►
From YouTube: DevoWorm (2022, Meeting 8): Graph Embeddings, Daisyworld, Cell Info Processing, and Regulation
Description
Updates and more information on Google Summer of Code projects, Overview of ongoing projects, journal special issue on networks, Daisyworld and rein control, niche construction, and system dynamics. Cellular information processing and regulation in development. Attendees: Susan Crawford-Young, Karan Lohaan, Jesse Parent, Richard Gordon, Jiahang Li, Ishan Shanware, Harikrishna Pillai, and Bradly Alicea.
B
B
C
D
C
B
A
B
B
A
A
G
A
B
A
C
D
D
D
With
processing
in
the
past,
so
I
was
hoping
you
know,
I
could
learn
more
about
the
project
and
you
know
get
some
more
insight
into
it
because,
as
far
as
I
know,
there
was
the
youtube
video
was
just
an
overview
and
since
I
haven't
really
attended
any
meetings
in
the
past,
I'm
pretty
new
to
this
organization.
So
I
was
hoping.
F
C
C
Exactly
why
we
have
these
meetings,
so
people
can
get
sort
of
used
to
what's
going
on
so
welcome.
I'm
gonna
actually
go
over
some
things
on
that
front
today.
I'm
also
gonna
we're
gonna
talk
about
a
special
issue,
we're
doing
for
a
journal
and
then
probably
papers
so
welcome,
jesse
and
and
dick
and
karan
to
the
meeting.
C
So
I'm
gonna
first
thing,
I'm
gonna
do
is
go
over
some
of
the
stuff
that
we
have
some
of
the
resources
we
have
for
like
code
and
some
of
the
data
sets,
although
I'm
not
gonna
get
into
the
data
set
so
much
today
and
then
kind
of
giving
you
some
background
on
some
of
the
things
we've
been
doing
and
then
I'll
talk
about
the
special
issue.
So
let
me
share
my
screen
here.
C
C
This
should
give
you.
If
you
go
to
this
effect,
it's
very
slow.
This
should
give
you
an
overview,
there's
a
pdf
here,
all
right,
I
think
there
might
be
somewhere
in
one
of
these
and
it'll.
Give
you
info
at
this
link
here.
It'll.
Give
you
information
about
this.
We
have
an
abstract
and
a
paper,
so
I
mean
that's
that's
a
paper
that
we've
done
and
if
you
you
know,
I
know
it
can
be
over
your
head,
maybe
because
it's
a
academic
paper
or
whatever.
C
But
if
you
have
specific
questions,
please,
you
know
you
can
bring
them
up
in
the
slack,
bring
them
up
here
in
the
meeting
or
whatever
we
can
answer
them,
or
I
can
answer
them.
If
it's
the
networks
work,
we
have
some
work
on
connectomes
as
well,
so
there's
the
emergent
connectome
and
there
might
be
some
other
things
in
here
that
you
might
find
interesting
or
inspiring.
C
From
that
point
of
view
now
we've
been
working
last
couple
years
and
we're
working
on
a
number
of
projects
having
to
do
with
like
deep
learning
and
machine
learning
techniques
for
looking
at
embryo
data.
So
we
have
this
diva
warm
ml
repository.
I
think
someone
might
have
asked
about
this
on
the
slack
what
this
is,
and
so
this
is
the
diva
mml
repository.
C
This
is
like
a
series
of
of
this
was
a
series
that
we
did
in.
I
think
2019,
where
we
did
a
number,
a
series
of
classes,
our
courses
on
different
topics.
So
you
know
we
did
current
topics
existing
tools,
little
five-minute
reviews
and
we
had
a
number
of
lectures.
So
this
is
the
syllabus
for
the
it's
maybe
a
little
bit
out
date
now,
because
this
field
is
moving
so
fast,
but
we've
had.
We
not
only
have
here
lectures
on
different
topics
like,
for
example,
we
have
input
data,
pre-tree
models
for
biology,
tensorflow
tutorial.
C
C
C
C
You
know
we
have
like
this
type
of
microscopy
data,
so
the
idea
is,
is
that
you
segment
cells
and
you
end
up-
you
know
using
some
algorithm
to
segment
cells
and
it
brings
a
number
of
different
or
it
brings
numeric.
Data
out
is
the
output.
What
we're
trying
to
do
in
this
project
is
maybe
use
some
of
the
deep
learning
algorithms
and
convert
them
to
numeric
values
and
then
model
that
using
these
neural
graph
embeddings
or
these
graph
embeddings
and
so
there's
a
connection
between
feature.
C
What
was
that?
Oh
feature?
Extraction,
yeah
yeah
yeah,
so
we
have
a
note
yeah.
These
are
a
number
of
tools
that
we
have
for
feature
extraction.
