lot
cleaner,
but
you
know
there's
a
trade-off
here.
If
you
really
set
the
threshold
high
on
some
images,
you
will
need
to
set
the
threshold
high
because
there's
going
to
be
a
lot
of
masking
of
noise
but
you're
going
to
lose
some
detail.
So
this
is
something
to
keep
in
mind
if
you're
using
this,
and
so
you
can
do
this
for
any
of
these
examples.
This
is
the
lineage
predictor.
B
So
this
is
from
you
know
this
example
image.
You
can
also
upload
your
own,
and
this
is
a
c
elegans
embryo
and
this
pulls
out
just
for
I
guess
this
is
just
set
up
for
c
elegans
now,
because
different
embryos
have
different
lineages
different
classification
schemes,
but
basically
you
have
these
different
classes
and
it's
able
to
pick
out
based
on
their
location
kind
of
what
cells
you
have
in
here
of
what
lineage.
B
So
this
is
nice.
It
kind
of
works
with
the
labeling
that
we
already
have
for
c
elegans
and,
of
course,
that
works
in
c
elegans
in
other
organisms
like
zebrafish.
It
might
not
work
so
well,
because
we
have
a
different
scheme
of
of
lineage.
So
many
edges
plus
there's
a
lot
of
what
they
call
regulative
differentiation,
which
is
where
cells
just
kind
of
divide,
and
then
they
get
they
they
take.
You
know
they
get
take
out
a
fate
based
on
the
chemical
signaling
around
them,
so
c
elegans,
that's
typically
not
the
case.
B
You
can
tell
what
a
cell
is
as
soon
as
it's
born.
So
this
is
a
nice
feature.
You
know
we
might
in
future
years.
We
might
work
on
this
a
bit.
We
might
work
on
linking
like
a
classification
scheme
to
an
image
array
like
this,
where
you
have
a
bunch
of
cells
in
an
image,
and
you
know
trying
to
figure
out
where
the
you
know
what
the
lineage
is
without.
B
A
label,
so
what
I
mean
is
I,
if
you
have
this
embryo
and
you
see
you
can
maybe
take
a
bunch
of
images
in
a
time
series
and
watch
them
divide
and
start
to.
You
know
this
cell
might
give
birth
to
like
16
cells
around
it,
be
able
to
track
that
lineage
out
and
just
build
like
a
lineage
tree
from
just
the
images
instead
of
like
the
labels.
B
But
that's
not
that's
beyond
the
scope
of
this
year
and
maybe
even
next
year,
because
I
don't,
I
don't
necessarily
know
how
he
would
do
that.
So
that's
that's!
Basically,
you
know
maybe
like
we'd
have
in
the
future,
so
I
don't
have
a
diagram
for
oh
hello.
Actually,
I
don't
know:
okay,
oh
okay.
My
knock
has
something
to
show
us
so
I'll,
wrap
up
the
diva
learn
stuff
and
then
we
can
get
into
the
droplet
stuff.
B
Okay,
all
right
good,
so
I
don't
have
a
map
of
this,
but
basically
we
have
now
we
have
the
open
devo
cell
model,
which
is
not
a
pre-trained
model,
but
it's
a
machine
learning
model
for
I
guess
it's
general
embryos,
but
it's
was
trained
on
c
elegans
and
we
know
it
works
on
c
elegans.
We
can
use
it
for
different
things.
B
We
also
have
the
divo
learn
platform
which
has
in
2020
we
started
with
the
the
basic
pre-trained
model,
and
this
year
we've
augmented
it
now
to
have
this
web
interface,
and
now
we
can
actually
just
do
this.
Now
we
have
the
open
devo
cell
and
the
diva
learn
platforms
that
are
both
two
different
types
of
models
and
they're,
both
web
interfaces.
B
So
now
we
can
actually
just
go
in
and
analyze
data
in
two
different
ways,
and
so
now,
if
you
don't
know
what
you're
doing
or
you
come
to
the
group
and
you're
like
I
wanna,
you
know
try
to
analyze
some
data
to
get
started.
Now
we
have
some
really
nice
tools
for
you
to
use,
and
people
are
downloading
this.
You
know
in
the
general
public.
So
hopefully
we
get
some
feedback
from
that
now,
speaking
of
which
I
wanted
to
go
over,
maybe
next
steps.
B
B
Okay,
so
this
is
the
diva
learn
platform
here,
let
me
show
you
okay,
so
this
is
the
diva
learn
platform
and
this
so
you
have
the
diva
learn,
evil
learn
repository,
and
this
is
where
the
software
is,
and
this
is
where
you
know
you
can
download
a
lot
of
the
code
and
everything
we
also
have
an
education
location.
We
have
theory
building,
we
have
the
c
elegans
diva
learn
which
is
a
web.
The
sort
of
the
broader
web
app
that
supports
a
lot
of
the.
B
B
So
they've
all
contributed
something
to
these
data
science,
demos-
and
there
are
different
things
in
here-
that
if
you
want
to
look
at
like
gans
networks,
centroid
extraction
this
one
here,
which
is
a
command
line-
basics
in
python.
