►
From YouTube: DevoWorm (2023, meeting #13): life's origins (real/artificial), graph models, curving and jamming
Description
Update on docs and web infrastructure for DevoLearn/DevoZoo. New models for life's origins, history and usefulness of Artificial Life models. Learning on Graphs 2023 and graphical modeling. Papers on curvatures, modeling, and jamming phase transitions. Attendees: Sushmanth Reddy Mereddy, Susan Crawford-Young, Jesse Parent, and Bradly Alicea
B
I'm,
just
explaining
to
Jesse
that
I'm
having
trouble
with
my
Linux.
B
B
I've
got
I
I
downloaded
the
latest
in
Ubuntu
and
then
I've
got
a
Windows
subs
subsystem
for
Linux.
A
I
don't
know
I
heard
that
Ubuntu
might
be
having
problems
with
the
last
build,
but
I
don't
know
it's
pretty
platform,
our
it's
very
platform
independent,
sometimes.
B
A
B
B
Because
I'm
getting
kind
of
fed
up
with
my
console,
multi-physics
yeah.
A
B
Graphics,
one
just
wants
to
give
me
Springs
to
work
with,
and
those
are
not
linear,
so
I
can't
really
use
them,
but
I'm
not
sure
I'm,
not
sure
what
to
do.
A
It's
a
robot
yeah.
So
yes,
so
we
had
a
number
of
people
apply
for
the
the
graph
neural
networks
project
for
gsoc,
and
we
had
I
think
five
applicants,
that's
pretty
good,
so
I'll
I'll
be
going
through
those
with
some
of
the
mentors
I
know
that
mayook
and
jiahang
from
last
year.
A
You
remember
Mayo,
of
course,
has
involved
in
Diva,
learn
so
they're
going
to
be
mentors
this
summer
for
that
project,
as
well
in
some
capacity,
so
we'll
be
going
through
the
applications
on
that
and
then
we'll
be
making
a
decision.
May
4th,
so
May
4th
will
have
an
official
announcement
and
hopefully
we
can
get
more
than
one
spot.
I,
don't
know
what
the
constraints
are
on
that
this
year,
so
it
should
be.
Hopefully
we
should
have
a
fun
summer
be
able
to
do
some
things.
A
I
also
got
a
message
from
Hari
Krishna
in
the
slack
news
from
last
year,
Krishna
pillai,
and
he
was
asking
about
Axolotl
data
or
data
from
the
one
of
the
specialty
microscopes
that
we
have
for.
B
C
C
Middle
of
that
I
couldn't
get
into
this
week.
Meetings
but
I
will
give
my
small
update
and
I'll
leave.
Is
it
okay,
yeah
yeah.
C
C
C
And
everything
these
are
my
updates
and
the
blog
was.
It
is
really
tricky
parts
we
are
I
am
trying
since
two
weeks
to
implement
it,
but
so
many
servers
are
coming.
I
was
using
sugar
framework
for
that
it
was
creating
some
progress.
Maybe
I'll
solve
it
by
next
weekend
and
I'll.
Let
you
show
you
a
minute.
C
A
C
A
C
A
C
Was
yeah.
A
That
was
social
month
yeah.
We
were
talking
about
doing
some
well,
he
wants
to,
you
know,
work
on
putting
blog
posts
together.
He
wants
to
do
this,
not
just
for
gsoc
but
for
like
a
longer
term
project
so
like
for
his
schoolwork,
so
yeah
I
mean
we
had
a
number
of
people.
A
Well,
we
had
some
work
on
the
website,
so
we
have
this
website
on
weebly.com,
and
that
has
a
lot
of
the
core
stuff
for
the
entire
group,
and
then
we
have
another
website
on
github.io,
which
is
a
sort
of
a
a
mix
of
things,
special
things
that
we
have
and
I'll
show
this
I'll
go
through
the
screen
share
here.
A
So
this
is
this
is
dvormai,
it's
actually
evilworm.github.io
and
we
have
things
like
what
we
call
the
Devo
zoo
and
we
have
like
a
onboard
sort
of
an
onboarding
guide
there,
as
well
as
an
onboarding
guide
on
our
GitHub.
But
one
of
the
things
we
have
is
this
Diva
worm
AI,
and
this
is
something
that
sushma
has
been
working
on
with
a
few
people,
so
we
had
this
week.
We
had
a
pull
request.
A
From
geothy
GLT
ready,
so
GOC
worked
on
this
two
weeks
ago,
issued
a
number
of
pull
requests
and
then
made
blogs,
which
is
what's
happened
today,
and
so
this
is
social
month
emerged
that
commit
yesterday.
A
So
we've
talked
about
a
number
of
models
over
the
last
few
years
and
this
kind
of
get
this
is
kind
of
a
place
where
we
can
put
some
of
these
models.
We
have
segmentation,
Tools
c
elegans
models,
so
the
Devo
learn
is,
is
you
know
that's
on
hugging
face?
That's
what
sushma
is
talking
about.
That's
where
we
have
some
of
our
segmentation
models
broken
down
into
different
types.
A
So
we
have,
let's
see
if
I
can
find
these.
This
is
going
to
send
me
back
to
default.
Okay
with
the
organization,
so
we
have
a
number
of
different
models
on
hugging
face.
A
We
have
the
lineage
population
model,
the
nucleus,
segmenter
model
and
the
membrane
segmenter,
and
so
those
are
on
hugging
face,
and
then
we
have
Diva
warm
AI,
which
is
inclusive
of
things
like
digital
basso
area,
General,
segmentation
tool,
C
elegans
model
and
so
forth.
So
there's
a
lot
of
stuff
up
here
that
might
be
of
interest
to
people.
A
A
How
are
you
that's
good,
yeah?
So
yeah,
you
know
if
you
wanna
I,
don't
know
if
you
want
to
say
anything:
if
you
don't
it's.
Okay
put
it
in
the
chat.
Anyways
I
was
just
talking
about
our
our
sort
of
where
we
put
a
lot
of
our
models
and
our
AI
models.
So
we
still
haven't
really
updated
it.
It
made
a
push
to
update
it
since
last
year,
so
we
don't
have
any
of
the
graph
neural
network
stuff
up.
We
have.
A
You
know
I
mean
we
have
a
basso
area
stub,
but
that
may
need
to
be
updated.
Certainly
some
of
this
needs
to
be
updated.
I'm,
not
really
sure
how
we
should
do
this
I
I
I'm,
not
sure
if
they
hit
geothy
and
sister
month
have
been
updating
things
but
I'm
not
really
sure
where
that's
translating
onto
these
pages.
A
So
I
think
this
one
was
updated
and
this
one
but
I
would
really
like
to
go
through
it
and
make
sure
that
we
have
like
you
know,
even
if
it's
just
like
going
through
and
making
a
video
of
like
what
we
have
for
people.
So
they
can
see
what
kind
of
resources
we
have
because
it
looks
like
you
know
you
go
through
there
and
you
don't
really
know.
A
But
anyways
that's
something
to
think
about
as
we
head
into
gsoc,
because
that's
something
we'll
be
we'll
be
doing
a
lot
of
documentation,
we'll
be
doing
a
lot
of
updating
things
and
and
then,
of
course,
towards
the
end,
we'll
be
putting
together
an
exposition
of
things
for
the
projects.
But
also
you
know
to
sell
some
of
what
we've
done
and
we're.
A
Yeah
well,
I
mean
you
know:
it's
not
I,
don't
know
how
useful
it'll
be
for
everyone,
but
just
for
like
if
you're
into
machine
learning
the
machine
learning
part
of
it.
That's
definitely
a
focus
at
one
point
we
were
looking
at
doing.
Devo
zoo
is
like
a,
and
it
still
is
it's
just
kind
of.
We
haven't
really
done
much
with
it.
It's
for
model
organisms
and
getting
secondary
data
for
model
organisms.
A
So
we
have
data,
for
you
know
things
like
splators
and
and
persophila
and
C
elegans,
of
course,
and
Nassau
area
just
to
have
all
those
things
in
one
place.
A
I,
don't
think
we
have
any
well,
we
don't
have
any
data
for
the
X
model,
but
would
I
I'd
like
to
update
it
a
little
bit
more
too,
because
we,
you
know
it
was
like.