So
you
know
if
you
go
through
here,
there
are
a
number
of
data
sets.
This
links
to
we've
done
some
work
with
c
elegans,
which
is
a
nematode
which
has
a
very
it's
a
very
good
system
to
study.
If
you
want
to
see
look
across
the
developmental
process,
there's
also
at
the
adult
c
elegans
here,
then
you
have
this.
C
C
C
We're
going
to
read
on
this
diva
learn
is
here:
this
is
there
we
go
so
this
is
the
d
we'll
learn
repository,
so
this
is
a
pre-trained
model
for
using
a
deep
learning
framework
and
it
allows
you
to
do
that
sort
of
feature
extraction.
So
we
have
all
these
tools.
The
point
being
is
that
I
wanted
to
expand
upon
that
and
go
to
these
graph
embeddings
as
a
next
step.
So
these
graph
embeddings
can
be
integrated
with
oh
hello,
j
hung
they
can
be
integrated
with.
C
I
can
travel
here
with
my
camera.
Okay,
they
can
be
integrated
with
these
deep
learning
algorithms
or
they
can
stand
alone,
but
they're
supposed
to
represent
these.
These
microscopy
images,
so
we
have
say
like,
for
example,
an
image
of
an
embryo,
and
we
can
do
feature
extraction.
We
have
tools
existing
tools
to
do
that,
and
so
what
we
want
to
do,
though,
is
we
want
to
create
graph
embeddings.
C
Now
there
was
another
question
on
the
slack
about
whether
this
is
a
research
tool
or
an
engineering
tool,
so
whether
it's
like
something
that
we'll
just
use
for
like
extracting
things
or
whether
we
want
to
build
a
pipeline,
and
so
the
answer
is,
is
that
nominally?
We
want
it
to
be
a
research
tool.
We
want
to
just
build
some
sort
of
library
of
graph
embeddings
based
on
some
of
these
data
sources
that
we
have
and
I'll
make
the
data
sources
available
in
the
slack.
C
I
don't
have
them
at
my
fingertips
right
now,
I'll
just
send
you
the
links
to
some
of
the
data
that
we
have.
We
have
some
standard
data
sets
that
we've
used
for
some
of
these
other
projects.
You
know
they
include
embryo
genetic
data
sets
like
in
c
elegans.
We
have
some
other
data
sets.
Some
of
them
are
easier
to
work
with.
Some
of
them
are
harder
to
work
with.
So
this
is
something
you
can
decide
on
what
you
want
to
use.
C
C
I've
hit,
you
know,
we've
had
students
go
out
and
find
other
data
sets
that
are
publicly
available,
and
so
I
can
put
together
a
list
of
data
sets
that
you
would
want
to
use
because
that's
going
to
be
an
important
part
of
this
is
to
get
data
sets,
and
I
don't
know
you
know,
maybe
using
the
right
data
sets.
You
can
build
better
embeddings,
but
you
know
I
don't
know.
I
don't
know
what
the
answer
is.
C
C
I
want
people
to
sort
of
think
about
the
best
answer
for
so,
if
you're
looking
to
do
this
project
know
that
we
have
these
resources,
we
have
the
different
types
of
machine
learning,
feature
extraction,
algorithms
that
exist
now,
so
you
can
actually
build
those
into
your
project
or
you
can
go
and
you
know
sort
of
develop
on
your
own.
You
have
data
sets
that
are
available
and
then
that
diva
or
ml
repository-
and
I
want
to
go
back
to
that-
because
it's
kind
of
important-
I
think
from
educational
standpoint-.
C
C
It's
like
we
have
game
theory
and
machine
learning
and
development,
and
that's
a
you
know,
a
lecture
where
we're
tying
machine
learning
to
game
theory
and
development
and
kind
of
talking
about
gans,
and
things
like
that.
So
there's
a
lot
of
that
sort
of
thing
if
you're
interested
and
that
might
guide
you
as
well
in
terms
of
the
types
of
approaches
you
want
to
take,
or
you
know
the
types
of
things
we're
thinking
about,
because
we
don't-
I
mean
we're
taking
we.
This
isn't
something.
C
That's
you
know
it's
sort
of
the
typical
approach
where
we're
a
multi-disciplinary
group
and
our
our
goals
are
mainly
for
research,
but
we
also
have
an
interest
in
building
pipelines,
but
I
think
for
this
for
the
gsoc
project,
I
think
the
first
priority
is
sort
of
a
research
orientation
where
you
know
you're
trying
to
build
these
graph
embeddings
for
this
sort
of
set
of
tools
that
we
already
have.
So
it
kind
of
fits
into
that
and
then
you
know
we
might
be
interested
in
building
a
pipeline.
