►
From YouTube: DevoWorm (2021, Meeting 27): GSoC #8, Developmental Agents, Thermotaxis, Cell Regulatory Networks
Description
Update on Summer of Code activities (Mainak Deb, Improving DevoLearn), Review of submissions document and group-wide task board. Developmental Agents, Germ Layers, and Differentiation Trees. Thermotaxis and environment in C. elegans, Cell Regulatory Networks, Flexibility of Turing Morphogenesis. Attendees: Bradly Alicea, Shruti Raj Vansh Singh, Sanjay from Amrita University, Jesse Parent, and Mainak Deb.
A
C
A
C
C
To
basically
convert
the
pth
file
to
this,
and
then
after
that
I
started
to
work
on
like
I,
I
started
to
work
from
scratch
again
on
the
devol
and
gmail.
At
this
time
I
was
using
another
library
called
streamlight,
which
is
which
is
very
similar
to
radio,
but
it
has
more,
but
it
has
more
capabilities.
C
C
C
C
C
C
A
A
D
C
C
C
D
A
Sounds
very
good
all
right
nice,
so
that
looks
pretty
good.
Thank
you
for
presenting
that
and
we'll
be
updating
on
that
next
week
too.
So
we
just
keep
going
and
welcome
sherdy
and
jesse.
Thank
you
for
attending.
Did
you
have?
Did
she
or
jesse
have
anything
to
add?
I
know
that
trudy
just
finished
her
neuro
match
course
yeah.
So
how
was
it.
E
E
F
B
F
E
D
F
A
So
they
were
looking
at
fmri
data
and
the
alex
net,
and
they
are
looking
at
how
the
different
layers
of
alexnet
correspond
to
the
different
parts
of
the
visual
stream,
and
it
turns
out
that
there
are
a
couple
groups
doing
a
similar
thing.
So
there
are
a
couple
groups
that
took
different
approaches
to
that
same
type
of
question
and
it
was
really
interesting
to
watch
like
some
of
the
results.
A
You
know
some
people
were
looking.
Some
groups
were
looking
at
like
the
nature
of
the
input
data
and
how
that
corresponds
to
what's
going
on
in
the
brain
and
then
those
activations
what's
going
on
in
in
a
deep
net.
So
you
could,
you
know
they
had.
They
have
a
type
of
analysis
called
representational
similarity
analysis,
that's
what
they
were
typically
using
and
analyze
the
data.
So
you
just
take
each
level
of
the
deep
net
you
take
each
part
of
the
brain,
each
like
they're.
A
You
know
what
they
call
regions
of
interest
and
then
you
just
do
a
bunch
of
correlations
to
see
where
there's
more
correlation
in
the
output.
So
you
know,
like
the
correlations,
are
pretty
low.
They
don't
usually
ever
get
above
like
0.35,
which
is
not
great
for
correlation,
but
you
know
you
can
see
differences
in
the
correlations.
A
We
don't
really
know,
but
we
know
that
like
there's
some
interesting
relationship
there,
and
so
I
thought
that
was
really
interesting
and
then
there's
so
many
different
deep
net
models
out
there.
It's
it's
interesting
to
see
how
the
results
varied
between
the
different
models.
So
you
have
alex
net
and
you
have.
I
can't
remember
some
of
the
other
ones
that
they
tested,
but
there
are
a
number
of
different
models
that
were
tested
and
they
all
have
a
little
bit
different
results.
A
A
B
B
A
Yeah,
so,
let's
see
today
we're
going
to
talk
about,
I
want
to
go
through
some
things
to
talk
about
some
papers.
Talk
about.
I've
been
promising
to
do
this
for
a
while,
but
I'm
gonna
talk
about-
and
this
is
something
I
do
with
my
other
group-
how
we
can
take
agent,
computational
agents
and
conceive
of
them
as
like,
a
a
series
of
germ
layers.
A
A
A
In
the
hermaphrodite's
case,
it's
going
to
have
both
eggs
and
sperm
and
it's
going
to
produce
eggs
that
they
lay,
and
then
they
become
new
worms.
But
that
germline
is,
you,
know,
segregated
pretty
early
in
development
and
it
just
stays
there
and
it.
You
know,
there's
a
reason
for
that,
and
the
reason
is
is
because
you
want
to
protect
the
germline,
because
you
don't
want
to
have
too
many
mutations
in
it,
for
example,
random
mutations.
A
So
it's
you
know
there
are
the
reasons
for
that,
but
I'm
going
to
talk
about
like
how
you
might
we
might
be
able
to
take
the
germ
layer
model
and
improve
sort
of
our
way
of
thinking
about
computational
agents
so
and
then
I'll
probably
go
over
some
submissions.
We
don't
really
have
much
in
the
way
of
submissions
to
talk
about,
although
I
want
to
go
over
it
once
more.
A
A
We
have
the
diva
learn
paper,
that'll,
be
something
after
we
get
the
new
release
out.
The
boring
billion
you
know
is
that
going
to
be
a
book?
Is
that
going
to
be
something
else,
a
paper
and
in
in
the
boring
billion
category?
I
put
a
lot
of
the
stuff
that
we've
been
talking
about
about
very
early
embryos
from
like
maybe
a
billion
years
ago,
and
so
that's
something
that
is
sort
of
an
open
open-ended
project
it
could
be
a
paper
could
be
a
book.