The
idea
was
like
we
would
take
like
a
model
organism,
have
some
data
for
people
to
work
with
maybe
some
information
about
the
organism
and
then
that
would
be
like
a
place.
You
could
go
if
you
wanted
to
do
some
comparative
work.
Ideally,
so
you
could
grab
like
two
or
three
species
and
try
things
and.
A
Yeah,
we
haven't
really
spent
a
lot
of
time
on
it
in
the
recent
past.
So
it's
it's
not.
A
Yeah
I
mean
ideally
too
it'd,
be
like
if
you
wanted
to
model
something
in
an
organism.
You
know
you
need
data,
your
models
because
it
you
know
we
that's
kind
of
what
we
want
to
do.
We
want
to
encourage
people
to
use
models
across
species
or
models
for
a
certain
species,
because
not
every
species
is
going
to
use
the
same
type
of
model.
So,
like
you
know
some
organisms,
you
might
have
a
connectome
or
a
tensegrity
model
or
whatever,
but
they've,
probably
specialized
for
that
system.
So.
B
Well,
I'm
one
could
come
up
with
a
a
sphere
that
is
changing
inside
yeah.
That's
one
thing:
I
thought
of
doing
for
amphibian
embryos
because
they're
producing
basically
salt
water
and
pumping
it
into
their
their
interior
and
then
the
they
become
stiffer
like
yeah.
That's
called
the
blastocill
and
I
swear.
It's
it's
waste
material
that
they
pump
into
the
middle
of
out
of
the
glass
to
seal,
but
I'm,
not
sure
it
makes
sense.
B
It's
a
fluid
and
it's
foreign,
how
many
people
have
done
testing
on
what
salts
are
in
that
food?
A
Yeah
I
mean
that's,
that's
the
thing
too,
like
a
lot
of
the
computational
stuff.
Is
you
know,
people
I?
Don't
people
have
an
uneven
background
in
biology
and
especially
if
you're,
a
biologist,
you
have
a
pretty
uneven
background
in
modeling
and
computation.
So
it's
it's
always
like.
One
of
the
things
we
want
to
do
is
I.
Try
to
even
that
out
somewhat,
which
is
tough.
I
mean
you
know
it's
hard
to
like
biology,
is
so
many
little
different
sub
areas
that.
B
B
B
A
A
Yeah,
that's
well,
that's
kind
of
the
yeah
yeah
I
mean
I,
guess,
there's
the
anatomy
and
then
there's
sort
of
the
regulation
and
then
maybe
like.
A
Physiology
I
guess,
but.
A
B
If
it's
different
from
its
surroundings
and
quite
often
yeah,
they
are
there's,
they
can
be
stiffer
or
looser
than
than
the
surrounding
tissue.
But
more
often
it's
stiffer
negative
formula,
but
that's
not
always
the
case
and
it's
the
stuff
that
I've
been
reading.
Yeah
yeah,
mechanics
of
biology,
yeah.
A
I
think
yeah,
that's
an
interesting
kind
of
conversation
to
have
is
this:
you
know
where
do
things
fit
into
different
areas?
Where
do
the
fields
begin
and
end
yeah,
but
I
mean
so
yeah
you're,
definitely
doing
this
stuff
with
tensegrity
and
I.
Think
that's
a
great
Direction!
It's
just
you
know.
A
lot
of
the
a
lot
of
biology
is
really
Gene
Centric,
a
lot
of
the
stuff
I've
seen
anyways
and
there's
stuff.
Of
course
that's.
B
A
A
B
A
Think
that
a
lot
of
the
physics
oriented
people
are
in
like
biophysics
departments
or
in
physics,
and
it's
like
there's
a
definitely
a
split
between
like
people
who
would
assay
a
bunch
of
genes
or
get
like
ex-gen
sequencing
data
set
and
like
mine,
it
versus
people
who
would
look.
Look
at
the
physics
intently,
like
the
mechanisms.
It's
just.
B
A
difference
they're
looking
at
how
the
the
nucleus
changes
shape,
how
that
affects
gene
expression.
So
there's
a
lot
of
that
going
on
there's
at
least
I've
run
across
papers
about
that,
of
course,
I'm
more
interested
in
how
the
cell
changes
shape
and
then
maybe
it
changes
the
nuclear
shape,
but
still
on
the
other
end
of
it.
Yeah.
A
Yeah
I
think
I'm
gonna
move
on
to
some
other
features
that
I
have
here
so
last
week,
I
was
had
a
scheduling,
conflicts
I'm.
Sorry,
if
I
missed
the
apparently
a
lot
of
people,
I
didn't
send
out
a
message.
A
lot
of
people
didn't
know
what
was
going
on
so
I
apologize
for
not
making.
A
You
know
and
clear
I
mean
we
had
mentioned
it
the
week
before,
but
and
so
like
you
know,
I
wanna
go
over
some
things
now,
so
we
had
a
I
found
a
couple
things
this
week
that
are
interesting
that
were
kind
of
following
up
on
some
of
the
work
that
we've
talked
about
and
previous
weeks.
A
The
first
is
this:
article
new
model
should
Light
On
Life's
origin,
and
this
is
a
press
release
on
this
bit
of
work.
That's
I
think
it's
a
paper.
The
research
reveals
clues
about
the
physical
and
chemical
characteristics
of
earth.
When
life
is
thought
to
have
emerged,
and
then
you
know
this
is
a
area
of
early
life
research
where
people
were
doing
things
like
looking
at
astrobiology
and
other
things
to
get
Clues
on
how
life
emerged
and
evolved
started
to
evolve.
A
So
this
is
the
summary
of
this
is
researchers
from
the
University
of
Rochester
in
the
University
of
Colorado
Boulder
used
experiments
in
Zircon
chemistry
to
build
more
accurate,
computable
computer
models
of
fluids
that
act
as
Pathways
from
inner
earth
to
Earth's
surface.
So
this
is
sort
of
earth.
Science
meets
biological
science.
A
A
So
this
is,
you
know
we
we
talk
about
like
events
in
Earth
history,
like
the
great
oxygenation
events,
which
are
these,
which
was
an
event
that
increased
the
amount
of
oxygen,
the
atmosphere
and
it
enabled
life
to
sort
of
hit
a
new
Plateau.
So
you
know
it
allowed
for
new
things
to
evolve,
new
new
capacities
of
life
to
evolve
and
so
forth.
In
this
case,
you're
talking
about
sort
of
the
milieu
of
early
life,
so
early
life
was
not
necessarily
in
a
Cell.
A
It
was
maybe
in
a
vesicle
which
is
like
a
little
cavity
that
has
metabolism;
basically,
it
has
chemicals
in
it
and
it
has
to
synthesize
this
self-sustaining
synthesized
set
of
reactions.
We've
talked
about
this
in
terms
of
the
hyper
cycle,
so
you
know
this
is
the
kind
of
thing
that
people
are
interested
in,
and
so
there
are
a
couple
ways
you
can
do
this.
You
can
either
look
at
other
planets
and
say
those
three.
Those
places
are
good
candidates
for
life.
A
Let's
see
what
you
know
that
your
the
planetary
history
is
there,
the
planetary
makeup.
You
can
look
at
Earth
and
say
you
know:
how
did
life
evolve
up?
First
of
all
on
Earth
think
about
early
Earth.
A
You
know
maybe
back
in
the
proterozoic,
and
you
know
what
was
the
earth
like
then,
and
in
this
case
what
they're
doing
is
they're
saying
that
certain
metals
and
elements
have
to
be
available
at
the
time
that
life
is
synthesizing
in
order
to
sort
of
give
it
a
give
it
a
kick
start
or
give
it
the
raw
materials
to
start
so.
The
first
signs
of
Life
emerging
on
Earth
in
the
form
of
microbes
about
four
billion
years
ago.
Well,
scientists
are
still
determining
exactly
when
and
how
these
microbes
appeared.
A
It's
cleared
that
the
emergence
of
life
is
inter
intake,
intricately
intertwined
with
the
chemical
and
physical
characteristics
of
early
Earth,
and
so
they
interview
a
couple
people
here
they
talk
about
how
life
could
have
started
differently
or
not
at
all.