C
He
is
working
for
this
company
and
he's
working
on
these
ml
pipelines,
and
you
know
they
have
a
whole
library
of
things
where
you
know
they
have
different
algorithms
and
different
pipelines.
So
we
might
actually
end
up
if
you're
interested
we
might-
and
this
could
be
beyond
the
scope
of
the
gsoc
project-
integrate
your
project
with
those
pipelines.
C
So
there
are
a
lot
of
options
here.
It
you
know
I'll,
be
in
touch
with
people
as
we
go
along,
and
I
want
to
you
know,
share
things
with
you
like
data
sets
and
things
like
that
to
help
you
get
started
on
your
project
proposals,
so
I
mean
the
first
thing
you
know
to
do
is
I
guess,
just
kind
of
think
about
a
project
proposal.
Think
about
how
you
would
you
know
how
you
would
approach
the
problem?
C
C
So
I'm
happy
to
look
over
people's
proposals
as
they
work
on
them.
I
mean,
probably
you
know,
you'd
want
a
first
draft
before
I
see
it
where
you
lay
out
the
problem,
you
maybe
do
a
little
bit
of
a
demo.
You
say
what
tools
you'll
be
using.
I
mean
if
you
want
to
use
tensorflow
or
pytorch.
That's
perfectly
fine,
you
don't
have
to.
In
fact
I
encourage
it
because
we
have
you
know
those
platforms
or
those
packages
are
kind
of
proven.
C
You
know
that
they
they
work.
So
we
know
how
you
know,
we
kind
of
know
how
they
work
and
it's
it's
more
valid,
I
think,
than
building
your
own
tools
from
from
the
ground
up.
But
then,
once
you
work
inside
those
you
know
platforms,
you
can
then
build
your
own
algorithms.
You
can
tie
them
into
other
things.
Did
you
have
any
questions.
C
This
would
probably
well
I
mean
it-
it's
probably
going
to
be
separate
at
least
to
begin
with,
although
if
you
wanted
to
build
it
into
that,
that
would
be
okay.
I
would
really.
I
think
it
would
be
probably
better
if
we
had
it
as
a
separate
thing,
though
for
gsoc,
because
they
want
to
see
like
they
want
you
to
be
able
to.
C
E
E
E
A
E
A
H
E
C
C
Now
dick
and
I
talked
a
little
bit
about
this
last
week-
I
know
susan-
I
didn't
really
didn't
get
together
on
this
as
as
as
the
three
of
us,
because
it
listed
you
as
a
editor
co-editor.
So
I
I'd
like
to
hear
your
feedback
on
this.
I
know
you've
got
the
email,
but
we
haven't
met
in
person
on
this
so
or
virtually
so.
The
title
is
approaches
the
developmental
network
structures
and
tried
to
be
very
broad
on
that
some
of
the
stuff.
C
C
Some
of
the
work
is
focused
on
like
some
of
the
things
we're
doing
with
tensegrity
and
other
types
of
topics,
so
this
will
focus
on
networks
and
the
role
in
the
representation
and
actual
actual
usage
in
the
developmental
process.
Networks
are
a
powerful
methodology
that
allows
to
analyze,
understand
and
visualize
complex
developmental
processes.
C
As
complex
systems,
developmental
processes
and
many
interacting
components
with
spatiotemporal
structure,
multiple
scales,
both
intracellular
and
extracellular,
so
to
understand
the
structure,
we
must
look
beyond
standard
correlative
and
statistical
approaches.
So
this
is
where
networks
are
useful
understanding
the
heterogeneous
structure
of
network
topology.
C
We
can
gain
a
broad
overview
of
how
the
adult
phenotype
emerges
from
a
generally
spherical
egg,
and
so
are
interested
in
things
is
wide-ranging,
is
different
machine
learning
approaches
that
use
graphs,
structural
models
of
embryophysics
tensegrity
networks,
I
guess
gene
regulatory
networks,
signaling
networks
and
then
complex
network
analysis
and
theory,
and
so
that's
that's
kind
of
the
broad
network
casting
here.
So
the
topics
include
10
security
structures,
order,
developmental
graphs,
including
lineage
lineage
trees
and
differentiation,
trees
and
graph
theory.
C
More
generally,
so
you
know
something:
that's
an
ordered
graph
that
describes
development,
divergent
developmental
network
structures,
developmental
graph,
embeddings,
embryo
networks
and
connectomics
graph
dynamics
and
temporal
rewiring
connecting
genetic
and
morphological
networks,
developmental
hypergraphs
and
role
of
horizontal
gene
transfer.
Those
are
all
possibilities,
and
so
we
have
a
couple
of
papers
kind
of
listed
here.
C
Dick
proposes
this
paper
a
call
for
a
modern
test
of
williamson
species,
fusion
hypothesis.