A
You
know
we
haven't
really
touched
that
in
a
while,
the
mathematics
of
diva
worm,
which
is
this
poster
that
would
become
a
book
chapter
or
something
similar
where
we
have
all
the
major
equations
of
diva
worm
and
maybe
next
week
I'll
try
to
bring
that
back
together
and
maybe
we'll
try
to
see
if
we
can
add
to
it
or
if
there's
things
that
we
need
to
change
or
what.
But
I
think
I
have
like
we
have
like
four
or
five
categories,
and
you
know
they
may
be
equations
that
may
be
like
computational
models.
A
So
it's
just
a
way
for
people
to
see
like
you
know
what
kinds
of
things
we're
thinking
about
in
terms
of
mathematical
models,
and
so
you
know
you
have
deep
learning
models.
You
have
connectomes,
you
have
morphogenesis
models,
you
have
other
types
of
things
that
are,
you
know,
have
different,
fundamentally
different
ways
of
being
fundamentally
different
representations
of
development.
So
well
we
can
go
through
that
and
see
what
we
can
add
or
refine
on
that
and
then
eventually,
I
would
like
to
make
that
into
maybe
like
a
short
chapter.
A
A
We
have
this
williamson
symbiosis
test
and
this
is
using
molecular
data.
Last
week
I
talked
about
hybridizat
or
I
talked
about
metamorphosis
and
I
talked
about.
I
had
some
articles
that
I
showed
people
on
the
transition
from
a
caterpillar
to
a
butterfly,
and
so
that's
something
that
you
know
you
see
in
the
phenotype.
A
A
There
are
molecular
level
simulations
of
diatoms
they're,
these
morphological
simulations
of
diatoms
they're,
these
archaea
models
of
archaea,
where
we
have
these
archaea
bacteria,
which
form
these
different
shaped
phenotypes
or
single
cell
organisms,
and
then
the
idea
would
be
to
analyze
them
with
some
sort
of
approach.
We
talked
about
networks
where
these,
these
networks,
that
I
proposed,
which
are
these
euler
path,
networks,
different
ways
of
doing
that.
A
There's
this
eye
of
nature
book,
steve,
mcgrew's,
magnum
opus.
I
guess
he
could
call
it
and-
and
it's
really
an
interesting
book,
he
brings
up
a
lot
of
topics
and,
and
he
actually
in
reading
through
it,
he
does
hit
upon
a
lot
of
these
themes
about
like
non-neuronal,
cognition
and
other
things
and
tying
it
to
evolution.
So
it's
a
really
interesting
book
and
I
don't
know
you
know,
there's
a
I
guess,
there's
a
desire
to
like
make
it
sort
of
publish
it
post
posthumously,
but
that's
something
that
I
don't
know
like.
A
Who
would
be
interested
in
that
we
might
publish
it
as
a
series
of
blog
posts
or
novelizations,
and
you
know
there
are
different
ways.
We
can
do
that.
We're
not
a
publishing
organization,
so
I
don't
know,
but
definitely
it's
there
and
you
know
and
then
there's
this,
and
this
is
something
I'm
going
to
talk
about
today,
follow
up
on
today
with
this
talk
about
view
or
well,
the
agents
as
germ
layers-
and
this
is
differentiation
trees
of
the
brain.
So
this
is
something
that
you
know.
A
A
B
A
And
kind
of
just
give
people
a
reminder
what
we're
doing
no
deadlines
on
the
horizon.
I
think
we've
we
don't
really
have
any
deadlines.
There
is
this
dynamics
as
europe.
This
is
something
that
I'm
doing
called
game
theory
of
develop
or
I
guess,
of
developmental
processes.
Here
I
changed
the
name
of
it,
but
it's
it's
a
work
in
progress,
so
this
is
something
I'll
probably
present
to
the
group.
This
is
due,
I
think,
at
the
end
of
this
coming
month,
so
this
is
going
to
be
a
talk
and
this
is
dynamics
days
europe.
A
So
this
is
dynamics
days.
It's
it's
sort
of
a
physics
and
complexity
conference.
They
do
a
lot
of
people
like
from
the
dynamical
systems.
Community
are
there
and
there
are
a
lot
of
mathematicians
and
physicists
and
some
by
you
know,
computational
biologists
and
this
group.
They
they
vo
they've,
hosted
it
every
year
in,
like
europe,
in
the
united
states
and
and
with
kova
they've
kind
of
had
to
move
it
online,
and
so
the
european
group
has
been
active
in
doing
that.
A
It's
a
like
a
games
and
development
group.
So
they're,
like
you,
know,
six
lectures
in
this
series
and
they're
all
different
topics
around
it.
They
kind
of
synthesize
the
group
based
on
like
some
of
the
the
submissions
that
they
had.
So
it's
actually
I'll
bring
it
up.
When
I
give
the
presentation
it's
a
pretty
interesting
and
eclectic
mix
of
talks,
and
hopefully
it'll
be
recorded
online,
so
we
can
see
it
later.
A
So
if
we
have
deadlines,
please
let
me
know
I
can
add
them
on
here.
People
want
to
submit
things.
I
know
there
are
things
coming
up:
the
the
neurops
workshops,
for
example,
which
are
you
know,
for
machine
learning,
for
deep
learning
those
are
coming
up,
so
the
main
conference
is
actually
quite
competitive,
but
the
workshop
sometimes
can
offer
like
you
know,
sub
topics
and
those
are,
you
know,
maybe
less
competitive,
and
they
you
know
you
just
write
up
a
paper.
A
So
you
know
that's
a
good
opportunity
to
sort
of
get
involved
with
nurips,
and
then
you
know
I
I
they'll
be
advertising
these,
so
I
think,
maybe
more
towards
fall,
we'll
find
out
what
they
are,
but
maybe
there'll
be
some
that
are
relevant
to
our
group,
all
right.