So
there
are
a
lot
of
different
tutorials
in
there
that
you
might
want
to
look
at
if
you're
interested.
B
I
know
that
actually
that
jesse-
and
I
were
talking
on
saturday
at
our
other
meeting-
about
demos
for
different
things-
and
this
actually,
I
forgot
about
this,
but
this
is
something
this
is
focused
more
towards
data
science
for
biology.
But
this
is
some
another
way
that
you
know
we
might
do
these
technical
tutorials.
A
B
This
is
the
dc
gan
digit,
this
yeah,
so
this
is
a
nice.
This
is
something
my
noc
put
in
as
a
tutorial
and
it
just
kind
of
goes
through
like
I
think
this
is
for
gans
implementing
again
there's
a
collab
notebook
and
everything.
So
we
can.
You
know
you
can
go
through
that.
So
this
is
the
the
platform
and
I've
been
presenting
on
this
to
a
number
of
different
groups
emphasizing
the
soft,
the
divo
learned
software,
but
also
the
larger
platform,
and
so
this
is
the
diva
of
zoo
organisms.
B
And
you
know
there
are
other
organisms
like
ants
and
spiders,
which
you
don't
really
have
any
content
for
right
now
on
this
basil
area,
which
are
the
with
the
diatoms,
we
talk
about
siona
intestinalis,
which
is
a
marine
invertebrate,
the
simulated
embryo,
which
is
a
where
we
numerically
simulate
an
embryo
in
terms
of
its
cell
tracking
and
things
like
that
and
then
zebrafish
and,
of
course,
c
elegans.
So
this
is
the
you
know
the
devozu,
where
we
have
all
these
different
organisms,
so
this
is.
B
D
Yeah
so
see
the
code,
is
it's
not
really
important,
but
I'll
just
go
over
what
I
have
so
this
is
the
input
data
that
I
got
from.
So
let's
just
go
ahead
and
go
into
the
image,
so
this
shows
so
this
shows
basically
four
photos
which
are
taken
in
four
different
time
points.
So
this
shows
these
droplets,
which
are
changing
shape
with
time.
So
at
this
point,
they're
really
small
and
then
they
form
hexagon
like
shapes
and
then
they
become
larger
and
then
they
form
into
triangles
and
lines.
D
So
what
the
goal
is
is
to
so
the
goal
is
to
basically
find
out
the
number
of
shapes
that
we're
getting.
So
what
I
did
is
that
I
cropped
into
this
third
image
right
here,
which
is
this
and
then
I'll
just
show
you
the
results
that
I
got
so
this
was
the
input
image
that
I
had
and
then
I
could
basically
map
out
the
shapes
in
this
format.
Like
the
the
borders
of
the
droplets.
I
could.
D
I
could
use
some
opencv
based
transformations
to
get
this
map
and,
if
I
draw
it
up-
and
if
I
drop
that
on
top
of
the
input
image
we
get
this,
so
it
might
be
very
obvious
that
a
lot
of
these
shapes
are
being
missed
out,
because
there's
a
threshold
that
I've
placed
on
the
minimum
area
under
which
the
shapes
would
not
be
detected.
So
if
I
set
the
threshold
to
be
a
lower
value,
more
shapes
would
be
detected,
but
that
will
also
pick
up
a
lot
of
the
noisy
parts.
D
So,
like
this
part
right
here,
this
might
not
be
as
it
might
be,
something
else
it
might.
It
might
just
be
nice
or
these
lines
right
here
they
could
also
be.
They
could
also
be
seen
as
shapes
if
the
threshold
is
said
to
be
very
low,
the
threshold
of
the
area,
but
if
I
set
it
to
be
high,
it
keeps
up
this
level.
So
so
this
is
actually
still
a
working
process
and
I'll
be
working
on
this
in
the
upcoming
weeks,
so
yeah.
That
was
it
for
this
subject.
Oh
that's.
D
Yeah
so
five,
it
basically
shows
that
the
shape
of
this
droplet
has
five
corners.
But,
as
you
can
see,
it's
not
it's
not
really
obvious
that
there
are
five
corners
but
somehow
like
it's.
It
roughly
comes
down
to
five
edges.
Actually
like
it,
it's
a
bit
rounded,
but
it's
actually
a
bit
blurred
out.
I
wish
I
could
show
you
a
more
clear
image.
So
maybe
I'll
come
up
with
that
in
the
upcoming
weeks.
D
I'm
using
a
lot
of
opencv
based
algorithm,
there's
there's
this
thing
called
I'm
using
the
contours
and
I'm
also
using
some
gaussian
blurring
and
then
thresholding
is
also
there
to
binderize
the
image
and
then
there's
also
a
bit
of
kenya's
detection
network.
So
it's
actually
a
mixture
of
a
lot
of
open
cv
based
study
guarantees.
So
I'm
trying
to
find
out
the
best
match
to
find
to
get
the
shapes
properly
and
to
get
to
detect
everything
properly
so
yeah,
that's
something
I
will
be
experimenting.