At
the
early
chemical
characteristics
of
our
planet
were
different,
so
you
know
a
lot
of
people
observing
Mars
note
that
Mars
may
have
had
a
similar
early
history
to
Earth
and
even
Venus
Maybe
Edison,
a
similar
early
history,
but.
C
A
A
And
so
when
hypotheses
are
proposed
for
different
origin
of
Life
scenarios,
scientists
have
generally
assumed
all
metals
were
available
because
there
weren't
studies
that
provided
geologically
robust
constraints,
a
metal
concentrations
of
fluids
from
the
earliest
times
of
birth
history.
So
you
know
we
didn't
really
have
you
know.
This
is
a
good
example
of
sort
of
how
to
build
a
model
and
what
to
include
and
what
not
to
include
you
can
include
things
as
sort
of
constants,
meaning
you
just
assume
that
they're.
A
So
this
is
a
shortcoming
of
a
lot
of
models
to
address
this
shortcoming
trail
of
McCombs
that
need
the
compositioning
characteristics
of
fluids
in
the
lithosphere,
where
the
outer
layer
of
Earth
that
includes
the
crust
and
upper
mantle.
So
this
is
actually
back
when
the
surface
of
the
Earth
was
forming
out
of
molten
lava.
A
This
is
the
lithosphere
and
it's
you
know
it's
cooling,
but
not
entirely
cooled.
So
this
is
we're
getting
really
early.
This
is
really
early
years
and
these
lithospheric
fluids
are
key.
Pathways
to
transport,
dissolved,
parts
of
rocks
and
minerals
between
Earth's
interior
and
hydrothermal
pools
in
the
exterior
World
microbial
life
could
have
formed.
So
this
is
where
you're
getting
a
lot
of
transport
of
things
to
the
surface
and
the
formation
of
bedrock
and
things
like
that.
A
You
start
to
get
hydrothermic
pools
at
the
surface,
and
this
is
where
microbial
life
could
have
first
started,
and
so
a
lot
of
things
are
being
transported
around
as
actually
we
see
all
the
time
now,
you
know
we
have
volcanic
eruptions,
we
have
erosion,
we
have
drawdown
of
different
chemicals
from
the
atmosphere
and
so
forth,
and
so
you
know
this
isn't
like
totally
irrelevant
to
Modern
Earth.
A
It's
just
that
this
is
particularly
relevant
to
earlier,
so
they
can't
really
directly
measure
the
metal
fluxes
and
and
where
they
were
present
or
absent
billions
of
years
ago.
But
you
can
infer
what
metals
and
what
concentrations
they're
at
for
your
model,
so
you
can
kind
of
figure.
You
know
again
add
this
in
as
a
parameter,
fluctuate
the
value
and
see
what
it
does
to
the
simulation,
and
so
we
can
get
this
from
geologic
information
from
billionaire
of
rocks
and
minerals.
We
can
do
experiments
in
high
pressure
and
high
temperature.
A
We
can
apply
this
in
different
ways
and
so
that
yeah
this
helps
us
understand
this
yeah
there's
a
paper
associated
with
this
parallel
column,
relatively
oxidized
fluids,
fed
Earth's,
earliest
hydrothermal
systems,
and
that's
a
recent
issue
of
science.
A
So
this
is
the
paper
here
and
so
yeah,
so
they're
using
they're
looking
at
the
chemistry
of
aqueous
systems,
all
right,
so
that's.
A
All
right
and
then
the
second
feature
is
again
on
sort
of
this
extraterrestrial
light
theme,
but
this
is
from
a
different
perspective,
so
this
is
actually
life
evolves.
Kind
of
attempts
to
create
artificial
life
evolve
too,
so
do
efforts
to
create
life
by
cooking
up
imitations
in
computers,
robots
and
molecules
linked
towards
a
universal
definition
of
biology.
A
So
this
is
a
recent
article
here
from
scientific
Americans.
So
so
the
question
here
is:
what
is
life?
A
You
know
we're
trying
to
figure
out
what
life
is,
and
so
we
can
get
down
to
these
building
blocks
that
were
around
right
at
the
earliest
points
in
Earth's
history,
you
know
with
where
we
know
life
was
synthesizing.
We
just
talked
about
that,
so
we
you
know
today,
our
knowledge
is
so
Advanced
that
we
can
precisely
manipulate
life's
building
blocks
the
DNA,
RNA
and
proteins
the
build
biological
machines
and
engineered
new
genomes.
A
Yet,
despite
all
we
know,
no
Universal
consensus
currently
exists
unless
fundamental
definition,
and
so
we
know
that
life
exists
on
Earth,
that
this
is
an
N
of
one.
So
we
know
maybe
that
life
probably
exists
on
other
planets,
but
we
don't
know
anything
about
it
really.
So
we
still
have
the
set
of
one
problem,
and
that
is,
we
only
know
of
one
way
that
life
emerged,
and
that
was
a
way
it
emerged
on
Earth,
and
so,
if
it
appeared
on
other
planets,
we
don't
know
anything
about
that.
A
There's
also
this
idea
of
panspermia,
which
is
that
life
came
from
outer
space
from
you,
know
different
other
planets,
perhaps
and
transported
via
meteorite.
But
again
we
don't
really
know
anything
beyond
what
we
see
on
Earth.
So
this
is
an
N
of
one
problem,
and
so
we
can
make
a
lot
of
predictions.
We
can
build
models
like
we
did
with
the
early
life
paper.
We
just
finished
in
the
group,
but
at
the
end
of
the
day
we
don't
know
anything
else
about
how
life
might
have
evolved.
So
in
a
life.
A
However,
people
have
been
working
on
this.
For
a
long
time,
people
have
tried
to
build
models
of
life-like
things
and
sort
of
synthesize
them
from
sort
of
a
standstill
or
sort
of
an
initial
condition,
and
so
in
a
life.
We've
done
a
lot
of
this.
A
It's
the
systematic
attempt
to
spell
life's
fundamental
principles,
either
by
studying
lifeless
natural
systems
that
exhibit
life
like
Behavior
or
by
building
artificial
systems
to
compare
against
nature
Nature's
creations.
Many
of
these
practitioners,
so-called
a-lifers,
think
that
that's
the
best
way
to
really
understand
what
place
is
build
first
and
explain
later.
That's
what
they're
proposing
as
a
way
to
get
at
this
problem.
So
you
build
a
model.
It's
a.
A
A
A
Complexity,
scientists
at
the
University
of
Tokyo
is
tired
of
such
complaints.
His
field
is
just
like
any
other
basic
science
and
seeks
knowledge
or
knowledge's
sake,
so
asking
about
the
point
of
a
life
might
well
be
missing
the
point.
The
existence
of
a
living
system
is
not
about
the
utility
of
anything.
Some
people
ask
me.
So,
what's
the
Merit
of
artificial
life,
do
you
ever
think?
What
is
the
Merit
of
your
grandmother?
What
is
the
Merit
of
your
dog?
So
basically
you
know
it's
not
really
easy
to
to
say
with
the
Merit
of.
A
Like
you
know,
studying
life
is
other
than
to
study
life
and
it's
a
pretty
big
question.
So
there
are
a
lot
of
ways
you
could
do
that.
So
you
know
this
sort
of
field
of
artificial
life.
You
can
study
it.
You
can
look
at
like
the
origins
of
life,
which
is
nice.
It's
it's
a
sort
of
a
trendy
or
interesting
topic,
but
there
are
a
lot
of
other
things
you
can
do.
You
can
create
novelty,
generators
or
systems
of
endless
capacity
for
creating
novelty.
A
You
can
do
other
kinds
of
things
like
look
at
a
more
focused
problems
in
evolution,
so
this
is
something
that
you
know
can
stand
in
contrast
to
AI
So.
Currently
in
AI,
you
can
build
these
monstrous
deep
learning
systems,
but
at
some
point
these
systems
can't
learn
anymore,
and
so
what
does
it
take
for
a
system
to
continue
to
learn?
Nobody
knows
so.
This
is
where
you
know.