This
involves
a
an
evolutionary
tree,
so
this
would
be
sort
of
our
ordered
graph.
I'm
also
planning
to
contribute
a
paper.
I
still
haven't
come
up
with
a
title
on
what
it's
going
to
be,
but
I'm
going
to
put
that
in
there
and
then
I
guess
if
susan
wants
to
put
in
a
paper-
or
if
you
know
someone
who
wants
to
put
in
a
paper,
I'm
not
really
sure
how
many
papers
we
need
to
have.
C
I
emailed
the
person
who's
doing
the
special
issue,
the
organizer
of
it.
She
says
it
just
won't,
be
made
public
until
you
get
like
a
certain
number
of
papers.
But,
like
you
know,
we
we
should
send
this
to
her
to
to
give
her
sort
of
a
heads
up
about
her
intentions.
And
so
that's
that's
where
we
are
with
that.
C
E
C
C
C
B
C
Of
for
you
know,
like
diva,
will
learn
specific
things,
but
we
might
actually
use
that
as
well
for
some
of
the
some
of
these
other
things
that
are,
you
know
specific
to
these
projects,
so
I'll
I'll,
be
posting
and
I'll
be
I'll.
Be
tagging
people
and
you'll
be
getting
messages
on
this
point.
You're
still,
okay,.
C
Well,
I
think
so
I
mean
you
know,
I
know
that
stuff
he
does
with
bioelectricity
it's
basically
looking
at
cells,
I
mean
some
of
the
stuff
he
does
is
like
the
cells
are
working
together
to
sort
of
coordinate
like
regeneration
or
morphogenesis
and
they're
having
to
coordinate
their
electrical
activity.
So
that's
a
network
of
like
coordination,
network
of
some
type
where
okay.
E
E
C
E
A
C
Yes,
so
let's
see,
why
don't
we
move
on
to?
Let's
see
we
have
some,
I
wanted
to
go
over,
maybe
some
of
the
things
we
do
have
some
submissions
that
we
I
want
to
go
over-
that
I
don't
think
it's
been
updated
to
recently,
but
I'm
going
to
share
my
screen
again
and
I'm
going
to
go
to
the
our
submissions
document,
which
again,
I'm
not
sure
it's
been
updated
really
recently,
but
let's
see
what's
there,
so
we
have.
This
is
the
way
we
work
on
this.
C
You
know
we
work
in
this
way
where
we
review
these
things
at
the
meeting
and
then
we
kind
of
follow
up
on
them
and-
and
you
know
we
see
what
gets
done.
There
are
various
deadlines
for
conferences
and
books
and
other
types
of
journals.
Now
you
know
we
have
a
special
issue.
We
have
to
organize
that
as
well.
So
so
we
have
a
number
of
different
things
that
are
outstanding.
These
are
projects
that
have
been
proposed
and
you
know
we're
looking
for
collaborators
on
them.
C
C
Organisms
have
developmental
programs
from
different
parts
of
the
evolutionary
tree
and
they
get
expressed
in
different
parts
of
development.
So
this
is
something
that
could
be
tested.
If
people
are
interest,
you
know
if
they
have
a
background
in
genomics,
they
can.
You
know,
look
at
different
genomes
from
different
organisms
and
do
like
blast
searches
to
match
up
potential
genes
that
are
shared,
and
you
know
maybe
there's
some
relationship
there.
So
well.
C
E
C
C
The
acceleration
of
the
of
the
cells
moving
in
a
colony
and
you'd
have
to
be
able
to
like
detect
that
movement
you'd
have
to
break
down
the
video
of
the
micro
micro,
microscopy,
video
and
different
frames.
You'd
have
to
resample
it
and
then
determine
whether
it's
you
know
movement
is
smooth
or
whether
it
exhibits
variations
along
its
path.
E
E
C
C
Right,
yeah
yeah.
We
have
a
lot
of
the
data.
You
know
it
comes
in
video
form,
so
you
have
like
a
number
of
different
images
sort
of
in
a
time
series,
and
then
you
have
to
basically
find
like
you
know.
You
have
to
find
features
of
they're
these
long
elongated
cells
and
you
have
to
find
you
know
you
do
some
sort
of
feature
detection
to
find
the
centroid
or
you
know
some
marker
that
you
can
use
to
track
cells
across
images.
C
So
the
idea
would
be
that
you'd
have
these
cells
and
they'd
you'd
be
able
to
track
them
over
time
and
see
how
they
move
and
their
displacement.
And
so,
if
the
displacement
is,
you
know,
forms
a
smooth
trajectory,
then
that's
smooth
and
it
forms
something
that's
highly
variable.
Then
that's
a
jerky
trajectory.
What
was
that.
