►
Description
DevoWorm meeting: February 24, 2020. Attendees: Richard Gordon, Vinay Varma, Devansh Batra, Thomas Portegys, Ujjwal Singh, Bradly Alicea, Yash Patel, and Jesse Parent
A
B
B
B
Okay,
welcome
to
the
meeting
today
hope
everyone
had
a
good
week.
I
was
talking
to
Richard
before
the
meeting
and
he
said
he
just
finished
his
the
paper
that
he
presented
last
week.
So
I
don't
know
if
he
wants
me
to
distribute
it
to
the
group
at
some
point,
but
so
will
still
be
discussing
that
you
know
in
subsequent
meetings,
maybe
at
the
end
of
this
meeting.
B
If
anyone
had
any
questions
or
comments,
they're
thinking
about
from
that
presentation,
thanks
again
for
the
presentation
last
week
this
week,
we're
gonna
turn
to
Tom
Portage's
who's
been
a
member
of
the
group
for
many
years
and
he's
gonna
present
on
his
his
he's
doing
a
lot
of
simulations
with
cellular,
automata
and
bio-inspired
modeling,
and
so
before
we
get
to
that.
I
wanted
to
share
our
paper.
I,
don't
know!
Maybe
time
we'll
talk
about
this
paper
in
the
in
the
hello.
B
B
A
journal
called
the
still
and
for
those
who
aren't
familiar
with
the
still
it's
a
machine
learning
journal.
That's
you
know
at
the
present
things
not
PDF
format,
but
they
present
them
in
this
interactive
format.
So
it's
an
actual
journal
where
people
publish
these
interactive
pieces.
So
let's,
let's
talk
about
this
for
a
couple
minutes,
so
this
is
a
paper
called
growing
neural
cellular
automata,
it's
called
the
subtitle
is
differentiable
models
of
morphogenesis,
and
so
this
this
Journal
distill
is
actually
a
machine
learning
journals.
B
So
this
is
a
little
bit
different
than
what
they
usually
publish.
They
usually
publish
things
in
more
mainstream
machine
learning,
but
this
is
a
morphogen
morphic,
genetic
approach
to
complexity
and
the
authors
here.
A
couple
of
authors
like
eleven
is
the
last
author
on
this
paper
and
he's
actually
well
known
in
the
developmental
biology
and
morphogenesis
community.
So
this
is
the
paper
I'll
put
a
link
to
it
in
the
chat.
B
Ok,
yeah,
if
I'm
says
distill
is
cool
this.
It
is
ok,
so
that's
the
link,
and
so
this
is
the
paper.
So
it
starts
off
with
this
animation.
So
the
way
this
works
is
that
you
can
write.
Let's
start
over,
you
can
set
the
speed.
You
can
set
your
faint
your
target
shape,
so
you
can
start,
maybe
with
a
a
gecko
here
or
a
lizard.
B
You
can
start
with
a
smiley
face,
a
human
eyeball,
a
fish,
a
Christmas
tree,
a
pretzel
I
mean
they're,
trying
to
be
cute,
but
they're
also
trying
to
cover
a
lot
of
different
shapes.
Well.
Why
don't
we
settle
on
this
butterfly,
so
you
can
settle
on
this
target
morphology
and
the
idea
is
that
you're
gonna
take
this
cluster
of
pixels
and
you're
gonna
run
a
simulation
that
forms
this
morphology,
and
so,
if
we
go
back,
we
reset
it
and
it's
starting
to
grow
at
a
very
slow
speed.
B
B
C
B
Yeah,
so
that's
that's
the
regenerating
regenerating
actually
has
like
where
you
can
cut
it,
and
then
it
reforms.
So
what?
How
does
this
do?
How
does
this
work?
So
this
is
actually
you
can
try
this
in
a
collab
notebook
for
those
of
you
are
interested
in
that.
So
if
you
want
to
download
the
notebook,
it
gives
you
the
code
and
everything.
So
how
does
it
work
and
then
they'd
so
in
the
paper
they
get
into
like
how
this
works?
B
So
you
get
a
description
of
what
you
know
morphogenesis
and
that
it's
an
example
of
self-organization,
so
they're
using
cells
as
the
building
blocks
of
these
bodies
and
they're
using
rules
and
those
rules
are
very
similar
to
what
you
find
in
a
cellular
automata
and
they
walk
through
the
model
here.