So
this
is
our
task
board.
I
didn't
really
want
to
go
over.
I
went
over
this
last
week.
I
think
pretty
deeply.
A
We
have
again
our
finished
category.
We
have
some
things
that
have
fallen
off
the
radar.
I
updated
this
after
last
meeting,
so
there
are
a
bunch
of
things
that
are
kind
of
off
the
radar,
but
you
know
those
are
things
that
so,
if
you're
interested
in
the
group
meetings
task
board,
let
me
go
in
there
put
that
in
there,
and
so
this
is
for
major
tasks
for
this
season.
A
A
A
A
So
let
me
put
this
up
here
so
these
are
developed.
These
are
computational
agents
that
they're
not
like
c
elegans,
necessarily
but
they're,
very
similar
in
that
they
have
these
very
minimal
nervous
systems
and
so
in
a
typical
breitenberg
vehicle.
You
have
this
vehicle,
which
is
this
embodied
agent
and
it
has
a
body
and
it
has
sensors
at
the
front
effectors
at
the
back,
and
you
can
configure
the
sensors
and
effectors
any
way
you
want.
A
But
the
critical
point
is
that
the
sensors
and
effectors
are
linked
somehow
so
in
the
classical
greenberg
vehicles,
they're
linked
to
the
straight.
You
know
a
straight
connection
sensor
to
effector,
and
then
you
can
have
combinations
from
like
the
left
sensor
to
the
right
wheel
or
the
right
sensor
of
left
wheel,
or
you
can
have
these,
what
they
call
contralateral
connections
or
you
can
have
like
a
ipsilateral
connection,
which
is
the
left
sensor
to
the
left
wheel.
That's
just
borrowing
terminology
from
the
brain.
A
A
A
A
And
then
we
have
this
more
complex
nervous
system,
where
you
have
not
just
these
singular
connections
between
sensor
and
effector,
but
these
intermediate
or
what
we
might
call
interneurons,
which
actually
you
know,
allow
you
to
mediate
some
of
the
signals
or
attenuate
them.
So
you
have
some
sensors
or
you
have
three
sensors
at
the
front
of
this
vehicle
they're
coming
into
this
single
interneuron
and
then
but
they're.
A
Also,
two
of
them
are
also
mapping
to
this
interneuron
and
then
these
two
interneurons
are
mapping
to
this
middle
internal
on
here
and
then
back
down
to
this
one
out
here,
which
then
goes
to
the
wheels.
So
you
have
a
multiple
and
then
this
one
kind
of
goes
from
the
first
inner
neuron
out
to
the
wheel
directly.
So
you
have
all
these
layers
of
control
that
you
can
use
and
they
develop
in
development.
A
So
they
give
you
these
developmental
stages
that
you
can
work
on
and
you
can
develop
behaviors
and
then
you
know
hard
code
them
into
the
agent.
Now,
where
does
development
come
in?
So
development
comes
in
at
the
morphogenesis
level,
but
also
we
want
to
get
a
little
bit
more
detailed.
We
want
to
say:
okay,
we
can
exploit
this
idea
of
triple
blastic
development,
so
we
can
use,
we
can
take
the
body
and
the
brain
and
think
of
it
as
a
single
sort
of
developmental
system.
A
So
I
mentioned
before
that
you
have
these
dipolastic
systems
and
triboblastic
systems.
So
in
the
you
know,
in
different
in
the
tree
of
life,
you
have
these
tenophora,
which
show
like
these
dipoplastic
phenotypes,
and
then
you
have
nidaria.
You
also
have
a
dipoblastic
phenotype,
and
then
you
have
this
evolution
of
something
called
the
mesoderm,
which
you
only
see
in
tripoblasts
or
bilateria
and
bilateria
are,
of
course,
us
c
elegans
and
a
number
of
other
organic.
You
know
a
lot
of
different
organisms
and
they
have
biola
terrier
defined
by
their
bilateral
symmetry.
A
A
So
this
is
these
layers
emerge
in
the
embryo,
not
the
early
embryos
so
much
but
like
later
on-
and
you
know,
they'll
become
different
types
of
tissue,
and
this
sometimes
this
phenotype
will
be
conserved
so
that
you'll
have
like
the
gut
will
remain
in
the
middle
or
at
the
midline
of
the
organism.
And
then
you
have
these.
These
tissues
that
sort
of
emerge
around
it
and
sometimes
they
move
around
and
become
different
tissues,
but
they
they
maintain
the
same
layering.
So
we
use
this
as
an
inspiration
for
these
agents.
A
We
start
with
this
sort
of
tripoblastic
origin
point:
it's
just
some
concentric
circles
and
then
from
there
we
can
take
like
this
black
inner
part.
You
know
we
kind
of
play
with
the
metaphor
a
little
bit
that
black
inner
part
will
then
become
like
the
sensors
and
effectors,
so
they
kind
of
can
move
around
to
the
edges
of
the
of
the
agent
you
have
this
outer
layer
or
this
outer
shell,
which
is
remains
throughout
development,
and
then
this
middle
layer,
which
in
this
case
is
going
to
become
our
neural
network.
So
we
have
this.
A
These.
These
cells
are
the
space
that
can
become
the
neural
network,
and
then
this,
wherever
this
moves
will
be,
will
define
where
the
network
evolves.
These
sensors
and
effectors
will
position,
so
the
idea
would
be
you
could
do.