B
B
B
D
B
D
D
D
D
A
D
D
B
D
The
proposal,
but
I
mean
like
I,
had
to
go
through
so
many
papers
and
I
had
to
try
deriving
one
like
equation
on
my
own,
so
I
could
actually,
you
know
verify
that
it
will
happen,
but
yeah
now
I
actually
tested
it
out
by
you
know
coding
it
and
it's
actually
working
so
on
to
the
next
step
that
is
like,
but
one
doubt
I
had
bradley
is
like
wait.
I'll
show
you
my
proposal
talk
also
so
that
you
can
view
it.
D
B
Okay,
so
yeah,
if
you
have
the
sphere
and
you
you're
looking
say
like
you-
have
the
globe
of
the
earth
and
you're
looking
at
the
north
pole
and
this
image
a
you're
kind
of
looking
down
at
the
earth
from
the
north
pole.
So
you
have
like
different
continents.
You
know
towards
the
pole,
they're
they're,
right
kind
of
in
the
middle,
so.
B
Would
be
the
north
pole
for
this
and
then
you'd
have
the
north
pole
and
all
the
land
that's
near
the
north
pole.
You
could
see
pretty
clearly
and
then
at
the
edges.
It
starts
to
deform
as
you
move
towards
the
sort
of
the
edges
of
like
maybe
towards
the
equator
and
then
what's
happening
here.
Is
that
the
earth
is
flipping.
B
B
B
D
D
Susan,
so
this
this
screenshot,
I
actually
took
it
from
your
paper,
so
you
can
see
right
like
so
what
I
I've
actually
copied.
The
statement,
which
was
also
written
over
the
image
in
the
paper
to
my
proposal.
So
you
can
see
right,
like
a
is
the
flip
to
the
bottom
image,
so
I
actually
had
a
doubt
whether
it
is
the
so
if
you
are
considering
earth,
it
would
be
the
north
polar
it
would
be.
The
south
pole.
A
D
B
B
D
Yeah
yeah
so
yeah,
so
this
is
also
just
a
basic
implementation,
but
now
right
now
I
would
have
to
bring
the
data
set
images
like
from
bottom
and
like,
like
you
said
from
moving
from
down
to
up
so
it
would.
I
am
actually
confused
with
the
plotly
libraries
but
yeah,
but
so
far
now
I
have
got
a
clear
idea
on
what
to
do.
Thank
you
both.
E
D
B
B
Oh
yeah,
all
right
so
yeah,
that's
so
that's
what
akshay
was
showing
this
work
that
he's
been
doing
on
the
axolotl.
He
was
gonna
tile
this
with
that
magic,
but
it's
not
visible
right
now,
but
the
idea
would
be
to
tile
this,
this
reference
frame
with
images
and
then
be
able
to
flip
it.
So
you
know
you
can
do
this
to
explore
the
space
and
the
features
on
the
image
would
have
like.
B
B
Yeah
yeah
look
forward
to
it,
so
let
me
go
back
to
the
submissions
document,
so
we're
talking
about
the
diva
learned
paper
and
that's
something
that'll
be
probably
on
the
front
burner
now
that
we're
done
with
the
summer
period.
So
this
will
be
like
updating
the
paper
and
then
pushing
it
out
into
a
preprint
server.
You
know
putting
more
stuff
in
the
paper,
so
let's
be
on
the
lookout
for
that.
B
I
want
to
go
over
the
meeting
and
get
some
feedback
on
some
of
where
we
are
with
it,
but
it's
not
quite
ready
yet
for
that.
So
all
that's
going
to
happen,
probably
in
the
next
few
weeks,
and
then
we
have,
we
still
have
some
of
the
outstanding
issues
here
we
have
this
boring
billion
idea
for
a
book
contribute
set
of
book
contributions.
B
We
have
the
kindle
book
and
I
talked
to
krishna
about
that
and
he
says
he's
still
working
on
it,
so
that's
still
sort
of
in
the
works
mathematics,
diva
worm.
I
think
I
showed
that
last
week.
That
is
something
that
I've
been
adding
to.
So
we
not
only
have
these
different
network
models
and
and
spatial
models
and
and
von
neumann
neighborhood.
Models
like
this
is
for
a
cellular
automata,
but
we
also
have
these
other
models.
This
is
from
a
paper
we
did
on
game
theory
applied
to
embryos.
B
B
So
we
have
like
these
different
models
and
then
we
have
them
in
front
of
us
with
sort
of
what
their
importance
and
then
you
know
we'll
go
from
there
and
see
where
we
can
put
this.
Maybe
you
know
make
it
as
another:
pre-printer,
maybe
put
it
in
some
sort
of
journal.
You
know
it.
I
think
it's
a
nice
resource
because
there's
nothing
like
it.
I
think
this
originally
started
when
I
think
we
were
caught
in
the
open
worm
group
in
the
group
of
senior
contributors.
B
We
were
contemplating
like
that
worm
book,
which
is
this
reference
for
c
elegans
researchers,
why
they
didn't
really
have
any
computational
information
in
it
like
it
was
all
biology
and
there
was
no
computational
stuff,
and
I
don't
know
if
this
would
be
a
good
fit
for
worm
book.
