►
From YouTube: DevoWorm (2020, Meeting 20): GSoC updates, General Biological Models, Euler Paths for Life.
Description
Attendees: Bradly Alicea, Susan Crawford-Young, Mayukh Deb, Steve McGrew, Ujjwal Singh, Richard Gordon, and Vinay Varma.
A
A
B
A
B
A
A
A
Right
so
I
guess
I
don't
know,
Klaus
is
gonna
show
up
for
the
meeting.
You
know
dick
is
taking
alternate
weeks
off
the
summer,
he's
going
out
into
the
woods
and
fishing
opportunities
where
he
lives,
and
so
you
don't
have
very
long
of
a
summer,
so
they
have
to
take
advantage
of
it.
So
he's
gonna
take
like
you
know,
rather
weak
off.
A
So
I
guess
we'll
hear
from
my
o'connor's
law
on
there.
Ji-Suk
updates,
just
weekly
updates,
give
a
quick
update.
If
you
have
demos
show
that
if
you
have
questions
raise
them
and
then
I
think
Krishna
was
supposed
to
show
up
and
give
a
talk
on
his
idea.
I
don't
know
if
he's
gonna
do
that
yet
he's
still
gonna
do
that
and
then
I
was
gonna,
give
a
talk
on
this
Euler
cycles,
multi
cell
system
stuff,
that
I've
been
working
little
areas
and
then
yeah.
That
should
be
the
meeting.
So.
D
A
C
A
F
A
F
B
A
F
F
F
All
of
them,
but
first
I,
want
to
understand
what's
going
on,
because
my
next
up
will
be
normalization
as
I'd
like
to
be
using
z-transforms
on
things
like
I'll,
see
to
that.
But
I
don't
know
about
this
block
and
cross
panel.
I'll
have
to
study
with
what
these
are.
I'll
have
to
do.
Probably
so
I
know
whether
or
not
do
not.
Let
me
know
because
I
bet
this
scale
gives
us
care
to
a
certain
leg.
One
this
one
I
don't
like
skating
them
would
make
sense.
Yeah
well.
A
A
A
A
Yeah
but
one
problem,
you're
gonna
run
into
or
there's
a
problem
with.
That
is
that
is
that,
there's,
like
you,
know,
you're
averaging
over
the
entire
lineage.
So
sometimes
like
you
know,
we
think
about
in
terms
of
lineages
and
sub
lineages,
but
you're
combining
a
lot
of
data,
different
kinds,
contexts
and
you
know
eventually
if
they
become
functionally
differentiated
and
so
it's.
A
Yeah,
what
I'm
saying
is
that
you're,
you
know
when
you
get
from
like
say:
the
256
else
is
just
something
larger
you're
even
like
from
you
know,
64
so
stage
to
something
larger.
You
get
things
new.
You
know
you
get
things
that
are
starting
to
differentiate,
so
you
know
it
might
be
worth
well.
We
can
talk
about
that
later,
but
suffice
to
say
when
you
look
at
a
subway,
you
know
you're
getting
a
lot
of
averages
over
time,
but
you
know
this
is
good.
Yeah.
B
A
A
F
A
A
If
there's
some
problems
when
this
goes
first
of
all
as
well,
please
let
me
know
as
soon
as
possible.
If
you
have
problems
with
computing
resources
or
something
running,
you
know
we
can,
you
know
either
I
can
help
you
or
just
put
it
aside
for
a
while
and
then
we're
on
to
something
else.
So
we
keep
the
summer
moving,
and
you
know
we
don't
have
a
bottleneck.
Oh.
G
D
A
That's
it
yeah
drop
me
the
links
and
slack
when
you
can.
Thank
you.
Thank
you
for
my
open.
Those
wall
for
the
updates
looks
like
you're
doing
pretty
well
wall
for
you,
yeah
yeah,
Krishna,
yeah,
I'm,
gonna.
Now,
I'm
gonna
move
down
to
you
and
you're.
You
wanted
to
present
on
your
idea
that
you
have
for
a
project.
G
A
E
G
G
E
G
F
E
G
E
A
A
What
it
you
know,
you're
gonna,
find
a
lot
of
similarities
that
you're
also
gonna,
find
a
lot
of
differences,
so
I
mean
one
of
the
things
that
you
have
is
you
know
people
think
about
this
when
they
think
about
like,
like
the
Tree
of
Life
or
like
when
people
couldn't
develop.
Very
broad
theories
in
biology
is
that
they
have
a
lot
of
variation,
and
it's
not
all
the
same
in
the
sense
that
you
can
really.