You
could
develop
an
agent
in
the
following
way.
You
could
define
where
these
sensors
and
effectors,
which
are
the
inner
layer,
are
going
to
be
along
the
edge
or
maybe
even
inside
the
agent.
A
A
We've
talked
about
hydro
before
in
this
group
and
they're,
a
marine
organism
that
has
this
thing
called
a
nervous
net
and
the
nervous
net
is
something
that
where
they
experience
their
the
sensory
world,
not
through,
like
this
formal
set
of
sensors
and
effectors.
But
there
are
things
that
propagate
through
the
organism,
so
they
get,
they
sense
things
and
then,
instead
of
having
a
central
nervous
system,
everything
just
propagates
through
to
different
parts
of
the
body,
and
so
their
neural
networks
are
more
integrated
with
most
of
their
body
than
other
organisms.
A
I
didn't
really
map
this
to
the
tree,
but
I
just
wanted
to
show
how
this
works.
So
you
have
these
different
germ
layers
and
we'll
call
them
a
b
and
c,
and
so
the
the
branch
length
is
important
because
the
longer
the
branches
are
the
more
chance
you
have
for
cell
proliferation,
so
you
can
have
like
a
bigger
layer.
A
So
this
layer
here
is
a
certain
size.
It
can
grow
in
size
by
having
a
longer
period
of
cell
division
and
proliferation,
and
so
that
can
be
mapped
to
this
tree.
And
you
know
this
is
something
you
can
use
like.
You
know
think
about
this
as
a
computational
mechanism,
where
you're
looking
at
like
you
know,
I
can
design
a
tree
of
this
topology
with
branch
lengths
of
a
certain
length
and
then
that
would
translate
into
a
certain
developmental
vehicle.
So
you
would
go
from
this
basic
tripoblastic
arrangement
implement
this
tree.
A
A
Is
basically
the
way
that
the
different
layers
are
are
differentiating
and
evolving?
So
there's
this
a
this
b
and
so
in
in
layer
b,
we
actually
see
that
there's
this
initial
layer
that
that
happens
and
then
there's
a
split.
So
this
b
would
be
roughly
analogous
to
these
sensors
and
effectors,
where
you
go
from
having
one
blob,
one
black
blob
to
two
black
blobs
that
divide
and
then
they
migrate
and
that's,
of
course
not
as
in
within
the
scope
of
this
tree.
A
But
you
have
now
you
have
these
two
b
layers
b,
prime
and
a
b
prime
prime.
That
then
you
can.
You
know
those
have
independent
instructions
on
how
those
will
proliferate
and
when
those
when
they'll
terminate
and
then
c
is
this
body,
which
then
is
roughly
the
same
size.
But
again
it
can
proliferate
for
a
longer
shorter
time,
depending
on
the
length
of
the
branch.
A
A
You
know
in
this
case
we
actually
are
doing
a
little
bit
different
than
this
model,
where
we
have
a
and
b
it's
like
a
and
then
this
bc
precursor.
That
starts
out.
So
you
have
this,
you
would
have
a
single
and
I
didn't
show
the
single
sort
of
germ.
You
know
the
the
sort
of
the
single
cell-
or
you
know
four
cell
case,
where
there's
no
real
differentiation
of
these
three
germ
layers.
But
once
you
start
to
you,
can
you
can
start
to
differentiate
these
three
germ
layers
using
this
sort
of
tree?
A
A
A
That
would
determine
what
this
looks
like
from,
say,
a
undifferentiated
sphere
to
this,
and
then
you'd
have
the
or
agent
would
interact
with
its
environment,
and
it
would
gain
information
that
way,
and
this
this
allows
it
to
become
a
mature
agent
with
a
nervous
system
and
then
training
the
nervous
system,
and
then
this
is
sort
of
the
progression
you
go
from
this
three-layered
three-germ
layered
precursor
to
something
that
has
different
layers.
The
layers
emerge
over
time
and
then
you
end
up
with
this
vehicle.
A
This
is
norbert
weiner
and
he
built
one
of
these
animats,
and
so
this
is
sort
of
the
inspiration
for
this,
among
other
things,
so
further
developments
on
this,
we
still
have
to
determine
sort
of
what
is
the
innate
component
that
enables
all
this.
So
this
requires
you
to
have
this
innate
component.
That
tells
you
tells
the
agent
kind
of
what
to
look
like,
and
so
there
are
a
number
of
options.
A
A
There's
less
control
on
something
like
that,
but
you
get
more
interesting
options
so
that
connected
to
that
is
sort
of
using
genetic
algorithms
and
we've
designed
some
genetic
algorithms
for
this.
But
we
really
haven't
been
able
to
specialize
any,
and
this
would
search
for
alternative
configurations
so
for
the
nervous
system.
Here,
we've
worked
with
some
genetic
algorithms
to
try
to
shape
what
this
looks
like
and
select
the
best
topologies
with
for
this
internal
nervous
system.
But
you
can
go
further
and
say
what
what
should
this
vehicle
look
like
in
development?
A
You
know
what
kinds
of
things
are
developmentally
advantageous.
What
kinds
of
vehicles
are
sort
of
the
optimal
phenotype
for
different
tasks?
So
you
can,
you
know,
have
a
fitness
function.
That
says
I
want
to
do
this
type
of
task
with
the
agent.
What's
the
best
phenotype
we
can
evolve
and
or
develop,
and
so
you,
you
know
you
might
get
something
that
looks
like
this
or
something
looks
very
different.