But
it
definitely,
I
think,
would
be
useful
as
a
reference
for
people-
obviously
maybe
not
in
this
form,
but
in
a
more
research
friendly
format.
B
So
we
have
these
other
things:
the
test
of
williamson
symbiosis,
the
molecular
level
simulations
the
diatoms,
those
are
kind
of
in
progress
and,
of
course,
minox
showed
us
his
work
with
the.
Where
is
this
thing?
Oh,
the
shaped
droplets
work,
so
I
know
surety
and
susan
weren't
in
the
meeting
at
the
time,
but
he's
been
doing
some
work
with
shape,
droplets
and
and
categorizing.
Some
of
the
droplet
shapes,
and
you
know
trying
to
analyze
it
by
the
number
of
sides.
So
it
looks
pretty
interesting.
B
B
B
He
wrote
a
book
called
the
physics
of
liquid
crystals,
and
so
I
think,
if
you've
heard
of
liquid
crystal
televisions,
you
think
what
does
it
have
to
do
with
biology
and
the
answer
is
well.
It's.
It's
biophysics,
basically
there's
this
theory
of
liquid
crystals
liquid
crystals
are
these
these
molecular
systems
that
you
know
behave
collectively
and
the
idea
is
that,
as
they
behave
collectively,
they
exhibit
a
lot
of
these
properties
that
you
can
classify
and
that
describe
sort
of
these.
These
broad
behaviors
and
people
have
done
this.
B
B
B
So
this
is
like
you
know
the
the
sphere
that
akshay
showed
is
just
like
the
you
know
the
beginnings
of
this.
You
can
have
much
more
complicated
topologies
to
put
these
phenomena
on.
So
you
know
if
we
want
to
explore,
say,
for
example,
the
edge
of
a
sphere-
that's
one
thing,
but
then
there's
so
much
structure.
B
B
Or
look
at
it
from
different
angles.
So
that's
why
that
kind
of
approach
where
you
have
things
on
a
sphere
where
you
have
things
where
you
can
rotate.
It
is
useful
because
it's
you
know,
depending
on
how
you
look
at
it
or
you
know,
depending
on
your
perspective,
you
can
analyze
things
a
lot
more
effectively
and,
of
course
this
is
right
at
sort
of
the
beginning
of
super
computing.
B
So,
as
super
computing
is
grown
in
power,
we
can
do
a
lot
of
that.
We
can
simulate
things
virtually
and
you
know
see
things
we've
not
seen
before.
So
you
know
in
in
liquid
crystal
biology.
We
have
these
building
blocks
here
we
have
small,
organic
molecules,
long
helical
rods
and
associated
structures.
We
have
these
different.
I
think
I
talked
about
this
last
week
or
the
week
before
we
have
these
different
three
different
phases,
where
you
have
a
pneumatics,
cholesterics
and
smectics.
B
A
B
B
They
think
about,
like
you
know
how
things
are
organized
at
these
different
range
length
scales,
so
whether
things
are
organized
locally
or
globally
makes
a
difference
in
how
the
collective
behaves,
and
so
this
could
be
cells.
This
could
be.
You
know,
you
know
biochem
biochemical
molecules.
This
could
be
a
lot
of
different
things
that
you,
you
know
you
can
think
of.
B
You
know
they
go
through
distortion,
textures,
dynamical
properties
and
then
good
into
like
cholesterol,
smectics
and
what
those
represent.
So
that's
that's.
This
is
a
a
lot
of
book
here,
but
this
I
get.
I
I'm
giving
this
to
people
as
a
reference
and
it
might
be
a
little
over
people's
heads
in
certain
parts,
but
don't
be
afraid
of
that
you
just
you
know
if
you
find
an
area
that
you
like
in
the
table
of
contents,
just
go
to
that
area
and
you
know
read
up
on
it
and
you
know
it.
B
B
A
related
paper
is
a
quantum
magazine
article
on
trying
patterns-
and
this
is
turn
patterns,
turn
up
in
a
tiny
crystal
so
making
this
connection
between
turing
patterns,
which
we've
talked
about
with
respect
to
biology
and
chemistry,
and
then
a
tiny
crystal,
which
is
the
system
that
they're
going
to
talk
about.
So
the
mechanism
behind
leopard
spots
and
zebra
stripes
also
appears
to
explain
the
pattern.
Growth
of
a
bismuth
crystal
extending
alan
turns
1952
idea
to
the
atomic
scale.
So
we've
talked
about
the
mechanism
behind
turing
patterns,
which
is
turing
morphogenesis.
B
B
You
know
organized
if
they
organize
pattern
at
the
organized
stripes
if
they
organize
fields
of
different
colors
and
things
like
that-
and
you
see
this
in
biological
systems
in
this
case
you're,
seeing
a
crystal
at
the
atomic
scale,
which
is
much
smaller
than
what
we
usually
think
of,
and
so
this
is
the
nanometer
wide
stripe.
Seen
in
crystal
growth
left
closely
match
those
form
and
formed
in
a
theoretical
model
based
on
turing
patterns
right.