E
C
C
That's
for
that,
then
we
have
some
work
on
quantitative
comparisons
of
archaea
and
shape
droplets,
and
this
is
something
that
maya
presented
in
our
group
back.
I
think
earlier
in
the
year
a
couple
weeks
ago-
and
this
is
something
that
is
sort
of
a
take
off
from
the
diva
learn
stuff
he's
using
computer
vision
to
characterize
different
shapes
of
droplets,
like
water,
droplets
and
archaebacteria,
and
using
sort
of
the
you
know,
feature
detection,
algorithms,
not
the
same
ones
in
diva,
learn
but
some
other
ones
to
make
those
kind
of
quantitative
assessments.
C
C
C
A
E
As
generators
of
these
patterns
raised
is
the
question
as
to
whether
or
not
the
wolfram
patterns
are
closed.
That
means
is
any
pattern
generated
by
something.
That's
like
a
wolfram
pattern,
generator
also
a
subset
of
the
obvious
patterns,
or
is
it
something
new
and
we
don't
know
the
answer
to
that?
Yet
so,
if
you've
studied
mathematics
and
you
know
the
concept
of
clothes
from
mathematics,
then
this
is
an
interesting
questions.
Whether
roll
from
patterns
are
closed
under
wilfred
rules.
C
H
C
E
Okay,
in
fact,
it
raises
a
curious
question:
if
you
look
at
the
standard
views
of
development
and
you
look
at
the
differentiation
tree
development
they're,
actually
at
two
opposite
extremes
in
the
differentiation
tree,
there
are
no
anastomoses
between
branches,
no,
no
connections
between
the
branches
in
the
standard
view.
That's
all
you've
got,
there's
connections
in
them
and
the
the
tree
doesn't
matter.
C
E
C
C
A
C
Lives
in
boston,
so
he
might,
but
I
don't
yeah
and
then
okay,
so
if
you
have
to
go
jesse
thanks
for
attending
the
meeting,
now
I'd
like
to
switch
to
our
papers,
we
do
papers
at
the
end
of
the
meeting.
You
know
I'll
be
doing
this
for
about
15
or
20
minutes.
So
if
you
have
to
leave
at
the
top
of
the
hour,
that's
okay!
I
just
wanted
to
have
enough
time
to
go
over
a
few
papers
here.
C
C
And
so,
if
you
change
the,
if
you
change
the-
and
this
is
thought
of
as
a
bunch
of
daisies
covering
the
world,
but
it
could
be
any
organism
that
covers
the
surface
of
the
planet.
If
you
change
the
color
of
those
phenotypes,
those
colors
will
change
the
albedo
and
change
the
atmosphere
if
you
have
heterogeneity.
So
if
you
have
daisies
or
organisms
of
multiple
colors,
they
change,
you
know
they
compete
for
dominance
or
they
fall
into
a
equilibrium,
and
that
also
affects
the
environment.
C
C
Polymorphic
stable
states
that
resist
perturbation
emerge
from
the
simulated
co-evolution
of
organisms
and
environment,
and
then
they
talk
about
different
state.
Multiple,
stable
states,
resulting
regulation
is
achieved
through
two
main
sub-populations
that
are
adapted
to
slightly
different
resource
values,
which
force
the
environmental
resource
in
opposing
directions.
C
So
this
this
is
basically
the
daisy
world
model
where
you
have
multiple
species
that
are
sort
of
co-evolving
and
they're,
actually
maintaining
this
dynamic
equilibrium
between
them,
so
they
don't
one
doesn't
overtake
the
other,
they
kind
of
exit
coexist
and
they
come
up.
They
talk
about
this
concept
called
niche
construction
and
so
niche
construction
is
a
newish
idea.
C
So
okay,
so
he
talks
about
how
these
organisms
you
know
talked
about
some
an
example
of.
Let's
see,
for
example,
burrowing
earthworms
change,
the
composition
of
local
soil,
where
is
the
amplification
of
silicate
rock
weathering
by
plant
life,
has
resulted
in
the
reduction
of
atmospheric
concentration
of
co2
by
10
to
100
times
without
the
latter
effect.
The
surface
temperature
of
the
earth
would
be
much
higher,
perhaps
in
excess
of
50
degrees
centigrade.
C
So
this
is
the
effect.
That's
called
niche
construction,
which
is
that
organisms
will
start
living
in
an
environment
and
modify
the
environment,
and
then
that
modification
changes
the
conditions.
So,
if
you
think
about,
if
you
know
anything
about
beavers
in
north
america,
they
do
this
ants
and
termites
will
do
this
with
termite
mounds.
They
build
large
mounds,
but
they
can
actively
change
the
environment
in
which
they
live.
They
can
change
the
properties
of
it,
and
so
they
you
know
they
can
build
structures.
They
can
do
all
sort
humans,
of
course.