A
We
can
also
then
use
genetic
regulatory
networks,
and
I
know
we've
talked
a
little
bit
about
these,
but
this
is
something
that's
separate
from
a
genetic
algorithm
and
it's
a
way
to
sort
of
model
how
genes
are
regulated
in
their
expression,
and
so
these
genetic
regulatory
networks
usually
use
like
a
some
sort
of
input.
They
use
a
signal
and
they
like,
if
you
have
a
genetic
algorithm
and
you
have
things
organized
as
genes
and
they
get
you
know
they
get
selected
for
they
get
expressed.
A
You
know
genetic
regulatory
networks
can
shut
them
down
or
turn
them
on
at
different
times,
and
so
this
is
kind
of
useful
for
tuning
outputs
of
a
deterministic
program
or
even
something
more
like
a
genetic
algorithm
and
then,
finally,
you
can
have
hybrid
programs,
which
you
know
combine
like
a
deterministic
program
with
a
genetic
algorithm,
and
so
you
can
see
that
that's
very
useful
in
a
case
like
this
or
you're
trying
to
map
the
differentiation
of
these
different
parts
of
the
agent
body.
And
then
you
want
to
figure
out.
A
E
E
A
Slides,
I
I
just
kind
of
gave
a
very
quick
overview
of
it,
but
we're
preparing
a
paper
for
like
an
artificial
life
special
issue.
This
is
something
we've
been
doing
on
embodied
intelligence,
and
so
this
is
something
we're
going
to
be
working
out
for
that
paper
too.
So
it's
a
sort
of
a
time
when,
like
we
need
to
have
like
in
our
paper,
we
already
have
the
paper
written,
but
we
need
to
revise
it
and
we
need
like
more
specific
things
to
put
in
it.
A
So
it's
like
a
matter
of
kind
of
figuring
out
what
works
and
maybe
what
doesn't
or
what's
plausible,
but
definitely
yeah.
It's
not.
I
mean
people
have
done
this
sort
of
thing
with
developmental
agents
before
with
embryos
and
they
you
know
they've
used
things
like
cellular
automata
to
to
define
the
shape,
but,
like
you
know,
saying
that
you
know
that
cellular
automata,
usually
just
kind
of
give
you
interesting
shapes,
and
so
there's
what
it.
A
You
know
it's
good,
because
you
can
see
what's
plausible
or
you
know
you
know,
you
have
a
common
set
of
rules
and
you
say
what
emerges
if
you
just
have
a
common
set
of
rules.
In
this
case
you
have
like.
Maybe
you
know,
your
different
germ
layers
have
maybe
different
sets
of
rules
or
you're
thinking
about
it
in
terms
of
like
you
know,
you're
going
from
this
blob
the
things
of
different
functions
and
then
you're
getting
this
agent.
A
That
does
you
know
maybe
has
a
behavioral
aspect,
but
also
has
like
this
body,
and
you
know
other
things
that
go
along
with
it,
so
different
different
objectives.
I
guess
so
they
do
a
multi-objective
optimization,
but
that's
very
hard
to
implement,
and
so
this
might
be
a
back
door
in
some
ways
to
that.
For
a
very
specific
thing,.
A
A
Well,
okay,
so
I'm
gonna
go
through
our
papers
now
and
I'm
gonna
as
if
you've
been
following
along
in
the
last
couple
weeks.
We've
been
talking
about
environment,
and
this
is
what
is
environment
three,
so
this
is
last
couple
times.
We've
talked
about
light
and
we've
talked
about,
I
think
thermal,
optimization
and
again
to
refresh
our
memory
about
this.
This
open
question
about
environment
and
the
questions
are:
what
is
it?
What
is
its
relevance
and
how
do
you
define
it?
So
this
came
out
of
a
discussion.
A
We
had
about
some
papers
where
there's
this
term
environment
it's
very
loose
and
we
talked
a
little
bit
about
environment
in
the
last
set
of
slides
about,
like
you
know,
there's
this
environment,
that's
giving
you
know
influencing
development,
but
what
does
that
mean?
And
so
we've
come
down
with
two
sort
of
things
about
environment.
One!
A
A
It's
transduced
in
the
biological
system
or
specific
information
ratios
patterns
and
codes
are
transduced
in
the
biological
system,
and
so
those
you've
been
at
neuro
match
know
something
about
like
how
maybe
ratios
patterns
and
codes
are
in
the
nervous
system
and
how
you
know,
maybe
you
have
things
in
the
environment
like
you,
know
the
position
or
the
the
length
of
days
versus
night,
so
you
have
things
that
trigger
circadian
rhythms.
You
have
you
know.
A
Rates
of
like
sounds
in
the
environment
that
are
used
by
birds,
their
their
song
has
like
a
sort
of
this
coding
or
this
ratio
organization,
so
that
those
are
things
that
you
know
and
then,
but
then
this
general
information,
these
stresses
or
intensities.
These
are
again
things
like
temperature
or
other.
You
know
other
things
that
we
can
quantify
or
at
least
get
our
hands
around,
so
we
can
quantify
it
where
things
like
colors.
A
So
if
we
go
back
to
this,
what
is
environment
today?
I
have
this
paper
and
this
is
actually
in
c
elegans.
This
is
eduardo,
escherto
is
from
the
open
worm
group
and
a
couple
other
people
in
this
paper
and
the
title
is
persistent
thermal
input,
controls,
steering,
behavior
and
c
elegans.
So
c
elegans
moves
around
these
very
stereo
typical
movements
and
they
have
these
steering
behaviors.
A
So
they
do
these
loops
or
they
do.