A
It
using
a
very
simple
model,
so
I,
you
know,
I,
don't
I,
don't
necessary,
worry
about
that.
It's
something
we'd
have
to
account
for
in
terms
of
like
feature
selection,
so
you
need
you're.
Gonna
have
like
a
bunch
of
features,
know
that
are
similar
for
the
different
categories
and
then
some
that
are
gonna
be
like
different
for
different
categories.
So
you
know
you
might
have
a
collection
of
cluster
of
features
that
are
like.
G
A
Know
something
the
shared
between
radiology
and
data
and
slant
data,
and
they
must
be
like
you
know,
technical
heart
obsessed,
or
they
might
be
like
actual
things
that
are
just
kind
of
thing,
that
you're
able
to
see
in
common
I
mean
you
know
it's
talking
very
generally.
Now
they
think
you,
you
know
you
praise
you
from
the
data
it
would
emerge,
but
then
you
know
you'd
also
have
things
that
are
shared
or
them
that
are
not
shared,
there's
unique
to
say
like
animal
data
for
cells,
you
know
yeah.
B
A
Image,
you
know,
data
with
the
microscope,
you'll
find
stuff,
that's
very
unique
to
that,
and
so
I
mean
that's
something
you
want
to
take
into
account.
I
do
kind
of
I
do
kind
of
worried
too
about
what
Maya's
saying
about
like
the
size
of
it.
At
least
you
know
in
terms
of
like
what
we
talked
about
this
before
about
balancing
your
classes
about
getting
like
right,
you
know
getting
getting
things
that
are
just
kind
of
spurious
things.
A
G
A
Yeah
yeah
I
think
that's
good,
you
know,
and
then
you
might
also
do
this
like
initially
balance
them
by
their
type.
But
then,
once
you
know
like
some
of
the
features
you
might
try
and
like
you
know,
rebalancing
them
like
yeah.
You
know
like
some
types
of
radiology
data
or
maybe
clozaril
plant
data,
because
you
know
they
do
this
with
like
phylogenies
all
the
time
and
I
think
I've
talked
to
people
about
what
those
are.
Those
are
trees
where
they
they
try
to
figure
out
how
things
are
related
in
life.
So
in
terms.
A
Bacteria
and
the
a
lot
of
times
those
are
very
contentious
because
the
fight,
those
fine
features
and
they'll
use
them
to
like
build
a
tree
say
build
a
hypothesis.
This
thing
is
a
common
ancestor
of
two
other
things,
because
it
has
this
feature
that
you
know
it
looks
similar,
but
it's
it's
changed
over
time.
A
So,
but
the
thing
is
is
that
they
constantly
update
these
trees
based
on
different
criterion
that
people
use,
so
people
have
incorporated
gene
sequences,
they've
had
different
tree
looking
trees,
they
incorporate
different
morphological
traits
that
the
trees
change
again
and
it's
you
know,
and
they
built
consensus,
trees
and
they're
different.
Even
still
so.
I
would
expect
a
similar
problem
here,
where
you'd
have
just,
depending
on
your
input.
C
C
A
E
G
G
A
E
G
A
G
A
A
G
A
Thank
you
very
much
Krishna
for
that.
Let's
see,
Jessie
had
a
question
wasn't
about
this,
but
it
was
about
some
questions
about
the
basic
non-neuronal
cognition
papers,
but
I'll
probably
ask
you
for
that
in
slack
yeah.
Please
ask
me
about
that.
In
slack,
that's
something
I
think
we
taught
we've
been
just
kind
of
talking
about
that.
They
exist
in
the
last
couple
weeks,
but
I
haven't.
We
haven't
made
great
strides
in
that
area,
so
you
know
in
the
recent
past.
So
we'll
probably
turn
to
that
in
a
while.
A
A
A
So
here
your
Euler
path,
so
Euler
pads
also
can
be
used
to
define
networks
but
also
geometric
networks
or
geometric.
Like
shapes,
or
you
know,
complex
geometric
shapes
so
here
on
the
Left,
we
have
an
octahedron
and
dodecahedron,
so
these
are
with
eight
and
I
think
twenty
sides
of
piece,
and
so
these,
where
you
have
this
these
cycles
inside
of
each
of
those
cycles
as
a
side
right,
and
so
you
count
up.
This
is
eight.
They
don't
know
how
many
are
in
here.
A
You
know
it's
like
12
or
yet
anyways.