So
here's
the
model
where
they
have
a
number
of
different
steps
in
the
model.
They
actually
is
a
pretty
advanced
model
and
it's
actually
more
than
just
a
cellular
automata.
B
There's
a
lot
of
machine
learning
involved
in
this
as
well
a
lot
of
encoding
and
decoding
of
things,
and
so
that's
I
mean
I'm
not
going
to
go
over
that
right.
Now,
but
you
can
read
about
that,
and
then
they
have
a
way
to
determine
the
cell
state.
They
have
your
cellular
automata
rules
and
yeah.
They
kind
of
walk
through
that,
and
then
they
have
their
different
experiments
that
they
profile
in
this
article,
so
learning
to
grow
experiment.
B
One
experiment,
two
is
what
persists
exists
and
then
experiment,
3
is
learning
to
regenerate,
and
so
they
kind
of
walk
through
all
of
this
yeah.
Oh
there's
an
experiment
for
rotating
the
perceptive
field,
so
each
of
these
cells,
I,
guess,
are
perceiving
things
in
their
environment
and
they're,
responding
to
it.
So
Michael
evan
has
actually
written
a
lot
about
decision-making
in
cells
and
so
he's
interested
in
that
area.
It's
they
kind
of
fit
feeds
into
this
work
as
well,
so
related
work.
B
He
talks
about
PD
es
which
are
partial
differential
equations
and
their
role
in
this
process,
now
all
networks
and
self-organization
and
swarm
robotics.
So
this
is
again.
This
is
probably
going
to
be
of
interest
to
a
lot
of
people
in
this
group.
I
would
go
through
it
and
maybe
if
someone
wants
in
those
times
gonna
talk
about
in
this
talk,
maybe
someone
could
present
it
on
its
own
because
it
probably
deserves
that
kind
of
attention.
B
C
C
C
It's
it's
a
longer
wise.
You
know
that
sort
of
player
of
what
you
just
saw
there,
where
you
have
various
objects,
kind
of
growing
into
other
shapes.
So
what
this
does?
Is
it
a
set
of
rules?
If
basically,
the
way
it
works
is
you
can
enhance
of
the
pinyin
morphogenesis
process
and
more
facade,
watches
that
and
henna
developed
some
of
their
comic-con
rules
and
using
those
rules
it
can
actually
recruit
the
more
colorful
genesis
process
and
you
can
you
can
you
can
even
feed.
C
C
C
A
graded
and
it
has
different
cells
in
different
states
and
the
way
this
works
is
you
have
a
basically
a
pyramid
of
expanding
in
space
neighborhoods,
so
you
can
have
an
immediate
one
which
is
like
the
morning,
rod
which
is
3,
2
3,
and
then
that
would
then
have
a
bunch
of
a
few
neighborhoods
in
a
neighborhood.
But
it's
actually
pretty
good
trading.
Exactly
it's
just
bigger,
and
this
is
the
aggregation
speeds.
You
don't
actually
get
this
effeminate
Oriole
explode
from
using
the
kind
of
a
histogram
cell
type.
C
C
C
C
C
C
C
C
C
I
wonder
how
this
looked
a
little
cracker
than
it?
Should
it
okay,
so
this
is
actually
the
work
was
bizarre
after
it
has
learned
how
to
do
Eagle
upper
pattern
by
watching
the
Turing
reaction.
You
diffusion
equations
that
work,
and
you
can
see,
comes
up
with
the
sort
of
it
will
come
and
get
better
and
better,
although
it
could
get
even
more
better
if
I
play
with
it
tomorrow,
you
can.
C
C
C
How
many
life
cell
regeneration
evolution
of
very,
very
rudimentary
calculation
sequence,
which
is
an
embryo?
Is
this
I'm
sure
you
know
more
about
this
and
I'm,
not
a
biologist
and
are
they
are
machine
learning
person,
but
I
have
interested
in
computational
biology?
So
it's
like
the
cancellation
is
where
this,
where
you
have
a
embryo
in
it
like
is
this.
C
C
C
C
You
need
you
saw
this
with
the
Lina
picture.
