►
From YouTube: DevoWorm (2023, #7): DevoLearn, Apoptosis + Mechanosensation, C. elegans GNNs, Network Universality
Description
Progress on DevoLearn/DevoGraph and reviewing Huggingface and Github. Multicolored embryo segmentation maps. Paper highlighting the proliferation of Mechanosensation-competent cells vs. Apoptotic cells in morphogenesis. Discussion on incommensurability (Thomas Kuhn) in different fields of biological science. Paper on applying GNNs (Graph Neural Networks) to movement of C. elegans adult movement patterns and stability. How a growing phenotype (blades of grass, stalk of wheat) maintain upright growth through coupled positive feedbacks. Paper and discussion on universality, rank-order trees and networks, and the hierarchy/heterarchy distinction. Attendees: Richard Gordon, Susan Crawford-Young, Sushmanth Reddy Mereddy, Gautham Krishnan, and Bradly Alicea
A
D
A
A
B
D
B
B
D
B
C
Oh
yeah
I,
they
took
off
the
interesting
final
part
about
the
cells
and
embryos,
but
I
think
it
went
fairly
well.
I
had
somewhere
I'm
very
interested
in
it
because
he
was
trying
to
use
hyperelastic
material
and
for
a
J
curve
and
then
I
said
well
can
segregated
and
he
was
floored.
I
was
asking
a
whole
pile
of
questions.
C
A
E
E
Took
a
break
for
four
days:
yeah,
that's
all
with
me
and
and
I
was
about
to
keep
a
PR
by
today.
That's
in
hugging
phase
to
create
a
complete
interface
of
Devo
land
and
I
was
thinking
to
upload
devograph,
also
because
it
all
belongs
to
devil
only
right,
so
keep
thinking
to
host
our
models.
There.
Maybe
I
was
thinking
to
keep
that
spaces
also
like
whatever
development
has
spaces
right
now.
I
want
to
create
a
3D
space.
I
mean
a
space
which
generates
the
4D
structure
of
a
graph.
E
Okay,
actually
there
is
a
thing
we
need
to
run
after
every
three
days.
We
just
need
to
create
a
check,
I
mean
I
mean
we
need
to
run,
run,
don't
run
them
and
they
will
just
stop
and
they
won't
work
from
now
on
on
roads.
I
will
take
care
of
that,
and,
due
to
that,
we
need
to
create
a
run
after
every
three
days
for
three
year
version
of
hugging
base.
We
need
to
run
that
whole
space
after
every
three
days.
Maybe
I
will
integrate
with
some
code
or
I
will
just
select
manually.
B
E
To
run
those
spaces,
okay,
yeah
I,
want
to
complete
I
will
I
will
completely
Host
this
devil
and
hugging
page
like
just.
We
had
spaces
right
now.
I
want
to
create
a
complete
development
models
will
be
hosted,
plexel
membrane,
nuclear
segment
of
lineage
population
model
with
that
demograph
also
and
read
me
about
our
devil
and
when
we
also
will
be
hosted,
I
will
create
an
API
and
I
will
share
you
with
my
today.
E
A
A
A
So,
to
get
this
two
different,
colors
I
had
to
make
some
small
changes
in
the
devil
than
the
post,
because
till
now
what
we
are
doing
is
we
are
just
passing
the
video
to
the
Netherland
and
they
used
to
give
us
the
output.
As
so,
we
couldn't
change
anything
about
it
in
the
Democrat
board.
So
I
made
the
changes
to
development
and
I'll
just
clean
the
code
a
bit
and
keep
the
pr
fight
tomorrow.
E
B
D
B
E
B
No,
this
is
yeah.
This
is
great
yeah,
so
yeah
I
look
forward
to
that
pull
request
and
thank
you.
Sushmath
for
your
work
on
the
hugging
face.
I
was
looking
through
the
hugging
face
set
up
yet,
like
the
the
account
yesterday,
it
looks
like
there's
like
there
are
elements
of
GitHub.
Is
it?
Does
it
interface
directly
with
the
organizational
GitHub
repository,
or
is
it
just
using
the
elements
of
GitHub,
because
it
looks
like
it's
using
like
a
standard
repository
structure
from
GitHub
and.
E
Actually,
what
are
the
lfs
files?
Actually,
whatever
our
reports
are
there?
They
contain
large
file
systems.
Actually,
the
reports
are
already
occupied,
the
all
nfs3
free
GitHub
resources,
so
I
can't
integrate
Hubble
space
and
this
that's
the
main
issue.
If
anyone
wants
to
contribute
incoming
ways,
they
can
just
to
keep
a
PR
for
that
also.
A
E
And
I
was
thinking
to
do
completely,
create
a
same
interface
like
GitHub
and
hugging
face
who
are
on
a
contribute
and
hugging
face.
They
can
contribute
there
and
it
will
be
used
for
Raging
face
community
and
we
can
showcase
that
and
GitHub
will
be
also
I
mean
I
need
to
do
changes
in
the
devil
and
read
me
final
answer:
I
will
do
complete.
Pr
I
will
show
you
a
small
presentation
of
time
thing
in
next
Maybe
by
night.
I
will
work
on
it.
B
If
you
go
to
the
gsoc
website,
they
they
have
the
dates
for
the
program.
You
can
kind
of
put
it
in
terms
of
the
dates
that
you
want
to
like
targets
that
you
want
to
hit
during
the
program
during
the
coding
period.
So
the
first
two
weeks
of
the
program
is
community
building
and
that's
not
really
expected
to
be
too
much
work
on
your
end,
but
you
do
have
to
you
know,
get
everything
ready
to
go
and
then
the
coding
period
you
have.