You
know
they
loop
back
around
if
they're
moving
in
a
certain
direction,
they
have.
They
call
these
omega
terms,
which
kind
of
move
in
an
omega
pattern
and
they're
different
ways
that
they
can
move
around
and
they
steer
their
behavior
towards
resources
or
away
from
predators
or
things
like
that,
and
so
this
is
an
important
type
of
movement
for
them.
A
So
the
abstract
says:
motile
organisms
actively
detect
environmental
signals
and
migrate
to
preferable
environments,
so
they're
looking
for
things
that
they
want
to
exploit,
or
maybe
they
want
to
avoid-
and
this
is
not
just
c
elegans
but
even
like
in
the
braidenburg
vehicles.
They
do
the
same
sort
of
thing,
especially
small
sized
animals,
convert
subtle
differences
in
sensory
input
into
orientation,
behavioral
output
for
directly
steering
towards
the
destination
so
they're.
A
So
this
is
something
that
you'll
see,
but
the
neural
mechanisms
underlying
steering
behavior
remain
elusive.
Here
we
analyze
the
c
elegans
thermotactic
behavior.
So
thermotaxis
is
where
the
organism
moves
towards
a
thermal
source.
So,
like
you
know,
if
you
think
about
like
a
heat
source,
it
might
move
towards
the
heat
source
or
if
it's
too
hot,
it
might
move
away
towards
the
heat
source.
It
models
the
environment
as
sort
of
this
gradient
and
it
moves
towards
a
certain
thermal
regime.
A
Okay.
So
these
are
just
very
simple
movements
and
they're
moving
towards
or
away
from
things.
So
if
you
were
to
move
the
thermal
source
around
the
movement
would
just
move
around
they'd
start
moving
around
towards
wherever
the
thermal
source
was
moving,
and
so
and
again
when
I
showed
the
slides
on
braidenburg
vehicles,
I
showed
the
sun,
so
bradenburg
vehicles
will
move
towards
the
light.
A
They
exhibit
these
taxes
these,
what
they
call
photo
taxes
in
this
case
we're
going
to
have
thermo
taxes,
so
it's
very
similar
which
a
small
number
of
neurons
are
shown
to
mediate,
steering
towards
a
destination
temperature.
So
this
is
a
very
small
number
of
neurons
in
the
c
elegans
nervous
system,
so
we
constructed
a
neuroanatomical
model
and
used
an
evolutionary
algorithm
to
find
configurations
of
them
all
that
reproduce.
Empirical
thermotactic
behavior.
A
I
don't
know
if
they're,
anticipating
it,
but
their
overall
movement
in
this,
you
know
as
they're
moving
it
around
or
putting
in
the
same
place
it's
kind
of
affecting
how
they're
doing
this,
maybe
they're
just
kind
of
moving
and
they
kind
of
learn.
Well,
you
have
to
move
a
lot
you
have
to
be
like
you
know,
you
have
to
maintain
this
sort
of
place
where
you
can
move
around.
You
know
you
can
prepare
to
move
so
persistent
temperature
increment
lessen
steering
rates
resulting
in
street
movement.
A
Further
spectral,
spectral,
decomposition
and
neural
activities
show
that
thermal
signals
are
transmitted
from
a
sensory
neuron
to
motor
neurons
on
the
longer
time
scale
than
sinusoidal
locomotion,
so
they
actually
are
doing
this
in
their
nervous
system,
where
they're
sort
of
integrating
the
information
over
multiple
moves
and
they're
sort
of
observing
it
more
more
long
term
than
just
kind
of
like
a
reaction,
but
the
behaviors
kind
of
show
themselves
as
a
reaction.
This
is
really
interesting.
A
So
this
is
again
like
they're,
using
this
temperature
gradient
and
they're,
observing
these
different
types
of
movement
and
then,
of
course,
they're
looking
at
the
at
the
wire
at
the
neural,
anatomical,
wiring
and
they're,
seeing
that
there's
this
longer
term
sort
of
activation
within
that
circuit,
that
shows
maybe
a
little
bit
of
anticipation,
so
they're
kind
of
going.
So
there
isn't
a
lot
to
this
paper.
A
A
How
this
works
and
then
they
show
the
the
neural
circuit,
so
we're
talking
about
six
neurons,
I
think,
are
seven
neurons.
So
we
have
these
networks,
where
you
have.
You
know
very
small
network
of
neurons
and
they're
controlling
the
steering
behavior,
and
it's
basically
like
you
know
in
sensory
input,
turn
right
turn
left.
You
know.
Maybe
reverse
course.
A
Maybe
you
know
it's
very
simple
and
we
can.
We
can
distill
this
in
c
elegans
down
to
about
six
neurons,
and
so
they
show
sort
of
the
time
scale
of
this.
So
they
look.
They
monitor
the
current
state
at
every
0.1.
Second,
there's
this
potential
for
a
move,
so
they
can
make
these
different
types
of
turns
or
curves,
and
then
they
exit
that
curve
and
they
make
another
curve
and
the
nature
of
the
body
is
c.
A
A
And
so
they
kind
of
go
through
this
neuroanatomical
models,
reproduce
thermotactic,
behavior,
so
they're
able
to
actually
model
this
with
an
evolutionary
algorithm
or
a
genetic
algorithm
and
they're
able
to
look
at
the
look
at
the
circuit
and
how
it's
activated
and
then
they're
able
to
to
they
have
the
experiment
in
the
simulation
and
they
match
up
pretty
well,
and
so
that's
so
that's
it
and
then
you
know
there
are
different
temperature
change
regimes.