You
can
learn
to
generate
a
whole
image
from
like
her
rough
sketches
relation.
We
have
shown
what
it
wants,
what
to
do
in
many
festivals
visit
the
escalation
super
simple.
This
thing
it
starts
out
of
the
square
which
is
in
I
think
this
cream.
Each
one
does
and
oh
we're
finding
we're
neuron
will
send
out
exons
to
target
Redis,
and
if
you
wanted
to
do
that
again
and
okay,
so
that's
pretty
much
what
I
want
to
say
about
MorphOS?
Oh,
so
that's
the
technique
that
it
uses.
C
A
C
C
C
C
Sort
of
static-
it's
dynamic,
that's
on
space,
but
animals
do
things
in
time,
so
this
behavior
is
Federer
goes
over
time.
It
changes
in
the
the
fish
is
changing
location
and
it's
moving
on
and
time
and
so
I
developed
this
kind
of
a
dated
sister
called
muffled
noises,
I
like
to
come
up
with
names.
It's.
C
More
venosus
me
tease
the
shape
of
knob
and
I'd
call
it
the
shape
of
nitrogen
spaces.
Higher
and
I
chose
not
only
time
but
also
space,
because,
as
you
might
be
wary
mammals,
well,
someone
could
a
lot
of
animals.
Mittens
are
one
of
them.
I,
don't
know
a
lot
of
other
animals
besides
animals,
but
we
have
these
special
structures
and
their
brains
called
root
cells
and
there's
other
ones
called
play.
Cells
which
are
specifically
wired
to
do
spatial
geometry
processing,
so
her
name
is
Randall
navigate.
C
Then
I
take
a
gauge
of
the
physics
of
the
world
that
has
a
tool
or
cleaning
geometry.
If
you're
to
be
a
German
on
a
flat
surface
and
use
that,
as
you
know,
so
it's
his
stuff,
that's
wired,
to
view
a
lot
to
leverage
the
fact
that
world
the
world
is
a
special
place
as
well
as
temporal
place,
because
most
now,
no
letters.
No
it's
just
the
temple
essentially
Alesi
at
your
current
normal
looks
that
I'm
interested
in
it.
So
here's
ten
a
picture
does
see
any
brain.
Behavior
again.
C
C
A
C
C
So
you
and
histogram
or
a
centroid
other
that's
feature
so
that
they
haven't
been
there
chunk
of
space-time
and
okay,
don't
say
thing
plug,
listen
to
it
now
in
our
open
nor
minute
thanks,
you
get
you
concisely,
because
you
get
to
the
fuzzy,
be
ability
to
get
noticed.
So,
let's
go
on
to
a
demo
of
this
next.
C
C
C
C
C
C
C
A
A
C
I'm
not
doing
this
on
because
we
don't
have
a
sign
in
our
toilet
on
fire.
So
it's
just
going
to
do
today,
making
mental
lamb,
which
is
more
than
any
bees,
actually
do
smaller
than
anything,
but
it
also
does
so.
The
direction
is
of
its
the
ants,
those
which
direction
the
beef
should
go
and
then
the
length
of
this
central
wiggle-waggle.
Thank
you
cuz,
you
know
her
go
and
then
the
bees
go
fly
off
and.
A
C
C
A
A
A
C
This
task
is
very
hard
for
a
conventional,
normal
heart
to
do
unless
it's
given
him
elicits.
You
know
the
potentially
so
much
to
start
like
compromises
so,
for
example,
like
a
very
popular
recurrent
Network,
which
called
long
short-term
memory,
but
if
you
try
to
get
a
long
trip
in
memory
to
serve
dead
reckoning
navigation,
to
figure
out
to
keep
track
where
the
hive
is
so
when
it
does
blunder
on
an
exterior,
it
immediately
gets
lost,
I
mean
lsdm
that
were
clearly
really
really
powerful,
but
they
can
handle
this
very
well
and
I.
C
Had
my
researcher,
she
seems
to
point
my
tractor
and
I've
actually
run
this
I
saw
his
brother,
Cody
I
had
felt
this,
and
indeed
he
gets
the
case.
So
you
need
some
kind
of
help
you
need.