B
You
know
you
want
to
hit
Milestones
every
week.
So
you
want
to
be
able
to
get
like
week
one
you
want
to
be
able
to
make
sure
that
you
have
everything
running,
no
no
problems,
and
then
you
know
week
two
you
want
to
start
with
a
small
task.
You
know,
I've
had
people
in
past
years
divide
their
work
up
into
two
or
three
sections,
so
they
start
with
like
one.
B
So
you
know
then,
like
you
should
always
build
in
a
little
bit
of
time
to
catch
up
the
enemy
one
or
two
weeks
at
the
end,
where
you
can
finish
up
things
and
then
that
that
should
even
out
so
you
know,
things
won't
go
exactly
as
planned,
but
at
least
you
have
a
good
sense
of
the
the
timing
and
the
Rhythm
of
the
project
in
past
years.
You
know
I've
had
people
who
have
run
into
problems
from
you
know,
based
on
what
they've,
what
they've
proposed.
B
So
the
proposal
will
say:
I'm
going
to
do
these
three
things
and
I'm
going
to
get
through
them
in
a
certain
amount
of
time
and
then
that'll
be
the
project,
and
then
they
hit
a
stumbling
block
where
some
plug-in
doesn't
work
or
you
know
something
isn't
working
the
way
they
want
it.
So
that's
a
problem
that
you
have
to
think
about
when
you
plan
out
your
project.
If,
if
I
can't
do
this
or
if
I'm
running
late
on
this
and
things
are
going
slowly,
then
what
else
can
I
pick
up
on
during
that
week?
B
So
I
don't
fall
behind.
So
you
have
to
think
about
those
things
when
you
write
your
proposal,
so
then,
once
you
write
your
proposal,
it's
that's
pretty
much
it.
You
want
to
put
your
CV
at
the
end
your
resume
and
then
that's
the
whole
proposal.
That
should
be
probably
you
know,
five
to
eight
pages,
I
think
that's
what
they
want.
I,
don't
know
what
the
limits
are.
They
change
it
every
year,
but
that's
that's
pretty
much.
B
What
I'm
looking
for
that
can
help
you
with
you,
know
stylistic
things
or,
if
I
think
that
something
is
too
ambitious
or
too
broad,
then
I'll
tell
you,
because
I
I've
had
that
in
the
past,
where
people
have
proposed
things
that
were
really
Broad
and
it's
like
well.
That
seems
like
a
lot
to
do
in
one
week
or
that
needs
a
little
bit
more
specificity.
B
E
A
E
B
B
C
B
C
But
let's
see
I've
got
mapping
the
Dynamics
of
force
transduction
at
cell
to
cell
Junctions.
That
came
up
that's
an
older
paper
and
then
cellular
mechanotransduction
from
tension
to
function.
So
there
were
a
couple
that
just
sort
of
came
up
when
I'm
searching
for
that.
So
that's
partly
why
it's
interesting!
It's
giving
me
some
other
papers.
Oh.
B
A
B
B
Allow
you
to
show
my
jamboard
here,
I'm
drawing
on
it,
so
it
starts
kind
of
as
a
web
structure
like
this
actually
more
like
this,
and
then
the
cells
in
between
the
digits
for
undergo
apoptosis,
and
this
is
just
a
made
up
limb.
This
area
here
is
the
area
that
dies
off,
so
the
cells
in
this
area
undergo
apoptosis,
and
then
you
end
up
with
these
digits
that
then
go
on
to
develop
the
fingers
or
the
digits
of
the
of
the
four
limb.
B
B
So
that's
what
they
mean
by
apoptosis,
and
so
only
a
restricted
number
of
neighboring
cells
are
selected
to
divide.
So
this
is
another
one
of
these
sculpting
mechanisms
where
you
get
cells
that
either
proliferate
or
die
off.
The
pattern
of
division
is
defined
by
IM,
homogeneous
Yap
median
and
mechano
transduction.
B
B
Cell,
you
have
mechanical
forces
coming
in
and
it
senses
the
mechanical
forces
in
the
cell
behaves
accordingly,
so
we
know
that
mechanical
forces,
when
it's
experienced,
pushing
or
pulling
on
the
cell
body
it
can
be.
You
know
it
can
have
different
responses.
They're
they're,
transcriptional
Cascades
that
occur
because
of
this,
so
the
the
forces
as
information
is
transmitted
to
the
genome-
and
things
are-
you
know,
transcribed
here,
and
so
we
get
this
sort
of
oh
I'm,
not
showing
this
part
of
the
screen.
Okay,
so
this
is
it.
B
This
is
my
drawing
of
the
mechano
transduction
in
the
cell,
and
so
you
have
this
Force.
These
forces
that
act
on
the
cell
and
changes
its
behavior
in
different
ways
and
what
they
mean
by
inhomogeneity
is
means.
It's
not
homogeneous,
it's
heterogeneous,
and
so
that
leads
to
the
third
point
in
homogeneity
arises
from
non-uniform
force
and
nuclear
size
in
neighboring
cells,
and
so
we
know
that
non-uniformity
is
sort
of
the
key
to
some
of
these
processes.
B
So
this
is
where
mechano
transduction
occurs.
This
is
the
cytoplasm,
the
nucleus.
This
yep
factor
is
activated
during
mechano
transduction,
so
mechano
transduction
happens
in
cells
with
a
large
nucleus
and
a
large
mechanical
stimulus,
and
so
this
is
an
example
of
apoptosis
or
the
cell
dyes
in
the
middle
this
one
cell.
Here
then
this
is
where
you
have
spatially
and
homogeneous
mechano
transduction
in
this
cell,
in
blue,
as
opposed
to
the
red.