B
Of
course
you
can
implement
them
in
cellular
automata,
so
we
can
build
a
computational
model.
I
don't
know
if
that's
what
they're
doing
here,
but
you
can
build
a
computational
model
of
this
and
you
can
replicate
something
you
see
in
nature,
and
so
the
idea
is
that
you
know.
If
we
can
do
that,
we
can
build
a
model
that
replicates
those
mechanisms
and
if
it
works,
then
we
know
it's
probably
working
in
nature
in
a
similar
way.
B
So
in
this
case
several
years
ago
a
team
of
physicists
at
stanford,
led
by
this
scientist,
was
trying
to
grow
a
thin
layer
of
bismuth
crystal
on
a
metallic
surface,
but
instead
of
forming
a
uniform
sheet,
the
crystal
became
a
patchwork
of
une
even
growth.
So
you
know
they
tried
to.
They
tried.
They
thought
that
they
were
going
to
get
this
uniform
sheet
of
crystal,
but
it
became
a
patchwork
in
some
areas.
B
Those
were
the
crystal
layer
was
only
one
atom
thick,
a
striking
design
emerged,
small
stripes
filled
with
regular
patches
and
these
regions
butted
against
one
another,
these
stripes
oriented
at
different
angles.
So
it's
like
this
image
here,
where
you
have
these
stripes
that
are
sort
of
irregular,
but
they're
regular
in
terms
of
their
period
but
they're
irregular
in
terms
of
their
shape,
so
they're
kind
of
curved
here,
as
you
can
see,
and
so
they
couldn't
explain
the
stripes,
then
working
on
a
working
trip
to
paris
in
2017
showed
them
to
another
theorist.
B
Who
said
this
is
like
a
zebra
and
that's
you
know
kind
of.
If
you
look
at
it,
you
say:
well,
that's
like
a
zebra
okay,
why?
Why
is
that
relevant?
And
then?
And
if
the
stripes
were
real
like
a
zebras,
then
they
could
be
a
turing
pattern.
So,
okay?
Well,
that's
a
hypothesis!
So
what
happened?
Why?
Why
is
this
important
as
it
turns
out
zebra
stripes?
Have
this
sort
of
process
behind
it
called
the
turing
pattern
formation
returning
morphogenesis-
and
this
is
something
of
course,
alan
turing.
B
B
Trans
paper
described
a
theoretical
mechanism
based
on
two
substances:
an
activator
and
an
inhibitor
inhibitor
that
diffused
across
the
different
areas
of
different
rates.
So
what
happens?
Is
you
have
these
chemical
agents
that
diffuse
across
the
space
with
all
these
molecules,
and
they
act
either
activate
or
inhibit
behavior,
so
it
could
be
like
growth
or
it
could
be
differentiation
or
something
like
transformation,
or
something
like
that,
and
the
idea
is
that
these
activators
will
contribute
to
one
a
stripe.
B
B
The
interaction
between
these
two
morphogens
is
trying
called
them,
allows
one
to
interrupt
the
effect
of
the
other,
creating
a
pattern
of
colored
lines
on
a
tropical
fish
or
something
that
is
rather
than
a
solid,
color
stripe.
So
the
activator
will
override
the
inhibitor.
Sometimes
the
inhibitor
will
override
the
activator,
depending
on
the
concentration
of
each
in
some
location,
but
in
the
case
of
a
bismuth
crystal,
there
is
no
diffusion
molecules.
Don't
shift
randomly
and
spread
out
while
reacting
with
each
other
and
then
after
three
years
they
ended
up
with
a
simulated
pattern.
B
This
is
published
in
nature
physics.
They
have
a
link
to
the
paper
here
that
look
almost
identical
to
the
stripes
and
the
real
crystal,
and
so
this
is
a
nice
match
of
the
model
and
the
experiment,
and
so
that's
the
paper
here
it
what's
interesting
here.
I
think,
with
this
liquid
crystal
perspective,
is
that
you
know
we
have.
We
talked
about
gene
expression,
gene
regulatory
networks
before
and
gene
regulatory
networks
are,
of
course,
one
way
in
which
this
works.
B
You
can
have
the
expression
of
different
chemicals,
chemical
agents
in
a
you
know
in
a
certain
location,
but
you
can
also
just
do
this
with
physics.
You
don't
really
need
to
have
a
genetic
system,
so
this
is
something
you
know.
These
are
two
different
sort
of
hypotheses
as
to
how
patterns
are
formed.
You
can
have
this
sort
of
regulatory
network
where
things
are.
You
know
that
these
morphogens
are
expressed
in
certain
locations.
B
So
you
know
there
and,
and
of
course,
a
general
regulatory
system
or
something
controlled
by
one
of
these
is
also
sort
of
emergent
as
well,
but
those
are
two
different
mechanisms,
so
I
think
that's
an
interesting
thing
to
follow
up
on.
I
know
susan
wants
to
do
some
things
with
soft
active
materials,
and
that
is
so.
We
should
keep
that
thread
going
in
our
meetings.
Anyways.