We
have
this
sort
of
idea
that
in
AI
we
have
to
train
our
models
of
data
and
that's
the
only
way
I
can
really
learn.
A
But
it's
also
the
only
way
it
can
continue
or
sustain
itself.
So
you
train
it
with
data.
Then
you
have
you
know
you
asking
a
question
or
you
have
a
set
of
outputs
and
that
set
of
outputs
is
based
on
basically
the
training
set,
and
then
the
testing
set
and
identifying
things
in
the
testing
set
or
responding
to
a
prompt
that
you
give
it
so
in
a
life,
things
are
more
combinatorial.
A
You
know
you
can
generate
new
possibilities
based
on
sort
of
the
structure
of
the
system.
For
example,
you
can
have
a
set
of
genes
that
mutate
and
those
mutating
genes
can
produce
things
that
you've
not
seen
before.
So
it's
really
truly
better
in
the
generative
sense
than
a
lot
of
AI
models,
because
it
allows
you
to
generate
new
combinations.
A
That's
not
like
say
what
you
would
find
with
a
large
language
model
where
you're
just
combining
things.
You
know
we
don't
really
understand
how
it
combines
things
aolife
models.
We
actually
would
understand
how
it
recombines
things.
There
are
rules
for
understanding
how,
for
example,
genes
with
mutations.
You
know
there's
a
mutation
rate.
For
example,
there's
there
are
different
operators
like
crossover,
we
can
apply
and
then
you
can
actually
measure
the
output
of
the
model
at
some
point
can
compare
it
to
the
initial
condition.
A
So
we
have
methods
for
understanding
what
the
changes
look
like,
and
so
you
know
this
is
a
nice
article
kind
of
goes
through
the
history
of
the
field
of
a
life.
It
started
in
1987
at
the
first
a-life
conference,
and
you
know
kind
of
goes
back
to
the
you
know
to
the
middle
of
the
20th
century,
where
people
were
trying
to
understand
how
machines
can
think
and
replicate
and
is
actually
John,
Von,
Neumann
and
stennis
they've.
A
You
know
they
worked
on
these
kind
of
problems
together,
building
things
like
the
Von,
Neumann
machine
and
other
types
of,
or
the
Von,
Neumann,
replicator
and
other
types
of
things,
and,
of
course,
this
evolved
into
the
modern
a-wave
systems
that
we
have
now.
So
this
is
a
animation
of
the
progression
of
a
breeder
which
is
highlighted
in
red.
A
Breeders
are
a
class
of
cellular
automata,
so
this
is
on
a
cellular,
automata
grid.
This
is
where
you
get
different,
behaviors
from
a
single
set
of
initial
conditions
to
kind
of
revolve
across
this
grid.
You
know
a
lot
of
Conway's
Game
of
Life
involved
movement
and
evolving
movement
in
different
directions
and
different
types
of
behaviors
and
then
characterizing
the
behaviors
in
some
way.
A
So
this
is
a
an
example
of
Conway's
Game
of
Life.
This
came
in
around
1970
and
then
you
know,
we've
been
working
on
things
such
as
you
know,
soft
a-life,
which
is
software
like
this.
That's
generative
that
can
produce
things
that
look
lifelike
and
then
hearty
wife,
which
is
the
creation
of
robots
and
other
types
of
Hardware.
A
A
A
A
nice
article
kind
of
goes
through
a
lot
of
the
history
of
the
field,
and
it
goes
through
some
of
the
applications
and
I.
Don't
know
if
there's
anything
more
to
say
about
it,
I
guess
you
know
they
just
kind
of
give
a
critique
of
the
field,
and
you
know
maybe
that
it's
doomed
to
succeed,
but
I,
don't
know.
If
that's
true
I
mean
you
know,
that's
that's
up
to
people
who
want
to
do
things
with
it.
I
think.
A
But
you
can
test
a
lot.
You
can
test
a
high
evolutionary
hypotheses
using
a
wife.
You
can
do
a
lot
of
things,
so
you
can't
necessarily
increase
the
n
s,
but
you
I
mean
the
N
of
observables,
but
you
can
increase
so
you
can
increase
the
end
of
observables
if
you
think
that,
like
a
simulation
is
equivalent,
what
we
see
in
the
real
world-
but
you
know
it's
it's
it's
again
we
have
to.
We
have
to
have
this
caveat
of
simulations,
aren't
really
like
the
real
world.
We
kind
of
control.
A
What's
in
the
simulation,
and
that's
that
changes,
what
we're
observing
in
these
simulations.
So.
C
A
So
it's
a
good
article,
it's
from
Scientific
American.
It's
life
evolves.
A
A
B
No
I
I'm,
just
well
just
listening.
A
A
We
had
learning
on
graphs
in
22..
This
is
something
we
thought
about
submitting
to,
but
weren't
able
to
submit
to,
but
it
was
this
was
the
inaugural
conference
learning
on
graphs,
and
so
this
was
a
nice
conference.
I
had
a
lot
of
this
is
actually
the
biddy
or
well
that's
the
the
YouTube
channel
of
their
talks,
but
this
is
the
learning
on
graphs
for
23..
So
this
is.
B
Can
you
see
it
on
YouTube
yeah.
B
A
Yeah
I
mean
this,
so
this
is
graph
neural
networks
more,
but
this
is
you
know.
The
topics
range
from
sort
of
machine
learning
to
some
topics
in
mathematics.
So
last
year
was
really
interesting
because
they
had
you
know
mathematicians
and
they
had
computer
scientists
and
they
had
machine
learning
people
they
had
modelers.
So
I
I
don't
know
what
qualifies
as
a
a
good
submission,
but
it
the
submission
deadline,
is
coming
up.
A
I
think
this
summer,
sometime
so
we'll
have
to
Target
it,
but
the
conference
is
going
to
be
at
the
end
of
November
and
it's
gonna
be
a
nice
opportunity.
Even
if
we
don't
get
anything
accepted
from
the
group
to
do
like
you
know
to
attend.
Last
year,
I
did
a
blog
post
on
it.
Where
I
I
went
over
a
number
of
different.
You
know
areas
of
interest
that
that
were
out
there.
A
They
they
laid
out
I,
think
the
field
very
nicely
they
laid
out
what
they're
interested
in
so
these
graph
neural
networks
are.
You
know,
neural
networks
that
have
some
sort
of
geometric
structure
to
them.
So,
instead
of
being
like
just
these
layers
or
these
hairballs
or
these
circular
networks,
they
have
some
structures.
This
is
not
a
good
example
of
what
a
graph
neural
network
is.
Basically,
if
it's
any
piece
of
data
that
you
have,
that
has
some
graph
structure
to
it
and
graph
structure
is
Loosely.
Defined
people
apply
graph.
A
You
know
graphs
to
a
lot
of
host
of
problems,
but
basically,
if
you
have
a
graph
structure
in
your
data,
you
should
be
able
to
extract
it
using
a
graphql
network,
build
a
some
sort
of
representation
of
it
or
embedding
and
then
do
computations
with
it.
So
it's
really
there's
some
really
interesting
domains.
Although
the
domains
focused
mostly
on
molecular
biology
and
medicine
and
then
on
other
problem
areas,
so
there
isn't
really
a
lot
on
what
we're
doing
like
embryos.
There
wasn't
anything
an
embryos.
There
wasn't
anything
on.
A
You
know
phenotypes
things
like
that,
but
there
was
a
lot
of
stuff
on
like
characterizing
protein
structures
or
molecular
interactions,
and
so
that
that
was
something
that's
like
a
basic
standard
systems,
biology
problem
where
you
want
to
build
like
a
protein
protein
interaction.
Network-
and
you
have
you
know
you-
have
a
grass
structure
already
there.
So
then
you
just
need
to
apply
this
technique
to
it
and
you
get
a
better
graph.
A
A
So
yeah,
so
this
is
like
very
specialized.
You
know
they're
they're,
doing
like
machine
learning,
algorithms,
so
they're
writing
their
own
algorithms
and
implementing
them,
or
sometimes
they
have
I,
don't
think
they
have
any
specialized
software
packages
for
this
there's
some
python
toolboxes
that
you
can
use
to
build
these
but
they're
in
pretty
early
days.