B
So
the
red
cell
is
apoptosis,
then
the
blue
cell
is
this
this
inhomogeneous
response,
and
then
in
this
white
cell
down
here.
This
is
a
small
nucleus
and
small
mechanical
stimuli.
So
it
is
not
mechanic.
It
doesn't
experience
that
mechano
transductional,
whatever
whatever
it
is,
that
they're
proposing
here,
and
then
that
means
that
this
blue
cell
will
divide
and
fill
the
space
formerly
occupied
by
the
red
cell,
that
are
known
apoptosis,
and
so
it
fills
in
the
cells
that
die
off
those
spaces
get
filled
in
by
the
cells.
B
You
know
sort
of
selectively
speaking
get
filled
in
by
these
cells
that
are
experienced,
mechano,
transduction,
so
basically
you're,
selecting
for
cells
with
a
hot
large
nucleus
and
large
mechanical
stimulus,
and
it's
filling
in
eventually
all
these
spaces.
So
it's
not
like
the
example
of
going
from
the
web
Forum
to
the
two
digits.
It's
that
the
entire
mass
is.
B
It
remains
a
mass,
but
it's
getting
filled
in
with
cells
with
a
certain
competency
as
opposed
to
the
ones
that
die
off
in
apoptosis
or
the
ones
that
eventually,
maybe
even
the
ones
that
don't
have
this
competency
for
mechano
transduction
or
Yap
transmittal.
So
that's!
This
is
a
the
app
is
the
yes
Associated
protein,
which
means
I,
guess
it
goes
it's
accepted
by
the
nucleus.
B
B
Apparently,
because
people
like
the
cartoon-
and
they
just
gave
it
a
name-
you
know
they
gave
it
the
name.
It
was
a
weird
story
about
Gene
naming
like
people
would
name
them
things
just
randomly
and
then
and
then
like
they
do.
Gene
therapy
where
they
have
to
talk
to
people
whose
children
have
diseases
and
they
have
to
talk
about
the
Sonic
Hedgehog
protein,
and
you
know,
there's
been
this
debate
about
whether
they
should
say
well.
Maybe
we
should
change
the
names
of
something
more
serious,
I,
don't
know.
B
B
B
B
We
have
I
think
a
number
of
issues
that
have
been
closed,
but
we
have
new
issues
being
generated
here.
Thank
you,
sushmath
and
delpham
for
doing
work
on
this.
So
this
is
Diva,
learn
and,
of
course,
if
you're
applying
to
watching
this
and
applying
to
Google
summer
of
code,
you
know
we
have
a
lot
of
issues
that
if
you
go
to
the
repository
you
look
through
it,
you
may
want
to
create
an
issue
which
you're
welcome
to
do.
If
you
think
that
something
needs
to
be
improved.
B
B
If
you
want
to
propose
a
change,
you
can
issue
a
pull
request
and
you're
welcome
to
do
that,
for
both
evil
learn
and
for
evil
graph,
and
so
Devo
graph
is
the
newer
version
of
this.
It
sits
on
top
what
Diva
learn
and
it
has
the
graph
neural
network
capabilities,
or
at
least
it's
that's,
where
we're
headed
with
this.
We're
headed
towards
this
we're
building
out
this
pipeline.
E
B
Did
I
did
archive
that
and,
and
we
actually
have
some
new
public
repositories,
and
these
are
the
ones
that
are
the
bottles
and
hugging
face,
so
nucleus
segment
or
membrane
segment
or
lineage
population.
These
are
a
little
bit
more
specialized.
These
are
for
evil
learn,
so
these
plug
into
divale,
learn
and
we're.
E
B
E
B
B
E
B
The
graph
yeah,
so
that
would
be
just
yeah
separate
thing
here.
So
this
is
the
evil
worm
group
yeah
I've
made
an
organization
current
I,
don't
know
why
it's
not
coming
up,
but
and
then
this
is
the
the
hugging
face:
I
guess
they
have
like
organ
or
teams
or
organizations
or
whatever.
This
is
and
then
yeah.
E
I
will
create
this
whole
interface.
I
think
I
haven't
worked
this
week.
If
I
worked
out,
I
would
have
completed
this
whole
thing
right
today.
I
will
keep
a
point
request
for
those
three
also
I
will
create
a
model
card,
and
for
that
I
will
host
the
models
and
I
will
upload.
The
data
sets
what
we
have
actually:
okay,
I,
just
refracted
them
I
will
just
upload
there
and
we
can
show
everyone.
A
A
B
Yeah,
that's
great,
so
something
I
found
interesting
in
a
dick
last
week
mentioned
about
Thomas
in
commensurability,
so
I
found
this
or
I
was
looking
at
it
looking
into
it
afterwards
and
found
this.
This
is
a
philosophical,
blog,
I
think
the
multi-disciplinarian,
and
they
talked
about
this
sort
of
Thomas,
Coon's
work
and
some
of
the
work
he
did
on
this
idea.
So
Thomas
of
course,
wrote
the
book
the
structure
of
scientific
revolutions,
and
that
was
in
1962,
which
I
was
very
influential.
A
lot
of
people
learn
about
that
book.
D
B
B
So
structure
is:
where
proposes
this
idea
of
a
paradigm
shift,
which
is
where,
if
you
have
a
theory,
if
you
have
a
field
of
science,
you
know
the
you
have
a
set
of
theories,
and
sometimes
there
is
a
paradigm
shift
and
where
in
which
some
idea
or
finding
will
overturn
previous
theories
and
create
a
new
theoretical
Paradigm.