So
you
just
look
at
how
that
affects
the
curving
rates.
So
this
is
a
good
example.
A
So
that's
what
is
environment?
Let's
see
this
is
a
paper
we're
talking
about
genetic
regulatory
networks
and
this
paper
is
it's
a
new
paper
in
cells
and
development
and
it's
about
cell
regulatory
networks,
so
the
cell
in
the
age
of
genomic
revolution,
cell
regulatory
networks,
and
so
this
this
paper
is.
A
The
abstract
says,
over
the
last
few
years,
an
intense
activity
in
the
areas
of
advanced
microscopy
and
quantitative
cell
biology
have
put
the
focus
on
the
morphogenetic
events
at
shape
embryos.
So
that's
what
we're
interested
in
with
the
diva
learn
platform
are
these
events
that
you
know
where
the
cells
are
moving
around
and
they're?
Dividing
the
interest
in
these
processes
is
taking
place
against
the
backdrop
of
genomic
studies,
particularly
of
global
patterns
of
gene
expression
at
the
level
of
single
cells.
So
people
are
interested
in
what
the
cells
are
expressing
in
terms
of
genes.
A
What
genes
are
expressed
when,
but
that
doesn't
really
account
for
the
way
cells
build
tissues
and
organs.
So
people
have
these
very
elaborate
models
of
like
they
know
the
genome
of
c
elegans.
They
know
some
of
the
regulatory
aspects
of
the
genes,
but
then
that
doesn't
tell
you
a
lot
about
what's
going
on,
then
at
the
cellular
level,
where
you
have
migration
and
division.
A
We
know
things
about
cell
division
in
terms
of
like
what
genes
are
involved
and
how
they're
regulated,
but
we
don't
know
how
that
actually
maps
on
to
like
the
type
of
real
time
model
that
we're
trying
to
build
with
divalern.
So
there's
a
gap
there
here
we
discuss
the
need
to
integrate
the
activity
of
genes
with
that
of
cells
and
propose
a
need
to
develop
a
framework
based
on
cellular
processes
and
cell
interactions
that
parallels
that
which
has
been
created
for
gene
activity
in
the
form
of
gene
regulatory
networks.
So
this
is
the
grns.
A
We
begin
to
do
this
by
suggesting
elements
for
building
cell
regulatory
networks
or
crns
in
the
same
manner
that
grns
creates
schedules
of
gene
expression,
so
they
basically
determine
when
the
genes
are
turned
on
or
turned
off
or
turned
on
halfway
or
whatever
that
result
in
the
emergence
of
sulfates
over
time,
crns
create
tissues
and
organs
space
in
in
space,
so
they
have
this
regulation
of
space
and
they
regulate.
You
know
what
tissues
and
organs
are
built
so.
A
Almost
like
the
differentiation
trees,
this
is
very
similar
in
in
terms
of
the
way
it's
conceived.
We
also
suggest
that
grns
and
crns
might
interact
in
the
building
of
embryos
through
feedback
loops
involving
mechanics
and
tissue
tectonics.
So
this
is
the
sort
of
thing
if
you've
seen
susan's
presentations,
where
she
talks
about
cell
mechanics
and
development.
A
Some
of
the
things
that
she's
done
she's
talked
about
like
this
biological
fracking
that
goes
on
in
different
in
the
embryo
or
some
of
the
you
know.
You
know
pressure
the
role
of
pressure
and
the
role
of
like
other
types
of
physical
stimuli
on
cells.
This
is
sort
of
the
thing
this
is
the
sierras
are
getting
at.
The
grns,
of
course,
are
affected
by
these
things,
so
grn's
are
triggered.
Example,
for
example,
by
stresses
or
by
you
know,
other
types
of
signals
that
are
sometimes
you
know.
A
And
they
kind
of
give,
you
know,
motivate
the
problem,
so
they
talk
a
lot
about
like
what's
going
on
in
the
in
the
embryo
in
terms
of
different
movements.
So
you
have
you
know.
We
know
that,
like
a
lot
of
developmental
biologists
are
obsessed
with
like
gene
expression,
assays
and
looking
at
how
different
genes
are
expressed
in
the
embryo,
but
then
there
are
these
other
types
of
collective
behaviors
like
event,
invagination
convergent
and
extension
and
cavitation.
A
A
There
are
these
large
scale
movements
that
happen
like
folding
and
bending
and
other
types
of
displacements
amongst
the
cells
that
you
know
you
can't
really
characterize
with
a
grn,
and
so
this
is
what
they're
kind
of
trying
to
model
but
they're
doing
in
a
way.
That's
you
know
regulatory
network
approach,
so
they're.
Looking
at
these
they're,
no
grns
are
a
representation
of
gene
expression,
their
nodes
are
transcription
factors
and
sometimes
signals
and
signaling
pathways.
Their
assembly
can
result
in
what
they
call
circuits.
A
So
these
are
like
brain
circuits
except
that
they're
on
the
molecular
level,
and
these
are
small
networks
that
perform
specific
operations,
usually
associated
motifs,
and
so
these
motifs
are
like
sets
of
genes
that
are
linked
by
their
expression.
So
you
can
have
things
like
coherent
feed
forward
loops
which
provide
temporal
control
in
gene,
switching
and
incoherent
feed
forward
loops
that
lead
to
oscillations
and
perform
adaptation.
A
But
then
grns
do
not
contain
and
cannot
provide
spatial
information,
such
as
volumes,
geometry
densities
and
the
associated
mechanochemical
signaling.