A
So
that's
the
other
impression
I
got
is
that
it's
kind
of
like
in
some
ways
it's
like
a
wild
west,
where
they
don't
really
have
the
tools
built.
So
people
are
building
the
tools
and
seeing
what
works
they're
like
a
host
of
issues
like
how
expressive
is
a
certain
algorithm
so
like.
If
you
apply
this
algorithm
that
generates
a
graph
from
your
data,
you
know:
how
can
it
how
many
ways
can
it
express
your
problem?
A
And
you
know
some
methods
are
very,
not
very
good
at
expressivity
and
some
are
very
good,
and
so
you
know
that
means
that
you
know:
can
you
really
Express
the
entire
content
of
your
problem
using
an
algorithm?
Sometimes
it's
very
hard
like,
for
example,
if
I
just
use
like
a
matching
algorithm
for
some
problem
that
required
a
lot
of
deep
spatial
structure.
That
would
not
be
very
expressive
because
I
would
just
simply
be
finding
like.
You
know,
surface
level,
detail
by
model.
A
You
know
that
might
be
better
because
you'd
be
able
to
capture
more
of
your
data
using
the
capture
more
of
the
complexity,
so
that
that's
one
of
the
things
that
they
talked
about
I
think
is
really
kind
of
relevant
to
what
we're
doing,
which
is,
you
know
to
maybe
you
know,
propose
some
sort
of
approach
to
you
know
back
up
from
what
we've
been
doing
with
graph
neural
networks
and
say
you
know:
how
do
we
propose
to
build
like
networks
or
how
do
we
express
the
networks
that
are
in
developmental
biology,
like
in
embryos
or
in
other
types
of
systems
like
that
there
are
these
networks
that
are
in
like
sort
of
embedded
in
the
data.
A
So
you
know
we
have
these
problems
like
tensegrity.
These
problems,
like
interaction
that
are,
you
know,
unique,
because
it's
not
just
interaction
at
one
point
in
time.
It's
interaction
over
multiple
things
in
time
and
then
consider
like
what
you
know
what
what
kinds
of
things,
what
kinds
of
algorithms
maybe
could
express
that
as
opposed
to
just
throwing
something
off
the
shelf
to
the
problem.
So
you
know
part
of
that
is
what
you're,
seeing
with
some
of
the
programs
you're
using
is
that
they're
made
for
mechanical
tensegrity
instead
of
biological
tensegrity.
B
A
B
I
tried
to
call
it
and
what
was
it?
A
cable
lift
and
I
get
yelled
at
the
table
map
it
is,
but
I
don't
have
to
call
it
that
I'm
just
going
to
say
it's
hooked
up
to
its
neighbors
yeah
I,
give
up.
A
B
I
I
need
some
quotes
on
this.
Actually
because,
because
I
sell
I
mean
a
tissue,
is
repels
in
a
tissue
rely
on
their
neighbors
right
for
their
structure
and
if
they're
no
longer
connected
to
the
tissue,
they
become
an
amoeboard
cell
and
chrome
away
or
become
or
undergo
a
apoptosis,
or
something
like
I
need
to
to
quote
that,
because
that's
that's
kind
of
what's
going
on
with
this,
like.
A
A
A
B
A
You
sort
of
need
a
background
structure
for
it
in
order
for
it
to
start
it's
kind
of
like
what
they
do
with
in
bioengineering.
They
give
like
something
a
scaffold
where,
like,
if
you're
growing
like
an
organoid
or
some
tissue,
you
have
to
put
it
on
a
scaffold
and
then
the
scaffold
has
to
dissolve
and
you
have
this,
but
it's
not
you
know
not,
it
doesn't
always
work.
Sometimes
you
fail
and
the
cells
don't
grow
along
the
scaffold
in
the
right
way.
A
B
That
are
great
chemicals
in
them
and
then
sculpted
them
into
the
right
shape
so
that
the
cells
are
actually
going
around
the
corners
to
get
the
right
shape
and
and
some
of
them
work
fairly.
Well.
B
So
still
my
question:
please
set
this
organ
or
whatever
you've
made.
Is
that
the
same
type
of
cell
so
you're
getting
in
a
real
one?
That
somebody
else
say
it's
a
nose
or
an
ear?
Is
your
ears
the
same?
Does
it
have
the
same
cell
types
as
a
normal
ear?
B
Then?
How
can
you
tell
yeah
so
but
Carthage
is
pretty
good,
because
it's
normally
a
cell
that
cells
that
develop
in
low
oxygen
or
from
synovial
fluid
so
they're,
not
relying
on
a
blood
supply
per
se,
so
they're
they're
pretty.
B
Should
have
some
success
with
that
and
there
I
know
they're
always
hit
for
the
spinal
column
spinal
cord,
trying
to
recreate
that.
So
there's
a
lot
of
interest
in
research
in
that
area.
Yes,.
A
B
A
B
A
Yeah
yeah
I
explained
that
sort
of
worked
fascinating
with
it.
You
know
it's
we're
learning
a
lot
about
that.
So,
let's
see
so
I
had
some
papers
to
follow
up
on
from
the
stuff
we
were
talking
about
last
well,
we've
been
talking
about
curvatures
and
modeling,
and
things
like
that
and
so
I
have
two
other
papers,
so
this
is
Morpho.
A
This
is
the
Morpho
paper.
Actually,
this
was
on
the
archive
under
condensed
soft
matter,
they're
condensed
off
matter
repository.
So
this
is
this.
Is
the
Morpho
paper
I?
Guess
it's
like
the
seed
paper
for
the
software
or
something
this
is
yeah
from
Tufts
University.
A
It's
a
whole
group
of
people
from
Tufts,
and
so
this
is
where
they
talk
about
an
emerging
theme
across
many
domains
of
Science
and
Engineering,
because
materials
that
change
shape,
often
dramatically
determining
their
structure,
involves
solving
a
shaped,
optimization
problem
or
a
given
energy
functional
is
minimized
with
respect
to
the
shape
of
the
domain,
an
auxiliary
Fields,
describing
the
structure.
A
Such
problems
are
very
challenging
to
solve
enough
for
a
lack
of
suitable
simulation
tools
that
are
both
readily
accessible
in
general
purpose.
So
this
is
where
they
introduce
Morpho
and
it
they
demonstrate
the
versatility
by
showcasing
three
applications
of
different
areas
of
soft
matter.
Physics,
swelling
hydrogels,
complex
fluids
for
that
form,
as
spherical
droplets,
aspherical
droplets,
so
they're
oblong
shaped
or
non-pherical
to
Silk
films
and
membranes
and
advise
on
broader
uses.
A
So
you
have
these
soap,
films
and
membranes,
which
are
surfaces
that
have
like
bubbles,
and
you
know,
they're
sort
of
their
bubbles
that
have
these
edges
that
are
continuous.
So
you
have
this
network
of
edges
for
the
soap
film,
so.
A
Yeah
then
you
yeah,
so
then
swelling
hydrogels
are
just
things
that
swell
as
a
sort
of
a
surface.
So
it's
like
like
a
tissue,
growing
I
guess,
but
so
a
lot
of
these
soft
materials.
You
can
use
them
as
analogs
to
the
biology,
but
also
they
have
that
sort
of
the
same
properties
as
the
biology
like
in
terms
of
stiffness
and
things
like
that
so
yeah.
This
is
basically
a
shape.
A
Optimization
they're
using
an
energy
functional,
which
is
you
know,
physics,
basically,
a
physics
tool
to
determine
the
energy
function
and
then
to
minimize
on
that.
You
can
strain
the
energy
function
and
you
can
get
these
Minima
that
allow
you
to
solve
the
problem,
so
so
the
one
one
option
to
do.
This
is
a
level
set
methods
where
you
represent
the
free
boundary
of
the
system
as
a
contour
or
a
level
set
of
the
scalar
function
defined
in
a
higher
dimensional
space.
A
So
you
can
capture
changes
in
topology
and
handle
them
easily,
but
you
need
to
to
formulate
the
optimization
problem.