So
that
means
a
paradigm
shift,
and
so,
as
saw
it,
the
scope
of
a
paradigm
was
Universal.
B
It's
not
merely
a
theory
but
a
framework
in
world
view
in
which
a
theory
dwells,
and
so
it's
you
know
in
some
Fields.
You
know
it's
debatable
as
to
whether
that
shift
is
like
a
revolution
or
it's
more
like
just.
You
shift
to
a
different
theory
that,
because
I
mean
you
know
it
just
I
guess
it
depends
on
the
field
of
of
science.
A
B
The
theory
of
evolution
itself
or
the
near
the
idea
of
natural
selection
was
revolutionary
so
but
anyways.
This
is
the
the
sort
of
the
philosophy
that
he's
come
up.
You
know
that
that
he's
most
famous
for
is
this
idea
of
a
paradigm
shift,
and
so
he
explained
that
successive
transition
from
one
Paradigm
to
another
by
A
revolution
is
the
usual
developmental
pattern
of
mature
science
and
by
developmental
pattern.
He
just
means
like
the
history
of
ideas,
not
like
developmental
biology.
B
His
view
is
that
paradigms
guide
research
through
periods
of
normal
science,
during
which
any
experimental
results
non-consistent
with
the
Paradigm
are
deemed
erratic
and
discarded,
and
so
you
know
there
are
examples
of
that
throughout
the
history
of
science,
it's
just
basically
what
what
are
people
finding
the
most
relevant
in
any
one
time.
I
mean
you
know.
B
This
persists
until
overwhelming
evidence
against
the
Paradigm
results
in
its
collapse
in
a
paradigm
shift
occurs,
and
then
he
stresses
this
idea
of
in
commensurability
between
Associated
paradigms,
meaning
that
it
is
impossible
to
understand
the
new
paradigm
from
within
the
conceptual
framework
of
its
predecessor.
So
this
is
like
the
copernican
revolution,
Lake
tectonics
or
quantum
mechanics.
So
if
you
tried
to
explain
the
concepts
or
you
tried
to
like
it
well
yeah
you
tried
to
use
the
measure
as
a
quantum
mechanics
using
classical
mechanics.
You
couldn't
really
explain
it.
You
need
a
new
framework.
A
B
Doesn't
it
isn't
consistent
with
classical
mechanics,
so
it
would
be
very
confusing
to
sort
of
integrate
those
measures
into
a
classical
mechanics
model,
and
so
you
have
to
come
up
with
a
new
model
for
it,
and
so
that's
exactly
what
happened.
The
copernican
revolution
is
the
same
thing.
So
as
people
were
measuring
the
positions
of
the
planets,
you
know
they
came
to
the
conclusion
that
the
you
know
the
geocentric
Universe
wasn't
a
sufficient
model.
B
So
you
have
these
conceptual
models
of
how
the
world
works
or
How
the
Universe
Works,
and
you
have
your
sort
of
measurement
techniques
and
your
measurement
actual
measurements
that
you're
making
and
sometimes
they
are
inconsistent
or
incompatible.
So
that's
sort
of
the
idea,
I
guess
behind
incommensurability,
where
there's
this
mismatch
between
the
sort
of
the
framework
or
the
mental
model
and
the
the
actual
measurement
what's
being
done.
B
D
B
D
B
B
D
C
B
D
Oh
there
were
a
bunch
of
a
bunch
of
the
other
ones
like
a
bunch
of
the
pioneers
of
molecular
biology,
were
physicists,
okay
and
now
there's
sort
of
a
periphery
in
terms
of
molecular
Dynamics
and
stuff.
Like
and
engineering
and
stuff
like
that,
but
I
can't
say
you
can't
say
they
they're
not
Central
to
biology
yet
well
they're.
Coming
up
with
some
interesting
theories.
C
D
C
D
B
B
But
this
would
be
like
in
later
on,
after
the
60s
added
that,
if
it
was
a
relativism,
it
was
an
interesting
sort
of
relativism.
It
needed
to
be
thought
of
before
the
tag
was
applied.
Then
communities
outside
of
hard
science
20th
century
social
theory
in
particular,
couldn't
get
enough
of
in
his
Paradigm
shifts.
So
there's
a
very
it
was
very
popular
in
terms
of
the
sort
of
the
idea,
but
whether
that
actually
represents
what's
going
on
in
a
lot
of
Sciences
I
mean
that's.
You
know
that
was.
A
B
B
So
that's
that's
they're,
actually
quite
different
than
the
paradigm
shift,
in
the
sense
that
you
know
you
get
these.
You
get
these
things.
It's
like,
okay,
how
do
you
change
the
way
we
look
at
the
world?
Well,
you
know.
Sometimes
it
has
to
be
aesthetically
pleasing
more
than
actually
consistent
with
the
data,
so.
D
A
B
B
A
B
B
So
actually,
I
wanted
to
talk
about
this
paper
because
it
just
came
out
and
I
know
our
people
left
people
who
are
most
interested
in
this,
but
they
can
catch
up
later.
On
this,
this
was
posted
in
the
slack
the
other
day.
I
found
it
this
week
sometime
and
it's
a
new
paper.
It's
it's
brand
new
and
it's
on
using
these
graph
neural
networks
in
C
elegans.
B
So
one
of
the
things
about
our
project
on
graph
neural
networks
was
that
we
didn't
really
have
a
good
example
of
it
in
C
elegans
and
it's
kind
of
like
in
these
things.