So
this
is
where
your
crns
come
into
play
and
so
yeah.
There
are
a
lot
of
intrinsic
cellular
activities
that
you
also
need
to
model,
and
so
these
are
things
like
integrating
the
activity
of
the
cytoskeleton,
the
cell
cell
and
cell
matrix
adhesion
and
intrasolar
trafficking
systems,
which
are
actually
the
nodes
of
the
crn.
A
A
A
A
You
can
see
what
kind
of
a
network
they
have.
They
have
these
things
that
you
know
you
have
regulation
and
forms
of
these
red
arrows
and
the
blue
arrows
are
the
sort
of
connections
between
them,
and
you
can
see
this
thing
where
you
have
this
crossover
between
the
causal
factors
for
crns
and
the
causal
factors
for
grns,
and
they
they
affect
one
another.
So
that's
what
they're
trying
to
point
out?
They
don't
really
have
anything
implemented
here.
A
A
A
So
this
is
a
pattern
in
a
this
is
a
shark
embryo
or
a
very
early
shark,
and
you
can
see
that
this
whole
cat
shark
hatchling
stained
to
reveal
its
skin
teeth.
So
it
has
these
skin
teeth
along
the
along
the
back
here,
and
you
can
see
this
pattern.
It's
identical
patterning
which
looks
like
teeth,
and
you
have
two
parallel
rows
near
the
dorsal
fin.
A
So
this
shows
where
this
an
issue,
this
development
of
these
skin
teeth,
initiates
along
this
midline
section
here
and
then
the
tentacles
then
spread
out
to
dot
the
rest
of
the
body
over.
You
know
as
it
matures
in
keeping
with
the
action
of
a
turing-like
mechanism.
So
we've
talked
about
turing
mechanisms
before
and
they're.
These
mechanisms,
where
you
have
a
bunch
of
particles
and
they're
controlled
by
chemical
gradients,
that
control
the
expansion
of
different.
A
So,
for
example,
these
these
denticles
are
restricted
at
this
point
development
and
then
later
they're,
the
that
constraint
is
relaxed
and
they
spread
out
across
the
body
and
there's
a
reason
for
this,
and
the
reason
is,
is,
if
you
don't
have
this
precise
control,
you
can
end
up
with
all
sorts
of
developmental
mutations
in
vertebrates.
You
have
this
mutation
where
pec
it's
involves,
pac-6,
which
is
a
developmental
gene
and
if
pac-6
isn't
suppressed
at
the
right
time.
A
You
can
end
up
with
one
eye
instead
of
two,
and
they
call
that
like
a
what
is
it
the
it's,
the
one
you
know
the.
I
can't
remember
what
the
oh
the
cyclops,
so
they
call
it
a
cyclops
mutation,
and
so
you
know
this
is
just
a
matter
of
developmental
gene
expression
going
awry,
and
so
this
is
why
all
this
is
importantly
coordinated
in
time.
A
A
It
basically
outlines
of
endless
varieties
of
stripes
spots
and
scales
can
emerge
from
the
interaction
of
two
simple
chemical
agents
or
morphogens.
So
morphogens
is
just
a
theoretical
term
and
then
you
know
people
have
developed
other
types
of
theories,
but
this
is
one
theory.
I
think
that
is
very
interesting
in
terms
of
pattern.
Formation
really
in
terms
of
pattern
formation
hasn't
been
surpassed.
A
The
beauty
of
this
work
is
that
shows
that
there
might
be
a
very
strong
conservation
of
this
mechanism
performing
anything
from
shark
tentacles
to
bird
feathers.
This
study
bolsters
a
growing
theme
in
developmental
biology.
That
nature
tends
to
invent
something
once
and
then
plays
variations
on
that
theme,
so
we
can
use
turing
sort
of
tweaking
the
turing
model.
This
reaction
diffusion
mechanism
to
come
up
with
different.
You
know
different
types
of
phenotypes.
A
A
So
this
you
know
turing
models,
excite
developmental
biologists,
because,
despite
its
simplicity,
it
can
explain
a
lot
of
diverse
patterns.
Yet
in
practice
very
few
instances
of
patterning
in
nature
have
been
definitively
pinned
to
a
turing-like
mechanism.
So
it's
hard
to
prove
it's
just
easy
to
see
that
it's
consistent
with
it,
but
it's
hard
to
actually
prove
it
and
show,
like
you
know,
in
a
very
rigorous
way
that
this
is
actually
what's
going
on.
A
So
you
have
all
these
different
hair,
follicles
and
mice,
and
feathers
and
chips,
and
these
are
two
different
model
systems
that
sort
of
are
consistent
with
this.
But
again
we
don't.
We
haven't
proven
that
this
is
exactly
what's
going
on
so
yeah.
This
is
really
interesting
goes
on
about
sharks.
It
goes
on
about
turing
mechanisms
and
development.
So
that's
a
really
interesting
paper
as
well.
So
I'm
going
to
put
these
links
in
the
chat
and
I'm
going
to
kind
of
wrap
it
up
here.
If
people
do
people
have
any
questions
or.
E
C
E
A
Yeah,
well
I
mean
it's
like
you
have
this
common
mechanism,
so
you
could
have
feathers
or
you
could
have
scales
and
they
could
be.
You
know
they
could
be
shaped
in
the
same
way.
So
you
just
use
the
same
machinery
and
development
say
you
know
whether
this
organism
has
scales
or
there's
feathers
and
you
have
or
hair
and
it
just
you
know
it
allows
for
the
patterning
to
occur.