You
need
to
use
a
lot
of
sophisticated
techniques,
so
in
Morpho
they
use
the
explicit
discretization
of
the
problem
domain
and
Associated
quantities
so
from
these
Morpho
is
able
to
evaluate
the
object
function
of
Interest,
as
well
as
its
gradients
in
Hessian,
which
is
a
mathematical
operator
with
respect
to
the
mesh
and
field
degrees
of
freedom.
A
A
number
of
algorithms
for
constrained
optimization
are
then
available
within
the
Morpho
environment
and
the
approach
is
similar
in
spirit
to
the
highly
successful
surface
evolver
software.
So
this
is
33
and
34,
and
that
is
these
two
references,
the
surface
evolver,
which
is
from
the
90s
and
then
the
surface
evolver
and
the
stability
of
liquid
surfaces
also
from
the
90s.
A
So
you
can
see
this
has
a
history,
and
so
this
is,
they
were
doing
I,
don't
know
if
you
can
link
to
the
paper
looks
like
you
may
be
able
to
so
this
is
the
surface
evolver.
A
This
is
a
computer
program
that
minimizes
the
energy
of
a
surface.
Subject:
constraints:
the
surface
is
represented
as
a
cyclical
complex,
so
this
is
actually
goes
back
to
craft
neural
networks,
because
what
they're
doing
in
graph
neural
networks
is
they're?
Creating
these
graph
embeddings
and
one
thing
it
has
an
analog.
It's
analogous
to
our
types
of
the
types
of
structures
that
topological
data
analysis
would
be
interested
in.
A
So
it's
you
know
basically
taking
a
network
and
breaking
it
down
into
these
cyclical
complexes,
so
you're
finding
shapes
within
the
network
of
triangles
diamonds
and
circles
and
other
things.
So
this
is
actually
something
that
has
it's
kind
of
strangely
related
to
graphical
Networks.
A
So
the
energy
can
include
surface
tension,
gravity
and
other
forms
constraints
can
be
geometrical
constraints
on
vertex
positions
or
constraints
in
integrated
quantities
such
as
body
volumes.
The
minimization
is
done
by
evolving
the
surface
down
the
energy
gradient,
so
you're,
basically
using
the
energy
gradient
to
find
the
shape.
Look
at
the
surface
properties.
C
A
A
good
paper
for
the
origins
of
this
stuff
see
if
I
can
put
it
in
the
chat.
I,
don't
have
it
in
the
collection.
A
B
A
So
let's
see
I.
A
So
there
are
two
other
papers:
there's
this
Universe:
let's
see
unified
phase
diagram
of
reversible,
irreversible,
jamming
and
yielding
Transitions
and
cyclically
sheared
soft
sphere
packings.
So
this
is
a
paper
on
surfaces
and
phase
transitions
which
we
know
in
development
and
in
soft
materials.
We
have
these
jamming
phase
transitions
where
you
have
a
bunch
of
particles
and
they're
moving
through
a
space
and
when
you
pack
them
tightly
enough,
they
undergo
at
some
density.
A
The
undergrowth
is
jamming
transition
where
they
go
from
free-flowing
to
sort
of
jammed
into
place,
so
they
stop
moving
and
this
is
the
phase
transition.
So
it's
a
quick
transition
from
move
from
moving
to
immobile,
and
so
this
usually
happens
at
a
some
density
of
you
know
packing.
So
you
know
if
they're
flowing
in
a
pipe.
If
they're
you
know
a
sparse
packing
of
particles
flowing
through
a
pipe,
they
flow
through
fine.
If
they're
there's
a
denser
packing,
they
start
to
slow
down,
but
they
don't
it's
not
a
linear
thing.
A
They
slow
down
a
little
bit
and
then
they
stop
moving
it
some
critical
density.
So
this
is
something
that
they're
looking
at
here:
reversible,
irreversible
jamming.
We've
kind
of
talked
about
that
in
embryos
too,
where
you
get
this,
you
know,
motility
of
cells
in
the
embryo,
there's
their
flows
of
cells.
There
are
other
types
of
migration,
but
then
you
have
these
jamming
transitions
where
things
stop
moving
based
on
their
packing,
and
so
these
are
both
reversible
and
irreversible.
A
In
some
phases
of
development,
you
get
a
reversible
jamming
phase
transition
where
things
start
to
move
again,
and
so
these
are
transitions
in
development
that
aren't
necessarily
well
characterized,
but
people
see
them
in
a
microscope.
So
the
the
abstract
here
is
self-organization
and
transitions
from
reversible
to
irreversible
Behavior
of
interacting
particle
assemblies
driven
by
externally
imposed.
Stresses
or
deformation
is
of
interest
in
comprehending
diverse
phenomenons
off
matter
and
in
embryos,
but
we'll
just
say
soft
matter
for
help.
A
They
have
been
investigated
in
a
wide
range
of
systems
such
as
colloidal
suspensions,
which
is
like
a
like
mud,
or
some
other
type
of
you
know
like
liquid,
with
solids
in
it
glasses
and
granular
matter.
A
In
last,
train
phase
transitions
in
glass,
the
behavior
of
glasses
is
well
known,
so
you
know
that's
something
that
again
is
another
bottle
system.
It's
interesting
in
different
density
and
driving
regimes.
A
Such
behavior
is
related
to
yielding
of
amorphous
solids,
jamming
memory
formation
Etc,
since
this
memory,
not
like
cognitive
memory
but
like
shape
memory
and
other
types
of
like
things
that
involve
hysteresis
how
these
phenomena
are
related
to
each
other
has
not,
however,
been
much
studied
another
in
order
to
obtain
a
unified
view
of
the
different
regimes
of
behavior
and
transitions
between
them.
We
investigate
computationally
the
response
of
soft
sphere
assemblies
to
a
thermal
cyclic,
Shear
deformation
over
a
wide
range
of
densities
and
amplitudes
of
Shear
formation
or
deformation
Shear
deformation.
A
Cyclic
Shear
deformation
includes
transitions
from
reversible
to
irreversible
Behavior.
Both
unjammed
and
jammed
soft
serve
packings
well
above
the
minimum.
Isotropic
jamming
density,
which
is
this
parameter
transition,
corresponds
to
yielding
in
the
vicinity
of
the
jamming
point
up
to
a
higher
density
limit.
We
designate
this
parameter:
J,
sub,
6
and
I'm
jammed
phase
transition
emerges
between
a
localized,
absorbing
phase
and
a
diffuse,
irreversible
phase.
A
So
you
have
this
sort
of
heterogeneity
that
starts
to
emerge,
and
so,
let's
see
the
emergence
of
the
unjammed
phase
signals
the
shifting
of
the
jamming
point
to
higher
densities
as
a
result
of
annealing.
It
opens
a
window
where
sheer
jamming
becomes
possible
for
frictionless
packings.
A
So
this
is
where
you
get
this
friction
as
a
parameter.
You
know
frictionless
packings,
you
can
Shear
jamming
that
occurs
as
a
result
of
some
of
these
changes
below
J
are
two
distinct
localized
States
term
point
and
loop
reversible
are
observed.
We.
A
And
detail
the
different
regimes
and
transitions
between
them,
so
they
do
this
basically
they're
characterizing,
the
materials
they're,
looking
at
these
different
phases
of
jamming
and
unjamming
and
then
they're
putting
it
into
a
phase
diagram
which
is
see
if
I
can
get
to
the
phase
diagram.
So
this
is
well.
This
is
this.
This
is
not
a
stress
stream
curve.
These
are
strain
amplitudes.
A
So
that
way.
B
B
A
B
A
A
A
B
A
So
I
think
the
the
key
here
is
that
there's
a
discontinuity
for
PE,
which
is
I,
don't
know
what
that
is,
but.
B
This
is:
is
it
the
circumference
of
the
cell
to
its
area?
Sort
of
some
ratio
is
that
is
that
that's
their
pee.
A
Might
be
yeah
I'd
have
to
go
back
up
into
this
point
to
see
oh,
the
potential
energy.
Actually,
oh.
A
A
B
Days,
you
get
a
phase
transition
with
with
tissue
because
it's
got
a
J
curve.
So
it's
it
goes
along
one
A
J
curve
when
you're
in
a
stress
strain
curve.