We
just
had
to
imagine
how
it
would
be
useful
in
C,
elegans
and
building
these
you
know
embeddings,
that
would
you
know,
maybe
be
good
for
building
networks
of
cells,
but
this
is
actually
on
adult
C
elegans,
in
fact
aging
C
elegans,
and
this
paper
is
a
GNN
based
model
which
is
graph
neural
networks
for
capturing
spatio-temporal
changes
in
Locomotion
behaviors.
B
So
this
is
where
they're
looking
at
behaving
adult
C,
elegans
they're,
looking
at
their
movement,
Behavior
they're
tracking
this
you
know
movement
in
a
culture
dish
or
something
like
that.
They're
tracking
it
and
then
they're
using
that
tracking
data
to
build
these
graph
structures
and
then
they're
using
graph
neural
networks
to
Analyze
That,
so
the
abstract
reads:
investigating
The
Locomotion
of
Aging
C
elegans
is
an
important
way
for
understanding
the
basic
mechanisms
behind
age-related
changes
in
organisms.
B
However,
The
Locomotion
of
Aging
C
elegans
is
often
Quantified
using
insufficient
physical
variables,
which
make
it
challenging
to
capture
essential
Dynamics.
So
you
know
sometimes
we
with
Locomotion
we'll
use,
like
you
know,
tracking
the
motion
using
like
just
kind
of
like
visualizing
the
worm
in
some
way.
We
need
to
have
better
physical
variables
for
that.
So
this
is.
This
makes
it
easy.
You
know
this
makes
it
hard
to
sort
of
understand
the
movement
Dynamics.
Sometimes,
for
example,
you
might
use
like
some
fluorescent
marker.
B
You
could
put
fluorescent
markers
and
c
elegans
in
like
the
head
cells,
so
you
know
you
can
see.
The
head
is
a
little
Green
Dot
moving
around
or
in
the
tail.
You
might
stain
some
neurons.
So
you
can
see
what
this
trail
of
green
going
through
the
field
of
view.
So
there
are
different
ways
you
can
do
that,
but
that's
not
the
best
way
to
do
it
so
so
to
study
changes
in
The
Locomotion
pattern
of
Aging
C
elegans.
B
In
other
words,
it
tries
to
keep
the
bending
angle
unchanged
and
expects
to
change
The
Locomotion
of
adjacent
segments,
so
it
tries
to
like
not
Bend
its
body
that
much,
but
it
actually
also
expects
the
the
adjacent
segments
will
change
their
Locomotion.
So
it's
sort
of
like
when
they
undulate
back
and
forth
they're,
trying
to
maybe
keep
stability
going
forward,
so
they're
just
kind
of
moving
their
body
back
and
forth.
It's
kind
of
like
when
you
drive
a
car
and
you
try
to
keep
blowing
down
a
straight
line
down
the
road.
B
B
B
Our
model
is
anticipated
to
provide
a
data-driven
method
for
quantifying
the
changes
in
Locomotion
and
for
mining.
The
underlying
causes
of
these
changes
so
I'm
looking
for
some
figures
here
and
probably
at
the
end,
they
have
a
mathematical
model
in
there,
here's
a
here's,
a
figure
one.
So
this
is
the
optimization
of
the
model.
You
have
these
different
segments.
This
is
the
worm.
You
have
dorsal
contraction,
dorsal
relaxation
so
with
in
ventral
relaxation
ventral
contraction.
B
So
you
can
see
that
on
the
dorsal
side
of
the
worm,
you
have
these
contraction
and
relaxation
segments
or
these
these
lateral
line
segments
and
then
on
the
on
the
ventral
side.
You
have
these
I
guess
that's
the
muscle
and
then
on
the
other
side,
you
have
contrasting.
So
it's
like,
where
you
have
on
one
part
here:
four:
five
and
six
dorsal
contraction
on
the
ventral
side.
It's
relaxation!
So
it's
like
Contracting
on
one
side
and
then
relaxing
on
the
other
bristly,
it's
Contracting,
ventrally,
it's
relaxing!
B
So
ventral
is
the
bottom
part
of
the
worm.
The
dorsal
is
the
top
part
of
the
worm,
and
so
this
is
where
we
have
this
this
sort
of
action,
because
it
it
works
down
this.
So
this
is
where
the
worm
is
undulating
and
you
have
contraction
relaxation
in
different
phases.
So,
as
you
have
contraction
here,
it's
bending
in
relaxation
it's
bending
out.
C
A
C
D
What
to
do
with
it?
Okay,
Bradley
on
the
top
of
this
paper,
I
I,
did
some
work
actually
in
the
60s,
not
terribly
competent,
but
I
was
looking
at
the
question
of
why
grass
can
grow
straight
and
if
you
look
at
grass,
it
consists
of
columns
of
cells
which
are
dividing
great
and
I.
Figured.
There's
got
to
be
some
sort
of
major
feedback
mechanism,
because
every
time
a
cell
divides
it
seems
that
the
whole
grassly
will
swing
one
way
or
the
other.
A
D
D
C
D
D
C
A
D
B
D
B
B
C
D
A
C
C
B
Yeah,
so
with
a
model
this
is
they
model
this
for
Ben.
You
know
looking
at
bending
angles
at
different
time
points,
so
the
sundulation
is
over
time
it
kind
of
moves,
and
then
these
bending
angles
are
measured
either
for
the
past
M
time
points
which
is
this
yellow
step
or
the
red
and
the
blue,
put
your
different
steps.
B
C
A
C
B
So
I
think
yeah
I
think
the
bending
angle
is
sort
of
like
the
summary
of
like
the
forces
against
the
surface
so
like
actually
in
the
open,
Worm
Project
or
in
the
open
worm.
Foundation
there's
the
cybernetic
project,
which
is
cyber,
which
is
with
an
s,
and
they
have
modeled
like
worms
crawling
across
different
media.