C
A
C
The
organism,
so
this
is
something
that
you
know
is
they
work
out
this
this
and
they
actually
get
into
some
of
the
population
genetics
of
it.
So
we
talked
about
that
population.
Genetics
paper
a
couple
weeks
ago,
where
they
talk
about
the
population,
genetics
of
daisy
world,
and
they
talk
about
how
this
sort
of
niche
construction
actually
involves.
Some
population
genetics
that
kind
of
is
consistent
with
this
idea
of
co-evolution
and
co
sort
of
coexistence
of
multiple
polymers.
C
and
this
is
catastrophes
and
daisy
world.
So
this
is
a
little
bit
about
how
you
can
have
catastrophes
in
this
system.
So
you
know
the
system
is
what
they
call
highly
regulated.
So
you
get
this
regulation
that
emerges
from
these
organisms
kind
of
living
on
the
planet
and
changing
the
conditions
of
the
planet,
but
you
can
also
have
catastrophes
which
are
sudden
collapses
of
the
world
and
it's
in
its
environment,
and
so
you
know,
how
does
that
work?
Does
the
population
crash
or
what
happens
so
there's
this
research
tradition?
C
So
this
is
something
that
again,
it's
this
sort
of
catastrophism,
where
you
have
this
small
perturbation
and
it's
not
consistent
with
the
effect
that
it
has
on
a
system,
and
so
they
kind
of
talk
about
catastrophes
and
daisy
world.
This
is
interesting
because
you
know
if
you're
you're
thinking
about
like
something
like
cybernetic
regulation.
C
You
know
it's
always
interesting
to
figure
out
kind
of
how
these
non-linear
effects
propagate
through
the
system.
So
this
is
a
very
much
this
new
study
by
aqua
and
all
and
new
meaning
when
this
paper
was
published,
has
introduced
curvature
into
the
ca
world.
Daisy
world
ca,
daisy,
world
models,
cellular
automata,
they
build
a
daisy
world
out
of
cellular
automata
and
they're
able
to
incorporate
variation
in
the
input
of
solar
radiation
across
the
grid,
and
so
you
have
these
this
variation
in
inputs
and
these
variations
in
inputs
sometimes
lead
to
these
catastrophes.
C
So
it's
not
just
a
matter
of
like
you
know,
it's
something
you
wouldn't
expect
necessarily
can't
necessarily
predict
it,
but
you
get
this
paper
from
acquainted
all
was
entitled
catastrophic,
desert
formation
in
daisy
world
because
they
found
that
when
solar
luminosity
increases
to
a
critical
value,
a
desert
formed
across
the
wide
band
of
the
planet.
So
you
have
these
phase
transitions,
which
are
analogies
to
phase
transitions.
C
C
So
the
concept
of
rain
control
dates
to
pre-automotive
eras,
the
pre-automotive
era,
when
a
driver
would
control
their
team
of
horses
or
some
other
pack
animal
in
order
to
get
wagons
or
other
payloads
from
place
to
place,
the
driver
would
use
two
reins
or
guide
wires.
So
one
on
the
right
one
on
the
left
that
were
independently
coupled
to
the
team
of
animals
in
front
pulling
on
the
rains
in
different
ways,
would
result
in
a
very
crude
form
of
steering
controlling
direction
or
speed,
and
so
in
physiology.
C
You
can
use
this
model
as
a
model
of
control
and
of
regulations.
So
in
physiology,
brain
control
exists
when
two
entities
a
and
b,
which
could
be
gene
products,
paracrine,
signaling
or
hormonal
release,
act
upon
the
same
target.
So
the
target
being
the
the
thing
that
you're
trying
to
control
were
in
this
case,
maybe
the
horse,
and
then
you
would
pull
say
a
like
on
the
left
ring
or
pull
on
the
right
rein
or
pull
on
both
of
them
at
the
same
time,
and
it
would
result
in
some
sort
of
control.
C
It's
not
you
know
it's
not
deterministic
control,
because
there
is,
you
know
it
isn't
like
a
direct
control,
you're
just
kind
of
pulling
back,
and
you
don't
know
if
you're
pulling
back
hard
enough
in
the
physiological
context.
You
know
there
might
be
one
pathway,
that's
activated,
but
if
it's
not
activated
strongly
enough,
it
doesn't
really
have
much
of
an
effect.
Another
pathway
might
be
activated
in
the
same
way
where
both
pathways
would
be
activated
and
it
has
different
effects
on
the
system,
different
control
effects.
C
Down
so
one
of
these
rains
you
know,
doesn't
function
anymore,
and
so
you
either
can
control
the
system
with
a
single
rain,
or
you
know
you
can't
now.
Sometimes
these
rains
are,
you
know
redundant,
and
if,
in
that
case,
then
you
don't
have
a
problem.