I
didn't
have
much
more
to
say
about
it.
I
just
wanted
to
highlight
some
of
these
things
here.
B
B
So
this,
of
course,
you're
talking
about
this
turning
morphogenesis.
This
reaction
diffusion
process
that
I
mentioned
previously
that
generates
facial
standing
waves.
These
condensations
are
transformed
into
the
nodules
and
rods
of
the
cartilaginous
and
eventually
bony
endoskeleton.
So
there's
this
process
where
bones
are
built
or
bone
material
was
built
out
of
these
cells.
B
The
second
outcome
is
that
temporal
periodicity
results
from
intracellular
intracellular
regulatory
dynamics,
so
this
is
where
you
have
this
gene
expression
network
working
to
form
patterns,
and
so
this
is
happening
at
one
level,
but
also
this
turing
reaction.
Diffusion
model
is
working
at
another
level,
so
they're,
basically
saying
that
you
have
this
modulation
at
one
in
one
sense,
in
the
intracellular
sense,
but
also
outside
of
the
cell
across
cells.
Your
head,
you
have
this
general
turing
type
morphogenesis.
B
So
here
we
review
the
experimental
evidence
from
the
chicken
embryo
interpreted
by
a
set
of
mathematical
and
computational
models.
The
spatial
waveforming
systems
are
based
on
two
glycobinding
proteins,
so
these
two
galectin
molecules
in
interaction
with
each
other
and
the
cells
that
produce
them
that
the
temporal
oscillation
occurs
in
the
expression
of
the
transcriptional
co-regulator
s1.
So
you
have
these
different
proteins
and
transcription
factors
that
are
playing
roles
here,
so
you're
putting
some
biological
flesh
onto
these
bones
of
the
model.
B
The
multicellular
synchronization
of
this
oscillation
across
the
limb
bud.
So
this
is
across
cells,
so
wants
to
coordinate
the
biochemical
states
of
the
mesenchymal
cells
globally,
thereby
refining
and
sharpening
the
spatial
pattern.
So
this
is,
of
course,
something
you
see
in
turing
models
where
the
spatial
pattern
is
sharpened
by
you
know
the
synchronization
of
different
signals
significantly.
The
waveforming
reaction,
diffusion-based
mechanism
itself,
unlike
most
turing-type
systems,
does
not
contain
an
oscillatory
core
and
maybe
evolved
to
this
condition
as
it
came
to
incorporate
the
cell
matrix,
adhesion
then
enabled
the
pattern
forming
capability.
B
B
This
is
where
you're
starting
to
get
so
you
have
these
different
condensations
that
are
going
on
in
this
bud.
So
this
is
across
cells
in
the
bud.
Different
structures
are
exhibiting
periodicity
in
different
ways,
and
then
this
rose
shading
is
where
there's
a
morphogenetic
activity.
So
you
know
it.
You
start
off
with
this
bud
here
at
the
bottom
and
then
the
leading
edge
here
out
at
the
edge
is
the
thing
that's
expanding
and
growing,
and
everything
is
sort
of
differentiating
outward.
B
E
B
B
B
E
B
B
B
That
you
know
that's,
there
are
probably
a
number
of
reasons
for
that,
but
I
think
that's
an
interesting
point.
I
wonder
if
there
we
could
find
a
paper
on
that
where
they
kind
of
link
those
things
together,
because
I
mean
it.
I
don't
know
if
it's
like
that,
there's
something
going
on
in
the
developmental
program
or
if
there's
something
that's
like
if
they're
linked.
A
E
B
Yeah,
that
would
be
great
if
you
could
do
that.
It
would
be
an
interesting
thing
to
follow
up
on
yeah,
so
yeah.
That
would
be
really
nice.
So
yeah
I
mean
that's
like
again.
You
know
that's
something
we
can
follow
up
on
as
well,
so
this
this
is
following
up
on
our
discussion
on
replaying
the
tape
of
life
last
week
in
in
different
pathways.
B
This
is
a
paper
by
david
krakauer,
who's
at
the
santa
fe
institute
and
christopher
kemps
and
they're
talking
about
the
multiple
paths
to
multiple
life.
I'm
not
going
to
go
very
deeply
into
this
because
it's
a
pretty
involved
paper,
but
so
we
argue
for
multiple
forms
of
life
or
wise
through
multiple
different
historical
pathways,
and
we
talked
about
last
week
about
the
different
like
when
you
replay
the
tape
of
life.
If
you
start
it
over
and
you
allowed
life
to
evolve,
you
know
once
again
from
its
precursors
on
earth.
B
Would
you
observe
the
same
outcome
and
of
course,
if
you
had
alien
life
on
another
planet,
would
it
follow
the
same
outcome
as
you
see
on
earth?
You
know
you,
you
have
a
different
planet.
You
have
a
different
planetary
history.
Would
it
look
the
same
and,
of
course
a
lot
of
people
would
say?
No,
but
maybe
there'd
be
some
commonalities.
A
B
There
are
multiple
historical
pathways
one
can
take.
So
from
this
perspective,
there
have
been
multiple
origins
of
life
on
earth.