So,
like
you
know,
they've
got
it
growing
in
different
types
of
gels,
for
example,
and
they
they're
simulating
different
substrate
conditions.
D
B
C
D
C
D
B
Yeah,
so
this
is
a
map
of
bending
angle
and
measured
and
predicted
using
their
model.
This
is
where
you
have
these
loss
values
for
the
model
and
this
just
kind
of
very
technical
in
terms
of
the
model
performance.
It's
the
sensitivity
of
the
model
where
you
see
bending
angle,
measured
for
different
Center
points
and
time
points,
so
you
can
see
that
there
they
also
actually
have
HC
elegans
here
for
different
mutants,
so
you
have
like
different
mutations
and
these
mutations
on
25
egl3
unk709.
B
Those
are
all
movement
related
phenotype,
so
the
mutations
are
such
that
they
affect
the
way
the
worm
moves
and
how
they
respond
to
sensory
input.
So
the
wild
type
is
going
to
give
you
that
kind
of
undulation.
That's
it's.
You
know
normal
as
they
say,
and
some
of
these
mutants
Define
mutants,
of
course,
will
give
you
a
different
pattern.
Movement
we'll
have
trouble
moving
in
that
way.
So
that's
that's
something
they're
looking
here
and
they
can
see
that
for
some
of
these
mutants
the
loss
is
more
severe
for
their
model.
B
Not
surprisingly,
so
that's
something
that
they've
done
as
well.
I,
don't
know
really.
If
they
get
into
some
of
these
Network
embeddings,
which
is
what
I'm
looking
for
here,
they're
just
showing
more
of
the
sort
of
the
forward
and
backward
movement
connection
data
here,
so
these
are
changes
in
the
connection
variables
of
age.
So
they
have
these
connection
variables,
which
are
these
different
segments
here,
it's
basically
the
segments
and
how
they're
connected
so
they
show
some
of
these
changes.
B
So
in
the
front
you
have
more
changes
than
at
the
back,
I
believe
and
that's
true
for
all
these
different
examples
here,
D1
through
D15,
and
so
you
can
see
that
the
mean
for
that
is
in
the
across
the
adult
stages.
This
is
D1
and
D15,
which
are
the
adult
stages
of
the
worm
and
there's
variability
with
respect
to
age.
B
So
d11,
it's
a
little
bit
above
the
mean
for
some
reason,
and
then
it
reverts
back
and
D13
and
D15
not
really
sure
why
that
is,
but
I'm
never
really
familiar
with
the
adult
stages
of
Aging.
But
I
don't
know
what.
If
there
are
specific
things
going
on,
I
mean
later
in
adulthood,
there's
variability
because
they're
getting
old
and
they're,
maybe
not
moving
as
much
they've
laid
their
eggs
at
this
point,
so
so
yeah
they're
able
to
model
it
for
these
different
stages
and
they
show
these
different.
You
know.
B
So
these
are
the
different
parts
of
the
the
connect,
the
graphs
so
they're,
connecting
the
segments
together
and
they're
showing
differences
with
respect
to
age
yeah,
and
so
that's
that's
all
they
have.
But
that's
I
think
that's
quite
a
bit.
I
think
it's
interesting
work
again.
You
could
probably
apply
this
to
plants
or
other
organisms,
and
it
will
probably
give
you
some
pretty
interesting
insights.
C
B
A
B
So
he
gives
examples
of
pecking
orders,
so
the
idea
of
hens
pecking
on
each
other
in
a
fight
that
allow
for
dominance
hierarchies
to
emerge
or
other
types
of
pecking
orders,
like
you,
know,
who's
subservient
to
whom
in
a
population
could
range
from
humans
to
other
like
to
chickens
to
other
animals,
social
groups,
and
so
this
is
you
know.
This
is
something
that
you'll
deal
with
in
data
and
statistical
summaries
of
networks
where
you
have
pecking
orders
and
social
rankings
rankings
are
then
simplified
representations
of
hierarchies.
B
So,
if
we
think
about
like
in
a
organizational
structured
business
on
government
or
in
the
military,
you
have
these
hierarchies
that
are
strict
hierarchies,
meaning
that
there's,
you
know
definitely
rungs
on
the
ladder
or
that
some
set
of
individuals
reports
to
another
set
of
individuals
in
that
culminates
in
a
leader,
a
singular
leader
that
you
know
is
at
the
top
of
the
ranking.
So
this
is
a
hierarchy.
B
Basically,
you
know
we
can
use
rankings
to
derive
a
hierarchy.
Hierarchy
is
one
of
those
Concepts
and
precision
seems
to
get
its
popularity.
There
are
many
more
less
problem,
specific
definitions
of
hierarchy.
These
are
further
more
often
inconsistent.
Most
authors
hierarchies
are
collections
of
levels
from
top
to
bottom.
Members
of
the
hierarchy
would
inherit
their
ranking
of
their
levels,
so
that
one
could
order
numbers
of
different
levels,
but
not
necessarily
members
of
the
same
level.
So
Herbert
Simon,
who
has
talked
a
lot
about
hierarchical
organization
and
his
work
points
out.
B
B
B
They
provide
other
types
of
cues
for
differentiation
and
you
end
up
with
differentiated
cells
on
a
lineage,
but
this
is
more
or
less
the
structure
of
a
strict
hierarchy
where
everything
is
subservient
to
something
else,
where
it's
contingent
on
something
I
guess
in
biology,
it's
a
little
bit
difficult
to
use
the
human
analogy,
but
this
is
also
true
in
like
organizational
charts.
Look
like
this
and
so
forth.