The
point
is,
is
that,
as
you
know,
each
of
these
cycles
represents
a
component,
that's
great,
and
so
you
have
this
network,
and
you
see
this
kind
of
bin
even
involve
ox
colonies
where
you
have
a
bunch
of
cells
in
this
county,
and
you
have
certain
properties
in
shape
that
required.
Well,
they
don't
miss
or
require
optimization
what
you
end
up
with
this
optimization
of
shape
and
you
end
up
with.
A
A
That's
you
know
very
alien
to
the
other
part
of
the
organism
or
there's
going
to
be
some
sharing
of
resources
across
all
these
cells,
and
so
this
is
why
we
have
circulatory
systems
or
other
types
of
systems
where
nutrients
are
trained
like
in
plants.
There's
a
lot
of
nutrient
transfer
up
and
down
the
root
system,
and
so
this
is
you'll
see
this
in
a
lot
of
multicellular
organisms
where
you
have
to
share
resources
across
the
colony
and
so
Whaler
circuits
in
a
very
technical
sense
are
an
attempt
to
cross
every
independent
edge.
A
A
You
only
want
to
visit
all
of
them
once
and
that's
it
and
the
reason
I
introduced
this
constraint
is
because
it's
something
that
we
understand
mathematically,
but
also
because
you
can
imagine
in
terms
of
symmetry,
if
you
visit
a
bunch
of
edges
multiple
times
that
that
gives
those
edges
an
advantage
in
terms
of
flows
and
other
things,
and
so
we'll
just
use
these
constraints
and
you'll
see
how
this
works
in
a
bit.
So
for
a
network
of
triangles,
you
have
a
symmetrical
transformation.
You
can,
you
know,
divide
them,
we'll
take
a
network
of
triangles.
A
For
example,
you
can
have
a
symmetrical
transformation
where
you
go
from
one
triangle
to
two
triangles
and
you
you
know
you
end
up
with
either
an
odd
number
of
cells
or
edges
or
an
even
number,
and
then
that
has
an
effect
on
whether
you
can
complete
my
water
circuit,
so
I
think
I
should
just
get
on
with
the
graphical
part
of
this,
because
it's
more
informative.
So
you
have
this.
I
have
there's
a
programmable
graph.
Online
I
was
using
to
generate
these.
So
this
shows
you,
like
you,
know
in
a
geometric
Network.
A
What
kinds
of
you
know
what
kinds
of
shortest
paths
you
can
generate.
So
we
have
these
nodes
and
we
have
these
edges.
The
edges
are
in
yellow
the
nodes
are
in
red
and
the
nodes
are
sort
of
the
end
of
each
edge
and
where
they
intersect
and
the
edges
are
the
flows
across
this
network.
So
we
have
this.
This
graph
actually
exhibits
a
you
Larian
path.
A
This
program
will
find
a
Larry
because,
as
it
turns
out,
Larian
pads
just
like
the
TSP
is
np-hard
and
it's
hard
to
get
a
analytical
solution
just
by
hand
and
this
example
it's
a
lot
more
complex.
This
does
not
have
a
you
Larian
path,
so
this
is
a
combination
of
triangles
and
rectangles
and
this
configuration
does
not
have
a
Eulerian
path.
A
That's
the
important
thing
to
remember,
and
the
number
of
edges
I
think
is
odd,
but
there's
a
I
haven't
been
able
to
figure
out
any
interval
where
you
have
you
Aryan
edges
versus
not
or
a
Eulerian
path
versus,
not
even
it's
not
as
simple
as
the
even
odd
distinction
I
mentioned
earlier.
It's
something
else:
it's
where
you
take
like,
say
a
triangle
and
you
divide
it
into
two
triangles
and
that's
EU
Aryan
path.
A
But
then,
when
you
introduce
maybe
a
fourth
or
fifth
triangle,
you
don't
get
it
and
then,
if
you,
maybe,
if
you
introduce
a
rectangle
you'll
get
it
back,
and
so,
if
you
as
you,
grow
these
structures,
you
run
into
situations
where
you
can
no
longer
build
a
Eulerian
path
and
then,
if
you
add
different,
shapes
or
different,
maybe
another
cell
division.
You
can
get
it
back
and
so,
but
that's
not
the
only
thing
we
can
find
out
from
this
exercise.
We
can
do
things
like
rotation,
scaling,
shear
and
perspective
of
these
different
shapes.
A
A
We
can
conformally
map
like
these
nodes
so
that
we
can
stretch
them
out,
and
so
this
is
like
development,
where
you're,
taking
like
a
landmark
in
a
new
area
or
in
a
larval
organism
and
you're
stretching
it
in
certain
ways,
and
so
this
is
a
conformal
mapping
showing
these
dots
and
this
would
correspond
to
these
nodes
and
then
you're
stretching
this
node
I
was
up
here.