If
one
fails,
then
you
just
use
the
other
one,
but
in
diabetes
a
and
b
are
different
things,
and
so,
when
you
lose
a
or
when
you
lose
b,
you
can't
control
the
system
very
well.
C
C
They
propose
rain
control
as
a
means
to
control
the
daisy
world
and
its
and
its
operation
and
its
stability.
So
the
inter
the
the
main
thing
they're
interested
in
here
is
stability
of
a
system,
and
what
does
it
take
to
maintain
that
stability
over
time?
And
so
you
know
you
can
use
a
lot
of
different
approaches.
C
You
can
use
differential
equations,
you
can
use
other
types
of
things,
but
at
the
end
of
the
day,
it's
really
just
about
making
sure
that
that
stability
is
maintained
and
in
many
cases
and
we
can
draw
from
complex
systems
or
we
can
draw
from
control
theory
or
cybernetics.
We
see
that
they're
different.
You
know
there
are
different
ways
that
can
happen
in
different
ways
that
that
can
change,
and
so
I
thought
that
was
interesting
of
interesting
follow-up
on
that
simple
model.
C
It's
morphogenesis
is
bayesian
inference
a
variational
approach
to
pattern,
formation
and
control
and
complex
biological
systems.
So
this
variational
approach
is
something
that
carl
fristen
has
advanced
they're
doing
this
stuff
with
what
they
call
active
inference,
which
is
a
framework
to
look
at
like
how
cognition
emerges
in
the
brain
or
maybe
not
even
in
brains,
but
it
may
be
in
other
complex
systems,
and
you
know
it
uses
things
like
free
energy
and
other
types
of
mathematical
tools
to
sort
of
get
at
this
question,
and
so
it's
kind
of
an
interesting
thing.
C
It's
one
of
these
things
where
it's
almost
like
active
inference
is
almost
like
a
where
you
have
a
hammer
and
you're
looking
for
a
nail,
but
it
does.
It
is
actually
quite,
I
think,
informative
in
a
lot
of
ways
it
you
know,
so
they
wrote
this
article
on
what
happens
in
embryos
and
and
so
they
go
through
this
in
the
abstract.
They
talk
about
the
wonders
of
modern
biology,
and
so
we
are
now
getting
more
increasing
detail
about
molecular
mechanisms
underlying
development.
A
C
Regeneration
and
biological
organisms,
however,
an
overarching
concept
that
can
predict
the
emergence
of
form
and
the
robust
maintenance
of
complex
anatomy
is
largely
missing
from
the
field.
So
this
broad
like
whole
embryo
approach.
You
know
explaining
how
you
get
the
emergence
in
form
of
anatomy,
so
they're
not
interested
in
gene
regulatory
networks,
they're
interested
in
sort
of
these
massive.
You
know
these
large
scale,
maybe
networks
or
other
things
that
are
going
on
classical
approaches
such
as
least
action
principles
are
difficult
to
use
when
characterizing
these
type
of
systems.
C
These
are
open
and
far
from
equilibrium
systems.
These
are
these
predominate
in
biology
and
they
are
very
different
from
the
types
of
things
that
you
would
see
in
say
like
chemical
in
chemical
kinetics,
for
example,
in
this
neurobiology
setting,
a
variational
free
energy
principle
is
emerged
based
on
a
formulation
of
self-organization
in
terms
of
what
they
call
active
bayesian
inference.
C
So
they
use
this
approach,
which
is
a
bayesian
approach,
which
is
where
you
calculate
conditional
probabilities,
and
you
calculate
these
things
and
you
try
to
figure
out
the
world
from
that
approach
and
bayesian
inference
is
actually
a
way
to
like
look
at
evidence
and
evaluate
it.
If
you,
if
you're
interested
in
more,
we
can
go
over
bayesian
methods,
we
have
some
tutorials
and
bayesian
methods,
but
but
you
know
that
this
is
a
thing
that
you
can't
really
do
in
like
a
single
afternoon.
You
really
need
to
dig
into
that,
but.
C
Inference
here
is
the
point:
the
free
energy
principle
has
recently
been
applied
to
biological
self-organization
beyond
the
neurosciences.
So
now
they're
going
into
these
processes
of
development
and
regeneration,
the
bayesian
inference
framework
treats
cells
as
information
processing
agents.
So
each
cell
processes
information
where
the
driving
force
behind
morphogenesis
is
the
maximizations
of
a
cell's
model
evidence.
C
So,
basically,
the
cells
are
processing
information
and
each
cell
is
taking
evidence
from
the
environment
and
they're
maximizing
the
evidence
of
the
environmental
signals.
So
it
could
be
like
signals
like
chemical
signals.