So
it
has
one
elasticity
and
then
it
changes
to
another
elasticity.
So.
B
A
So
this
is
accumulated
strain
here.
They
again
show
this
discontinuity.
Jumping
diffusivity
is
a
function
of
strain
amplitude.
So
again
this
this
feasivity.
You
know
how
easy
it
is
for
the
things
to
move
around
based
on
a
functional
strain
amplitude.
How,
however,
they
Define
straining
amplitude
exactly
and
you
have
these
different
phases.
So
this
is
B
here,
plot
of
the
non-a-fine
path,
length,
L
and
the
percentage
of
new
collisions
C
Nu.
So
this
is
c
new.
A
This
is
L
which
clearly
differential
differentiate,
all
the
different
phases
and
all
densities,
and
so
you
have
these
different
phases,
observing
phase
on
Jam
phase
and
yielded
phase,
so
the
phases
are
actually
colored
in.
So
this
is
down
here.
This
is
a
the
star
is
yielded
I.
Think
the
star
is
up
here
and
unjammed.
His
his
blue
triangle,
which
are
down
here
so
that's
absorbing
and
unjammed
or
down
here
yielded,
is
up
here.
A
Then
this
is
another
one
where
you
see
regions
and
the
phase
diagrams.
So
this
is
where
you
have
yielded.
This
is,
let's
see
a
is
yeah.
It's
not
really
label
them
here,
but
you
can
see
that
they're,
these
different
phases
along
this
in
this
phase
diagram
here.
So
these
different
regions
that
represent
different
states
of
this
system.
B
A
Yeah
they're
they're,
you
know
nice
to
look
at.
You
can
like
kind
of
have
an
index,
but
you
have
to
sort
of
predict
for
all
these
different
thoughts,
so
you
have
to
do
like
two
parameter
sweeps
to
set
it
up.
You
have
to
do
this
parameter
in
this
one
and
see
a
different
values
with
the
value
you
know,
and
then
you
can
Define
these
lines.
These
boundary
lines.
A
I
mean
those
are
yeah
they're
useful
I
mean
they.
They
do
give
you
a
lot
of
information
about
sort
of
the
transition
points,
but
they're
you
know,
they're,
not
precise
I
mean
if
he
were
to
go
and
test
the
system
for
every
value
you
may
or
may
not.
This
bit,
Audrey
mirror
may
not
be
like
the
thing
you
come
up
with,
but
it
basically
tells
you
where,
in
that
space
you
get
a
certain
phenomena
so
yeah.
So
this
is
more
of
the
same.
So
this
is
basically
you
know.
A
There
are
a
lot
of
connections
to
surfaces,
biological
surfaces
and
that
study
they
didn't
look
too
much
at
the
biology.
They
looked
at
these
different
soft
materials
in
this
paper,
they're
attacking
my
jamming
on
curved
surfaces,
so
I
think
Susan
mentioned
Timothy
Atherton,
maybe
two
weeks
ago.
Okay,
so.
A
Yeah
yeah
one
of
the
Morpho
people,
so
this
is
so
the
abstract
here
reads:
colloidal
and
other
granular
media.
They
experience
a
transition
to
rigidity
known
as
jamming
if
the
fill
fraction
is
increased,
Beyond
a
critical
value,
so
the
fill
fraction
I
guess,
is
how
much
is
in.
In
a
space,
the
resulting
Jam
structures
are
locally
disordered
bear
applied
loads
inhomogeneously,
which
means
that
there's,
a
lot
of
you
know
heterogeneity
in
terms
of
how
they
compare
loads
across
the
jam
possess.
A
The
minimal
number
of
contacts
required
for
stability
and
elastic
properties
that
scale
differently
with
volume
fraction
to
crystalline
media.
So
this
is
again
I
guess
this
part
here
is:
it
possesses
the
minimal
number
of
contacts
for
stability.
They
scale
differently
with
volume
fraction,
meaning
that
they're
I
guess.
The
number
minimal
number
of
contacts
increases
not
linearly
with
the
size
of
the
jam,
I'm,
not
really
sure
but
anyways.
A
Here,
the
jamming
transition
is
studied
on
a
curved
ellipsoid
Surface
by
computer
simulation,
where
shape
Evolution
leads
to
a
reduction
in
area,
routing
the
particles
and
preventing
further
revolution
of
the
surface.
So
the
particles
get
jammed
into
a
smaller
area
and
then
they
can't.
A
This
kind
of
constrains
the
surface.
The
arrested
structures
can
be
unjammed
from
the
surface.
Further
evolved,
iteratively,
eventually
leading
to
a
rigid
metric
Jam
state
that
is
stable
with
respect
to
motion
of
the
particles
in
some
specified
space
of
deformation
to
manifold.
A
So
you
get
this
jamming
when
you
constrain
the
surface
and
you
get
a
more
packed
system
and
then
you
can
actually
that's
that's
arresting
the
structure
and
then
you
can
unjam
it,
and
then
this
allows
you
to
have
these
sort
of
deformations
as
a
result
of
unjamming.
This
is
kind
of
like
in
a
tissue
when
you
have
tissue
formation,
your
cells,
that
you
know
start
to
you
know
double
and
then
double
again
and
you
get
more
and
more
cells
and
eventually
you
get
these
jam
States.
A
And
then
you
get
this
some
sort
of
folding
or
bending
or
deformation
of
the
of
the
sheet.
And
then
you
get
this
so
you
get
morphogenesis
as
a
result
of
this
sort
of
need
to
move
to
a
different,
functional
state.
So
you
get
this
jamming
and
then
you
need
to
unjam
it,
so
you
get
folding
and
that
allows
it
to
unjam.
A
The
structures
obtained
are
compared
with
those
obtained
in
flat
space.
It
is
found
in
the
jam.
States
and
curved
geometries
require
fewer
contacts
per
particle
due
to
the
non-linearity
of
the
surface
constraints.
In
addition,
structure
is
composed
of
soft
particles
are
compressed
above
the
jamming
point.
It
is
observed
that
relatively
well
ordered,
but
geometrically
frustrated,
monodispersed
packings
share
many
signatures
of
disorder
by
dispersed
package.
A
So
that's
I'm,
not
sure
I,
understand
that
very
well,
but
they
talk
about
jamming
here
in
the
intro.
A
They
talk
about
some
of
the
powerful
tools
that
have
been
developed
to
classify
the
nature
of
jam
structures
for
hard
particles
for
Cotto
and
still
in
your
proposed,
the
taxonomy
of
jamming,
based
on
the
space
of
feasible
motions
available
to
the
constituent
particles.
So
this
is
where
they
can
move
in
different
ways
in
different
directions
and
that
sort
of
is
the
determinant
of
jamming.
A
So
a
packing
is
locally
jammed,
the
least
engine
category.
If
no
particles
are
able
to
move
all
of
the
others
are
being
fixed.
So
this
is
where
it's
every
nothing
can
move
like
a
traffic
jam.
It
is
collectively
jammed
if
no
subset
of
particles
is
movable
with
the
remainder
held
in
place.
So
I
guess
this
is
where
it's
each
cell
is
holding
the
other
place.
A
So
if
one
cell
were
to
move
out
of
the
way
or
be
removed,
you'd
get
this
Avalanche
of
cells
that
start
moving
in
response
to
that,
then
it
is
strictly
jammed
if
no
Collective
subset
of
the
particles
can
be
moved
at
the
same
time
as
a
volume
conserving
deformation
of
the
container.
A
So
this
is
where
you
can't
move
the
particles
at
all
and
then,
as
long
as
you
conserve
the
volume,
if
you
move,
if
you
you
moved,
if
you
change
the
volume
like
if
it's
a
pathway
or
if
it's
like
a
an
egg
and
you
increase
the
size,
then
the
cells
can
move
out,
if
not
they're
not
so
these
are
their
two
three
different
types
of
jamming
and
they're
kind
of
technical
definitions,
but
they're
kind
of
important
to
development.
A
A
In
contrast
to
hard
particle,
packings,
packings
of
particles
of
soft
finite
range
potentials
can
be
compressed
beyond
the
jamming
plate,
so
you
can
compress
cells
beyond
the
jamming
point.