You're
stretching
it
out
like
this,
but
you're
able
to
retain
this
Euler
path.
A
A
But
but
that's
how
this
this
works,
and
so
you
can
do
a
lot
of
things
of
this
type
of
simple,
geometric
approaches
and
understanding
how
these
things
work
now
this
is
where
it
gets
interesting.
Is
that-
and
this
is
what
sort
of
this
raise
it
closer
to
life,
so
this
is
very
similar
to
like
it.
Lineage
dreams
develop.
So
in
this
case,
what
we're
doing
is
we're
adding
we're
just
simply
using
our
replication.
A
Alright,
in
this
case
and
we're
building
just
take
this
triangle
and
we've
replicated,
and
then
we
try
to
do
one
of
these
build
one
of
these
illyrian
circuits
on
this
object
that
results,
and
so
in
this
triangle
we
can
build
a
new
illyrians
circuit
pretty
easily,
and
so
it
has
a
penile.
Our
cost
function
of
zero
cost
function,
value
of
zero.
You
go
down
and
you
add
a
triangle
to
it
like
this,
where
you
make
this
shape
and
it
has
a
cost
function
of
negative
one.
So
now
it's
no
longer
completing
that
circuit.
A
A
And
so
so
that's
there,
and
this
is
what
I'm
calling
modularity
so
you're
actually
creating
a
new
module
within
the
organism.
So
we
talked
about
modularity
in
terms
of
maybe
like
a
functional
subdivision,
and
this
is,
of
course,
a
module.
That's
different
from
the
rest
of
the
organism.
So
you
have
this
again.
You
have
this.
You
can't
complete
an
Euler
path
this
way.
So
you
add
another
triangle
up
at
the
top
and
you
recover
your
oil
or
completeness.
A
If
you
add
another
triangle
in
another
direction,
you
actually
score
worse,
but
then,
if
you
add
components
inside
these
objects
as
they're
replicating,
you
actually
recover
your
Euler
Euler
completeness-
and
this
is
again
modularity
because
you're
creating
new
modules,
phenotypic
modules,
there's
probably
an
analysis
to
be
done.
The
phenotypic
modules
themselves,
but
haven't
done
that
on
this,
and
so
this
actually
resembles
some
of
this
is
what
dick
was
talking,
boni,
that
his
triangle
triangular
cells
talk.
So
this
is
a
triangular
cell,
so
you
know
cells
can
take
on
a
lot
of
shapes.
A
If
you
look
in
human
cells,
they
take
on
a
lot
of
shapes,
but
especially
like
an
LG
and
in
micro
organisms.
They
have
a
lot
of
different.
They
have
a
lot
of
shape
diversity,
and
this
is
one
way
we
can
like
simulate
this
I.
Can
we
can
do
this
with
hexagons
with
hexagons?
It's
interesting,
because
if
you
used
a
simple
replication
rule,
you
always
get
oil
or
completeness,
and
it's
interesting,
because
people
are
one
of
the
hypotheses
about
honey,
combs.
A
The
structures
of
honey's
build
these
build
to
store
honey,
and
things
like
that
is
that
they're
optimized
for
like
packing
for
packing
in
maximal
number
of
cells
in
a
small
space,
and
so
this
kind
of
suggests
that
that's
correct,
that
there's
some
optimality
to
hexagons
and
so
I
have
again
I
haven't
analyzed
these
any
further,
but
this
is
something
that
as
an
interesting
result,
so
there's
no
lineage
tree
here.
It's
just
kind
of
a
progression
of
replication
of
cells
and
then,
of
course,
you
can
use.
So
this
is
not
supposed
to
be
here
buddy.
A
This
has.
This
is
a
very
complicated
lineage
tree
here.
Where
you
have
this
square
and
then
you
divide
it
into
two
squares
and
then
you
can
continue
with
the
square
motif
where
you're,
adding
in
you
had
just
have
a
simple
square
replicator
that
continues
down
this
path
and
it
keeps
producing
squares
on
this
grid,
and
that
also
shows
this
serial
oiler
completeness.
So
it's
they
call
it
fractal
reproduction,
because
you
can
reproduce
as
many
squares
as
you
want
erecting
rectangles
as
you
want
and
always
get.
A
The
same
result
never
gets
broken,
and
so
there's
a
fractality
to
this.
There
are
no
modules
and
these
in
this
case
you
have
an
asymmetric
transformation.
So
say
you
had
a
rectangle
of
a
different
size.