It
could
be
signals
from
other
cells
and
it's
maximizing
its
evidence
of
what's
going
on
and
then
it's
making
the
appropriate
decision
should
I
differentiate.
C
C
This
is
realized
by
the
appropriate
expression
of
receptors
and
other
signals
that
correspond
to
a
cell's,
internal
or
generative
model
of
what
type
of
receptors
and
other
signals
it
should
express,
so
cells
express
receptors
all
the
time
in
development
and
in
adulthood
and
they're.
The
fact
that
they're
expressing
receptors
is
sort
of
almost
like
a
chicken
and
egg
thing
in
some
ways,
because
it
has
to
express
receptors
that
then
it
needs
to
receive
information.
C
C
This
provides
us
with
a
quantitative
formula:
isn't
formalism?
So
it
gives
us
the
set
of
equations
that
we
need
to
understand
this
information
processing.
They
derive
a
lot
of
mathematics
behind
bayesian,
inference
and
cells
in
this
paper,
and
then
they
do
some
simulations
to
show
this
formalism
can
reproduce
experimental
top-down
manipulations
of
complex
morphogenesis.
C
So
this
is
a
first
principle
approach
to
morphogenesis
and
they
also
consider
aberrant,
signaling
and
functional
behavior
of
single
cells,
and
then
they
also
look
at
like
first
steps
of
carcinogenesis,
which
is
the
formation
of
cancer
cells,
and
they
think
of
they
interpret
that
as
a
set
of
false
beliefs
about
what
a
cell
should
sense
and
do
so.
They
have
this
model
of
integrating
information
in
cases
where
you
have
cancer
or
you
get
on
this
cancer
pathway.
C
Where
cells
differentiate
from
like
some
cell,
it
could
be
a
differentiated
cell,
it
could
be
a
stem
cell
into
a
cancer
cell.
This
is
based
on
a
set
of
false
beliefs,
about
what
the
cell
should
be
expressing
it's
expressing,
maybe
the
wrong
receptors
or
it's
integrating
its
information
wrongly.
So
this
is
kind
of
a
weird
way
to
look
at
it
in
some
ways,
because
it's
it's
really
kind
of
projecting,
maybe
what
humans
are
doing,
but
it's
also
kind
of
a
unique
approach
because
it
does
give
you
this
sort
of.
C
I
know
people
in
cell
biology
will
use
terms
like
you
know,
signaling
and
then
they'll
kind
of
get
really
anthropomorphic
about
it
and
kind
of
put
it
in
the
framework
of
humans
exchanging
messages.
So
you
know
I
don't
know
if
that's
useful
or
not,
but
it's
you
know
it.
At
least
it
fits
into
that
tradition.
C
A
C
C
Might
be
a
problem
for
some
people.
These
are
examples
of
their
simulations,
so
they're
simulating
the
cell
from
the
head,
the
intestine,
the
pharynx
and
the
tail
and
they're
driving
this
from
a
flatworm,
which
is
something
that
levin
works
on,
and
they
show
that
this.
This
patterning
happens
in
in
flatworms,
where
you
have
this
programmatic.
You
know
you
have
this
organization
from
head
to
tail,
it's
very
much
regular
across
development,
and
you
get
this.
They
were
able
to
take
in
flat
worms.
C
You
can
do
these
interesting
experiments
where
you
can
dissect
out
pieces
of
the
flatworm.
You
can
put
them
on
a
plate
and
you
can
get
a
whole
new
flatworm,
because
all
the
cells
are
toady
potent
they
cannot.
Each
cell
can
generate
a
whole
new
organism,
so
they're
actually
able
to
look
at
some
of
the
prediction
and
error
of
each
cell
as
it's
regenerating.
Does
it
end
up
in
the
right
place,
and
so
they
can
show
that
this
is
exactly
what
happens.
C
So
they
do
things
like
this
and
they
use
a
model
organism
which
is
actually
amenable
to
this
approach.
So,
if
you're
looking
at
an
even
if
you're
looking
at
c
elegans,
it
might
not
necessarily
be
the
case
that
this
is
what's
going
on
so,
but
I
thought
that
was
an
interesting
paper
since
we
brought
up
levin
looks
like
we
have
some
messages
here
and
we
have
people
left.
Ishan
said
thank
you
for
getting
in
touch
on
the
project.
Susan
left,
thank
you
for
attending
susan
and
nishan
and
derivative
integral
control.
C
A
Oh
I'm
still
here,
the
derivative,
integral
control
is
just
a
standard
method
of
controlling,
say
a
large
plant
or
something
both
the
the
control
formula.
And
then
I
do
a
derivative
integration
of
of
the
resulting
signal.
That's
coming
out
to
try
to
to
keep
the
plant
under
control.
So
it's
not
exploding.