Life
is
not
a
universal
homology
in
earth
science.
There's
this
idea
that
life,
the
original
life
is
actually
there
have
been
many
sort
of
starts
to
life,
and
we
believe
that
there's
like
this
common
origin
of
life,
but
that
these
multiple
starts
of
life
were,
you
know,
really
early
in
the
history
of
life
and
that
there
was
a
lot
of
there
wasn't
a
lot
of
branching
back.
B
But
what
happened
was
the
the
the
signal
that
we
have
now,
which
is
usually
like
molecular
data
dna
and
so
forth,
have
a
point
to
a
universal
origin,
but
the
universal
origin
could
be
more
complex
than
we
would
like
to
think.
So
you
know
there's
maybe
a
lot
of
start
potential
starts
to
life.
Maybe
some
of
those
went
extinct,
but
the
starts
of
life
that
persisted
the
common
day.
B
B
It
could
be.
You
know,
the
sort
of
the
environment
that
it's
in
so
there
there's
this
old
idea
of
rna
clays.
I
think
that's
been
discredited
at
this
point,
but
there's
this
idea
that
you
know
life
takes
many.
For
you
know,
life
can
take
many
forms
of
a
substrate
that
where
life
can
replicate
and
survive,
and
then
software,
which
is
the
evolve
function.
So
how
do
things
get
inherited
over
time?
B
So
you
have
this
sort
of
imperative
for
optimization
or
fitness
maximization
or
something
you
also
have
energetic
constraints,
meaning,
if
you
don't
have
energy,
you
can't
live
or
replicate
and
then
a
level
of
material.
So
you
need
to
have
some
basic
material
and
a
lot
of
cases
is
dna
or
rna.
Sometimes
it's
both
in
a
cell.
That's
a
compartmentalized
thing,
and
so
this
is
how
we're
just
considering
different
ways
life
could
have
originated.
B
You
could
start
with
rna.
There
are
things
called
protocells
which
they
can
build
in
a
lab
where
you
have
this
container,
that
it's
filled
with
rna
molecules
and
those
are
sort
of
on
the
boundary
of
living
and
non-living.
So
you
can
have
this
sort
of
system,
and
you
know
this
could
have
been
something
in
early
life
and,
if
you're
on
another
planet,
so
you
could
might
find
protocells
and
maybe
protocells
that
have
evolved
into
other
things.
B
The
functions
essential
to
life
are
realized
by
different
substrates
with
different
efficiencies.
The
functional
level
allows
us
to
identify
multiple
origins
of
life
by
searching
for
key
principles
of
optimization
in
different
material
form,
including
the
prebiotic
origin
of
protocells,
the
emergence
of
culture,
economic
and
legal
institutions.
So
you
can
apply
this
broadly
to
cultural
evolution
to
software
evolution,
software
agents
etc.
B
B
Life
should
be
thought
of
as
a
special
class
of
convergent
evolution.
The
multiple
origins
of
life
on
earth
happen
to
have
a
common
historical
trajectory
in
lucca.
So
this
is
the
last
universal
common
ancestor,
as
has
been
noted,
if
new
life
were
created
in
a
computer
in
a
laboratory,
those
specific
substrates
are
set
up
by
humans
and
create
a
causal
link
with
luca.
B
So
you
know,
there's
this
and
then
living
across
levels,
so
they're
trying
to
sort
of
get
a
handle
on
these
layered
or
multi-level
structures
for
thinking
about
life,
inspired
by
mars
levels
of
information
processing
for
vision,
which
is
something
that
we
talk
about
a
lot
in
artificial
intelligence,
and
so
it's
an
interesting
take
on
it.
It's
you
know,
thinking
about
life
as
a
sort
of
sentient
thing
that
you
know
might
be
you
know,
a
lot
of
people
argue
about
the
definition
of
life.
B
This
is
another
way
to
look
at
this,
so
this
is
looking
at
different
layers
of
information
processing,
but
it's
a
useful
analogy
for
illustrating
the
type
of
theory
that
we
want
to
build
in
terms
of
the
origin
of
life.
So
it's
an
interesting
paper.
I
think
jesse
would
be
very
interested
in
this
paper
and
this
shows
the
materials
of
life,
the
constraints
and
the
optimization
principles.
B
So
you
have
these
constraints.
Physical
constraints
we've
talked
about
in
the
last
couple
papers.
We've
talked
about
engineered
constraints,
energetic
constraints.
We
have
these
material
possibilities,
so
you
can
have
life
as
crystals
or
life
as
like
biological
biomolecules
or
maybe
even
rocks
or
clay.
You
know,
I
don't
know.
I
mean
this
is
the
point:
is
that
they're
saying
that
you
could
have
a
wide
range
of
materials
that
could
possibly
be?
B
This
is
where
you
have,
depending
on
the
history
of
the
toxin,
that
you're
looking
at
life.
The
origins
of
life
may
be
different,
they
may
be,
they
may
converge
upon
a
common
ancestor.
So
that's
another,
that's
a
follow-up
to
that
discussion,
and
then,
finally,
I
wanted
to
talk
about
one
more
paper
here.