In
this
case,
you
would
have
an
asymmetric
transformation
that
still
preserves
that
oil
or
completeness
you
can
and
divide
this
bottom
cell
up.
So
it
can
divide
and
take
like
a
like
something
like
a
grid
like
this
and
that
you
know
that
preserves
its
oil
or
completeness.
A
But
if
you,
if
it's
off-center,
if
it's
an
asymmetric
division,
then
it
breaks
that
oil
or
completeness-
and
you
can
continue
down
that
road
of
no
module,
there's
no
modularity,
nor
is
it
a
complete
Euler
cycle,
and
so
that
might
be
a
non
bio
phenotype.
In
this
case
we
have
modularity.
But
then,
if
we
add
an
ax
here,
we
add
in
like
a
triangle
at
the
bottom.
You
can
see
a
here,
but
if
you
add
in
a
triangular
motif,
then
you
get
this
negative
one
and
you'll
get
modularity
because
you
can
recover
the
Euler
completeness.
A
It's
still
I've
still,
that
I've
had
a
lot
of
trouble
kind
of
defining
the
conditions
of
the
rules
in
terms
of
like
replication
or
you
know.
When
do
you
switch
to
a
new
shape
or
things
like
that?
Those
are
hard
to
map
to
the
biology,
but
I
think
this
is
just
the
same
thing,
but
I
finally
won
to
end
with
this
mention
of
on
growth
in
form.
So
there
was.
A
This
is
a
book
that
was
written
by
Darcy
Wentworth
Thompson
was
a
morphologist
in
a
developmental
biologist
about
a
hundred
years
ago
or
more,
he
wrote
a
book.
The
hundredth
anniversary
of
his
book
was
I
think
last
year,
where
he
it's
a
thick
book
and
in
this
book
he
talks
about
sort
of
a
mathematical
structure
of
seashells
and
other
organisms.
A
He's
famous
for
using
mathematical
transformations
will
look
at
different
phenotypes
and
fish.
So
you
have
these
different
fish
that
you
find
and
the
idea
is
you
take
a
single
fish,
a
bony
fish
and
you
put
this
grid
over
it
and
you
mark
the
landmarks
in
different
parts
of
the
grid,
so
different
intersections
of
the
grid.
Torus
correspond
to
different
landmarks
and
then,
when
you
put
a
different
fish
on
the
same
grid,
those
landmarks
are
misaligned,
so
you
realign
the
grid
according
to
where
the
landmarks
have
moved
to,
and
so
you
can
create
these
warps.
A
These
worked
grids
that
actually
described
changes
in
phenotype
across
fishes.
They
might
be
related
phylogenetically
or
they
might
be
very
different,
but
I
think
I
think
in
the
in
light
of
what
we're
talking
about
machine
learning
and
feature
selection.
But
this
is
an
interesting
thing
that
was
done
about
a
hundred
years
ago
and
when
he
did
this,
he
didn't
have
like
the
benefit
of
supercomputers.
A
He
had
to
do
it
by
hand,
and
so
this
is
really
I
think
a
remarkable
feat
considering,
but
we
don't
really
understand
a
lot
of
what
went
on
in
this
book
other
than
like
just
kind
of
what
he
displayed
I
mean
there.
People
have
tried
to
do
computational
for
ease,
and
some
of
this
there's
a
book
on
growth,
forming
computers,
which
is
a
about
20
years
old
now,
and
it
was
an
update
to
this
and
they
did
a
lot
of
morphogenesis
and
computers,
but
did
no
ones
are.
A
Oh
I
mean
I,
don't
think
people
have
looked
at
like
the
biological
significance
of
a
lot
of
it,
and
so
that's
my
talk
on
that
I
thought
that
would
be
a
good
talk
to
revisit
and
lighteth.
Well,
thank
you
for
the
+1
in
light
of
dicks
talk
a
couple
weeks
ago,
so
that's
I
mean
I
just
wanted
to
pull
up
that
slide
deck
just
to
show
people
what
was
going
on
with
that.
A
Not
sure
where
that
work
is
gonna
go,
but
it's
kind
of
like
you
know,
I,
don't
know
how
we're
gonna
push
it
forward,
but
it's
probably
something
there's
something
there
I'm
sure,
but
I
don't
know
what
it
is
and
then
it's
for
the
young
growth
inform
stuff.
You
know
this
is
something
that
we
can
talk
about
further
as
well.
There's
a
lot
there
actually
is
a
lot
of
like
work.
That's
been
done
between
like
mathematics
and
developmental
biology,
but
it's
it
was.