It's
just
that
they
don't
move
from
the
jamming
Point
Beyond,
so
they
can
be
so
hard,
packings
sort
of
follow
these
three
conditions,
and
these
are
also
maybe
applicable
to
soft
materials.
But
soft
materials
can
be
cons.
You
know
it
can
be
constricted
even
more
so
because
they're
soft
they
can
give
and
they
can
compress.
A
So
you
know
that
that's
something
to
think
about
in
terms
of
modeling
soft
materials
versus
herder
materials.
They
also
exhibit
critical
scaling
laws,
as
the
density
is
increased
above
the
jamming
point.
This
is
true
for
the
contact
number
Z,
as
well
as
the
bulk
modulus
B
and
the
shear
modulus
G,
and
these
are
all
like
different
aspects
of
shear
and
stress
and
strain.
So
you
know
these
are
things
that
soft
materials
are
going
to
have
that
hard
materials
are
not.
A
These
scaling
properties
reveal
the
non-linear
nature
of
packings
near
the
jamming
pool,
so
there's
some
non-linearity
there
that
has
to
be
account
for
as
well
in
flat
space
monodispersed
packings
in
2D.
The
two
dimensions
are
highly
crystalline
disordered
monodispersed
packings
can
be
produced
only
under
extreme
circumstances,
however,
for
packings
in
a
curved
2D
space,
which
is
where
you
have
curvature.
Instead
of
just
this
flat
surface,
that's
where
there's
packing
or
no
packing
crystalline
order
is
frustrated
by
the
surface
geometry,
necessitating
defects
in
inducing
strain
in
the
crystalline
regions
of
the
packing.
A
The
question
arises,
then,
of
how
geometric
frustration
affects
the
mechanical
properties
at
the
jamming
plane.
So
this
is
geometric
frustration
where
you
start
to
get
this
non-linear
non-linearity,
you
get
those
curved
surface.
Instead
of
the
flat
surface,
the
curving
introduces
these
instabilities
and
then
that's
that
introduces
non-linearities,
so
it
behaves
differently
when
it
when
it
folds
or
when
it
curves,
and
so
this
is
what
they're
talking
about
in
this
paper
or
about
soft
materials
jamming
on
a
curved
surface,
and
so
they
kind
of
go
through
some
of
the
math
here
and
they
talk
about.
A
You
know
all
the
different
aspects
of
this.
They
give
a
mathematical
model
for
unjamming
arrested
packings.
They
give
oh
here's
a
good
example
a
picture
of
what
we're
talking
about.
So
this
is
a,
and
this
is
B.
This
looks
kind
of
like
an
embryo,
but
it's
not
at
least
not
in
what
they're
modeling
here.
This
is
a
which
is
the
arrested
structure
produced
by
the
evolution
of
an
ellipsoid
surface
at
constant
volume
with
the
particles
colored,
but
hexatic
order
parameter.
A
So
this
is
where
we
have
this
order
parameter
from
zero
to
one
and
I
think
we
did.
You
know
there
was
a
paper
recently.
We
talked
about
with
drosophila
a
couple
weeks
back
where
they
have
this
very
thing
where
they
have
these.
You
know
cells
that
were
sort
of
in
an
unstable
regime
versus
a
stable
regime
into
the
flowing,
and
they
showed
this
kind
of
pattern
where
you
get
like
pattern
formation
resulting
from
this
type
of
packing.
A
B
A
A
Yeah
yeah
B
is
the
unjamming
motion
found
for
the
structure
using
the
linear
program.
So
this
is
where
this
is
the
unjust.
So
this
is
the
predictive
unjamming
motion.
So
a
is
where
it's
jammed
and
they're
different
or
order
parameter,
goes
from
zero
to
one,
which
is
just
basically
the
parameter.
We
use
to
say
whether
something
is
ordered
or
not.
A
There's
instabilities
in
these
regions,
where
closer
to
one
and
then
stability,
is
where
it's
closer
to
zero,
I
believe
and
then
or
it
could
be
the
other
way
around.
In
any
case,
then
this
B
predicts
the
motion
for
each
particle
that
they
would
have
to
take
to
unjam.
So
you
can
see
that
there's
differences
in
the
sort
of
the
unpacking.
You
know
it's
kind
of
like
a
Jenga
Tower.
You
know
what
part
you
need
to
pull
to
get
the
whole
thing
to
fall.
It's
not
obvious
from
the
gecko.
A
You
can
model
it
and
say:
okay,
you
remove
this
piece
and
the
whole
Tower
Falls,
because
they're
not
evenly
like
you
know
they
don't
all
contribute
an
order
in
the
same
way.
Some
of
these
are
highly
ordered.
So
if
you
remove
a
packing
from
here
or
you
remove
a
particle
from
this
part
of
the
packing,
it
doesn't
have
any
effect.
If
you
remove
a
particle
from
this
blue
part
of
the
packing,
it
has
a
bigger
effect.
C
A
It's
like
a
traffic
jam,
you
know
what's
causing
the
traffic
jam.
If
you
remove
that
you
can
unjam
the
traffic.
If
you
don't,
if
you
just
remove
cars
at
random,
you
can't
necessarily
unjam
the
traffic.
So
there
there's
that
ins,
there's
that
in
homogeneity.
There's
that
instability-
and
you
know
you
can
also
unpack
it.
You
can
move
the
cells
in
certain
directions,
and
so
that's
what
this
is
and
then
that
that
becomes
important
in
embryos.
Actually
they
do
this,
like
I,
said
this
unjamming
that
occurs
when
you
get.
A
You
know
folding
when
you
get
folding
or
curving
that
occurs
in
different
tissues.
So
this
is
something
that
is
relevant
to
that
as
well,
so
they
talk
about,
arrested,
jammings
or
arrested
packings.
They
ask
if
whether
they're
truly
jammed
or
not,
and
then
that's
and
then
a
Criterion
for
ice
iso-tasticity,
which
is
where
you
have
this
even
amount
of
stability.
A
Yeah-
and
this
goes
through
a
lot
of
stuff-
then
they
get
into
soft
particles,
so
that
wasn't
even
really
focusing
on
soft
particles.
There's
a
soft
particle
section
where
they
actually
have
an
arrested
packing
a
five
particle,
so
they
actually
have
this
Atomic
example
where
this
is.
This
looks
really
relevant
to.
A
So
this
is
a
the
linear
programming
finds
this
unjamming
motion
representing
rotation
about
the
ellipsoid,
but
this
motion
is
not
feasible
within
the
non-linear
surface
constraint,
so
these
arrows
show
where
the
center
of
I
guess
the
center
of
mass
has
to
move
in
order
to
unpack
this.
C
A
So
that's
so
they
give
a
lot
of
interesting
stuff
in
this
paper.
I
think
this
is
really
useful
for
maybe
like
understanding
what
Morpho
is
doing,
but,
moreover,
I
think
it's
really
under
interesting
from
a
theoretical
standpoint,
because
it's
really
kind
of
getting
at
this
idea
of
packings
and
why
they're
important
and
how
they
maybe
relate
to
Networks,
because
we
have
networks
here
we
have
a
network
here
in
the
in
this
five
packing
and
then
we
have
another
Network
up
in
the
in
figure
one
where
we
have
this
thing.
That
looks
like
an
embryo.
A
This
is
a
network
and
we
can
move
the
network
and
change
the
topology
in
a
way.
That's
maybe
leading
to
some
sort
of
change
or
phase
transition.
B
B
I
I,
don't
say,
I'm
still
very
interested
in
this
program
having
some
trouble
downloading
it
anyway,
yeah.
A
This
all
right.
Well,
thanks
for
attending
the
meeting
again
next
week,
we'll
be
back.
Do
you
have
any
questions?
I'll?
Send
you
those
papers,
Susan
I,
know.
A
Oh,
it
says:
oh
okay,
yeah
I
can
yeah.
I
can
put
these
in
the
in
the
open
one
Slack
for
Jesse,
so
yeah
check
that
or
I
might
put
them
in
the
other.
We
have
another
Slack
which
we
can
use
for
that,
so
I'll
get
them
to
Jesse
as
well.