If
I
can
find
it
this
one
here,
this
would
be
interesting.
B
This
is
the
ethology
of
morphogenesis,
reveals
the
design
principles
for
nidarian
size
and
shape
development.
So
this
is
talking
about
morphogenesis
again.
This
is
talking
about
cnidarians,
which
are
a
marine
invertebrate,
and
this
is
talking
about
their
size
and
shape
development.
But
I
think
in
this
case
they're
talking
about
how
behavior
regulates
that
so
the
abstract
reads
during
development
organisms
interact
with
their
natural
habits
undergoing
morphological
changes.
Yet
it
remains
unclear
whether
the
interplay
between
developing
systems
and
their
environments
impacts
animal
morphogenesis.
B
B
We
demonstrate
that
the
muscular
hydraulic
machinery
that
controls
body
movement,
which
is
this
idea
of
muscle
being
like
this
hydraulic
type
mechanism,
directly
drives
larva
polymorphogenesis
unexpectedly
size
and
shape
development,
are
differentially
controlled
by
antagonistic
muscles.
So
this
is
where
the
muscles
are
controlling
the
size
and
shape
development
in
this
organism.
A
simple
theoretical
model
shows
how
a
combination
of
slow,
priming
and
fast
pumping
pressures
generated
by
muscular
hydraulics
acts
as
a
global
mechanism,
mechanical
regulator
that
coordinates
tissue
remodeling
altogether.
Our
findings
illuminate
how
dynamic
behavior
modes
in
the
environment
can
be
harnessed
to
drive.
B
B
So
this
is
a
very
different
model
organism
in
c
elegans
you
have
this
larva
polyp
transition,
and
so
this
is
this
is
where
they're
looking
and
they're
looking
at
how
you
can
have
these
variability
in
development
based
on
some
of
these
muscular
behaviors
and,
like
I
said
in
c
elegans,
the
muscles
behave
in
a
way.
That's,
I
think
the
pictures
are
down
here.
Okay,
in
a
way,
that's
autonomous,
and
then
it
hooks
up
to
the
connectome
in
this
case
they're
largely
interested
in
that
sort
of
muscle
autonomy.
B
B
So
they
have
two
conditions:
low,
motility
and
high
motility,
and
then
this
is
where
you
have
different
expansions.
So
this
is
an
isotropic
expansion.
This
is
an
isotropic
expansion
where
it
expands
along
two
axes,
not
just
the
anterior
posterior
axis
this
oc
and
then
no
elongation,
and
so
these
behavioral
modes
sort
of
correlate
with
morphodynamics.
B
In
this
way
you
have
projection
of
the
oral
and
able
oil
pull
location,
necession
motile
animal.
Here
in
this
b
and
c
this,
these
graphs
are
where
you
have
time,
shifted
distributions
of
morpho
dynamics,
precession
motile
animals.
So
this
is
the
sessile
animal
for
these
different
conditions.
This
is
the
motile
animal,
so
a
degree
of
motility
and
the
number
of
animal
or
the
percent
of
animals
that
do
this.
B
And
so
this
is
an
example.
This
is
a
picture
of
the
organism
in
more
detail,
we're
going
from
this
stage
here
to
the
polyp,
which
is
the
stage
where
they
have
more
defined
morphology,
and
this
is
the
elongation
that
occurs,
and
so
you
can
see
that
this
is
what
they're
looking
at,
and
so
you
can
see
that
they're
different
behaviors,
based
on
what
the
muscles
are
doing,
that
determine
how
this
expands
again.
B
B
If
you
knock
out
this
tbx
20
shape
is
preferred
over
size
if
this
bmp
this
growth
factor
when
it's
knocked
out
size
expands
over
shape,
and
so
you
have
these
different
types
of
larva
that
grow
in
different
ways,
and
so
it's
an
interesting
way
to
characterize
this
process,
and
so
that's
a
behavior
driven.
So
that's.
B
Those
papers,
I
think,
that's
enough
for
today,
so
is
that
a
hydra,
it's
actually
not
a
hydra.
It's
a!
Let
me
go
back
to
the
papers
here,
it's
a
nigerian,
so
it's,
I
think
it's
closely
related
to
a
hydra,
but
the
actual
species
here
is
no
matter.
Stella
vectensis
and
I'm
not
familiar
with
that
organism,
but
the
it's
a
you
know
it's
a
nigerian,
it's
pretty
close
to
air,
but
they
look
very
similar
in
a
lot
of
ways.
B
And
so
surety
had
network
issues,
and
so
now,
actually
I
asked:
does
it
have
theories
about
surface
tension
of
liquid
crystals?
So
I
think
that
yeah
they
do
talk
about
maybe
surface
tension.
They
do
some
experiments
with
liquid
crystals,
where
they,
you
know,
they're
certain
experiments
where
you,
you
know
where
you
have
a
liquid
and
then
in
a
liquid
phase,
and
the
different
molecules
can
migrate
out
to
the
edge
of
the
liquid.
So
they
have
like
you
know
they
do
types
of
different
types
of
experiments
like
that.