There
is
this
competing
idea
of
heterocky
and
we've
talked
about
this.
B
We
have
a
paper
origin
of
the
embryo,
which
was
with
Rob
Stone
and
Thomas
portage's
and
George
michaelovsky,
as
well
as
myself,
and
in
this
paper
we
introduced
this
idea
of
a
hierarchy
where
you
have
a
network,
but
it's
very
different
from
this
hierarchical
thing.
You
have
this
set
of
nodes
that
are
interconnected.
The
inner
the
interesting
thing
about
the
center
connection
is
that
it
could
be
sort
of
hierarchical
in
the
sense
that
this
node
say,
for
example,
can
control
what
the
rest
of
these
nodes
are
doing.
A
B
B
It
would
be
like
you
know,
a
middle
manager
would
have
would
dictate
what
the
CEO
does
and
what
the
base
worker
does
the
entry
level
worker,
but
in
a
biological
Network
this.
Of
course,
this
kind
of
tree
structure
is
usually
unfolds
over
time.
So
it'd
be
a
little
bit
harder
for
things
to
be.
You
know
for
the
daughter
cells
to
control
the
mother
cells,
but
nevertheless
you
could
have
a
situation
where
one
of
these
nodes
controls
everything
in
the
tree
that
happens
afterwards.
B
There
is
subsequent
divisions
in
the
tree
and
or
subsequent
levels,
and
so
in
that
sense
you
could
have
a
hierarchy.
So
this
is,
albeit
you
know,
the
hierarchy
here
is
all
based
on
ranking.
We
rank
things
according
into
biology
to
their
age
in
organizations
according
to
their
sort
of
pecking
order
and
in
biology
of
pecking
orders
too,
just
not
in
this
example.
B
So
I
wanted
to
make
that
point
and
I
think
like
we're
talking
about
hierarchies
and
we're
talking
about
pecking
orders,
this
is
going
to
lead
us
to
networks,
and
so
in
this
case
this
heterarchy
also
forms
a
network,
but
it's
a
very
different
type
of
network
and
so
Herbert
Simon's
statement
about
it.
You
know
a
hierarchy
being
a
recorder
Matrix
with
very
low
values.
So
we
have
to
think
about
this
now
as
a
matrix.
B
The
entries-
and
these
entries
of
course,
then
are
going
to
be.
You
know
we're
going
to
have
to
put
numbers
in
here
and
so
now,
a
lot
of
these
numbers
are
going
to
be
close
to
zero
and
some
are
going
to
be
higher,
and
that
gives
us
our
hierarchy,
but
we
also
can
get
heterocky
out
of
this,
because
these
values
can
change
and
some
of
them
can
switch
from
near
zero
to
higher
levels,
the
higher
values.
B
A
B
Is
one
of
the
basic
aspects
of
networks?
Science?
Is
that
you
know
networks
like
technical
networks
and
all
sorts
of
you
know
biological
networks.
Social
networks
should
all
yield
the
same
types
of
patterns,
and
so
this
is
where
he
gets
into
some
of
this
stuff
about.
You
know
you
go
back
to
Kant,
who
said
if
fruit,
whether
Free
Will
exists
or
not,
human
actions
like
every
other
natural,
is
that
event
are
determined
by
universal
laws,
and
so
this
kind
of
goes
on
talking
about
some
of
this.
B
Perhaps
the
well-known
most
well-known
examples
are
zip
Salah,
which
is
a
a
varying
to
the
power
law,
but
Zips
law,
and
this
has
been
found
in
human
languages
and
across
other
types
of
systems.
Cities,
the
size
of
cities,
the
distribution
in
a
country
or
around
the
world
conforms
to
Zip's
law.
Zips
law
is
basically
the
frequency
of
the
case,
most
common
word
and
they're,
using
this
context
of
of
a
dictionary
of
words-
and
this
is
where
Zip's
law
originated,
the
frequency
of
the
case.
B
Then
pareto's
principle
of
Vault
distribution,
which
is
the
richest
20
percent
of
the
people,
have
80
percent
of
the
wealth,
and
you
can
have
this
sort
of
inverse
relationship
between
individuals
and
resources,
and
so
you
can
see
this
in
biology
as
well.
We
see
this.
Actually,
we
have
networks
and
we're
going
to
get
into
like
how
networks.
When
we
talk
about
Network
control,
certain
nodes
will
have
a
large
share
of
connections
versus
other
nodes
in
the
network,
and
those
nodes
are
called
hubs
in
those
hubs
that
serve
as
the
basis
for
hierarchy
of
Networks.
B
A
B
Physical
theory
of
critical
phenomena,
the
meaning
of
universality,
came
to
change
in
the
late
in
the
1960s
and
70s
critical
exponents,
describing
the
phase
transitions
seem
to
fall
into
a
few
so-called
universality
classes
describing
how
systems
change
with
length
scale.
The
renormalization
group
Theory
led
to
a
theoretical
foundation
for
the
scarcity
of
critical
exponents.
Moreover,
a
motivated
statistical
physicists
to
search
for
universal
laws
outside
of
physical
systems.
B
It
became
a
prominent
example
of
a
universality
can
appear
from
emergence
and
that's
another
defining
feature
of
complex
systems.
So
we
can
look
at
universality
and
ranking
Dynamics,
and
this
is
where
we
look
at
some
of
this
physics:
flavored
complexity,
science.
We
use
this
to
sort
of
get
at
this
question
this,
so
I
think
they
have
yeah
this
illustration
here
in
figure
one.
This
is
dominance,
relations
within
a
group
of
leghorn
hens,
so
this
shows
ranking
and
how
this
changes
over
time.
B
B
Okay,
so
there
are
two
papers
of
blue
metal
and
any
gays
at
all
and
they
built
they
both
use
these
sort
of
this
physical
way
of
doing
this.
This
is
for
physics,
they're,
probably
other
ways
to
do
this,
but
people
haven't
really
explored
those
because
the
physicists
sort
of
have
the
Monopoly
on
some
of
these
tools
and
ideas.
B
So
so
these
two
papers,
they
build
a
microscope
model
of
list
Dynamics
that
differ
from
the
forward.
So
these
these
these
models
are
the
same,
but
they're
different
in
two
aspects,
so
in
a
guess
is
the
first
paper.
They
study
open
lists
where,
like
many
real
world
rankings,
only
the
top
elements
are
explicitly
ranked.
So,
in
other
words,
only
the
top
parts
of
the
list
are
ranked.
The
rest
are
free
to
move
around
and
we
don't
really
care
about
those.
B
Second,
they
do
not
assume
the
ranking
is
based
on
a
variable
associated
with
the
items.
So
you
don't
have
like
a
variable
of
called
rank.
The
ranking
is
just
inherent
in
the
model.
The
model
of
any
gaze
consists
of
only
replacement
and
displacement,
so
these
are
where
items
can
move
within
the
list
and
offset
the
others
or
an
out
of
list
items
and
substitute
an
element.
A
B
So
we're
talking
about
these
distributions
that
are
long-tailed
so
again,
I
mean
I've
talked
about
this
in
other
meetings,
but
we
have
what
is
essentially
this
Power
Lawn
I'm
just
going
to
draw
it
in
a
way,
that's
sort
of
linear
log,
or
something
like
that
where
you
have
this
function
and
I
like
to
draw
it
this
way,
because
it
kind
of
drives
home
what
the
distinction
is.
This
is
the
upper
part
of
the
rank
here
in
a
rank
order.
This
is
rank
order
and
this
is
like
magnitude
or
something
right
and
then
so.
B
You
may
or
may
not
pay
attention
in
the
order
here,
but
this
part
of
the
rank
is
most
important,
and
this
is
where
you
see
the
most
variation
in
terms
of
magnitude.
This
is
where
you
have
the
most
connections
or
you
have
the
most
power
you
have
the
most
wealth
or
whatever,
and
so
this
is
the
part
that
matters
this
other
part
can
be.
B
You
know,
ordered
in
different
ways,
but
it's
not
so
much
important
because,
as
you
can
see,
it
doesn't
really
contribute
that
much
to
our
understanding
of
the
variation
in
the
rain
or
what?
What
it's,
why
you
know
things
at
the
top
top
ranking
things
are
sort
of
the
most
Salient
in
terms
of
our
understanding.
B
Now
there
are
things
that
can
move
around
in
the
ring,
so
things
from
this
category
can
move
up
to
this
category,
and
so
we
can,
you
know,
have
to
account
for
that
in
some
way
and
that's
what
they
talk
about
in
the
paper
here.
So
there
are
different
ways
of
doing
these
different
methods,
and
so
we
can
end
up.
B
We
end
up
with
this
fat
tail,
which
is
down
here,
and
this
is
basically
tells
us,
like
the
intensity
of
the
hierarchy,
intensity
of
the
inequality
or
whatever
it
is
this
being
larger,
meaning
that
the
rank
is
less
sort
of
strict,
and
you
know,
maybe
in
in
a
heterarchical
systems.
This
part
of
the
distribution
is
very
large.
In
fact,
there
might
not
be
a
fat
tail
at
all.
It
might
be
very
thin,
too,
in
a
heteroical
system.
B
B
B
And
this
is,
you
know,
this
is
a
difference
between
hierarchies
hierarchies
and
all
these
Universal
statements
that
we
can
make.
There
are
Universal
things
that
we
can
find
so
discovering
Universal
relations
of
simple
observables
and
list
Dynamics
have
more
practical
benefits
than
the
lofty
goes
of
content.
Others
they
give
us
precise
language
to
describe
a
phenomenon.
B
We
know
that
this
replacement
rate
is
sufficient
to
explain
the
state
of
ranking
Dynamics.
Thanks
to
bloom,
we
can
describe
an
item
in
ranking
just
just
by
the
phase
to
which
it
belongs.
So
these
graphs
that
I
showed
you
here
are
all
you
know:
they're
they're,
all
kind
of
speak
to
this
kind
of
list.
You
know
this
list
ordering
and
and
some
of
these
things
like
that,
so.
A
B
This
reference
here
number
five
also
opens
several
new
research,
Direction
Avenues,
so
the
first
of
all
we'll
go
through
them
here.
The
role
of
memory
in
real
world.
This
Dynamic
seems
to
be
Uncharted
Territory,
so
memory
would
be
if
I
had
a
list
and
that
list
we
had
had
some
sort
of
memory.
So
it's
kind
of
like
a
Markov
process,
but
in
the
list
list
form.
B
B
B
B
B
Is
in
stark
contrast
to
the
memory
memoryless
models
of
these
references?
Second,
because
of
their
even
more
fundamental
role
for
human
organization,
it
would
be
interesting
to
study
the
Dynamics
of
hierarchies,
the
big
caveat,
of
course,
being
that
there's
no
universally
accepted
definition
of
hierarchy,
which
we've
talked
about,
and,
of
course,
we
can
contrast
it
with
heter
key
so
to
fulfill
the
promises
of
complex
system
science.
We
need
to
keep
discovering
these
fundamental
relations
of
simple
phenomena
such
as
ranking
Dynamics,
and
they
form
the
basis
for
more
complex
systems.