Some
kind
of
animals
have
the
same
where
they
actually
have
some
emotions,
more
determination
of
space,
and
that's
what
you
need
to,
because
this
beauty
and
just
a
general
note
this
is
kind
of
where
I'm
at
is
animals.
C
A
C
C
B
Well,
thank
you
for
that.
That's
very
good
and
I'm
sure
we
have
a
lot
of
questions
about
what
you've
been
doing
and
I
I,
especially
like
the
comment
about
the
neural
networks
being
in
you
know,
unable
to
do
certain
things
that
you
know
we
people
talk
about
like
there's
this
debate
about
how
neural
networks
are
like
the
brain
or
not
like
the
brain
or
how
its
you
know,
able
to
generate
intelligent
behavior.
B
But
you
know
these
are
like
very
specific
things
that
these
models
are
being
trained
to
do
so,
like
you
know,
people
talk
about
like
how
artificial
intelligence
can
play
jeopardy
and
win,
but
it's
a
very
specific
type
of
task
that
they're
doing,
and
you
know,
there's
a
lot
more
to
behavior
than
just
like
you
know.
A
narrow
set
of
you
know,
like
maybe
an
expert
system
type
of
croch
or
some
sort
of
like
you
know,
a
pattern.
F
C
C
C
B
I
think
this
was
Gordon.
This
was
about
the
growing
cellular
automata.
He
said
it
contrasts
with
the
cybernetic
embryo,
so
I
know
people
here
probably
I've
heard
a
little
bit
about
the
cybernetic
embryo,
but
this
is
a
series
of
two
papers
that
stone
has
written
with
one
with
Richard
and
one
with
myself
and
Tom.
We
talked
about
like
cybernetics
and
how
that
kind
of
you
know
that
kind
of
approach
can
you
know,
explain
something
about
morphogenesis
in
the
embryo.
H
In
the
cybernetic
area
approach
is
that
the
everyday
does
not
have
a
Code,
Red,
okay
and
a
global
sense.
What
it
does
is
it
sets
up
that
a
system
which
has
a
specific
goal,
and
then,
when
that
goal
is
achieved,
it
includes
the
construction
of
two
more
cybernetic
systems.
So
you
end
up
with
a
tree
of
cybernetic
systems
which
results
on
embryo
and
that's
very
different
from
any
scheme
which
had
like
four
women
to
sort
11th
strikes
being
directed.
G
F
H
E
H
The
the
rules
are
very
localized
and
insensitive.
Each
each
system
has
its
only
goal
point
which
accomplishes,
but
at
the
end,
the
goal
that
includes
the
construction
of
the
next
cybernetic
system,
and
so
in
fact,
marvelously
constructs
as
paragraphs,
so
you
get
a
by
figure,
could
keep
you
treated
under
data
systems?
H
H
Will
start
first
when
you
get
some
proliferation
to
get
more
than
one
set?
Okay,
so
you
get
a
few
at
least
know.
Maybe
couples
or
whatever
you
got
a
pilot
cells.
No
happiness!
Is
you
get
a
pair
of
ways
we've
gone
over
the
organist.
We
call
it
to
accentuate
those
ways
propagates
through
the
cells
to
a
subset
of
the
cells
and
change
the
so
students
subtype
when
the
wave
is
done.
H
H
H
H
H
C
H
H
H
A
H
To
imitate
those
patterns,
so
we
had
a
a
chapter
in
the
book
and
this
is
coming
out.
Diatomic
matters,
a
tiny
pellets,
maybe
I'll
send
that
to
you,
it's
known
as
a
person
she's
a
much
better
mathematician.
I
am
so
it's
a
little
bit
steep
on
the
myth,
but
the
might
seem
that
way.
Actually
take
your
simulation
for
the
puffer
fish
and
started
acting
in
centric
diatoms
yeah.
H
C
F
C
G
E
H
G
C
B
And
so
this
is
a
very
interesting
topic.
Yeah
thanks
for
the
presentations,
so
will
we
might
revisit
this
in
subsequent
weeks?
I've
people
are
interested
in
talking
about
it
further,
please,
you
know,
chat
message
me
or
Tom.
We
can
talk
about
some
other
issues
related
to
this
I.
Think
it's
a
good
thing
to
follow.
Up
on
next
week
we
have
I,
think
Jesse's
been
and
a
half
his
he's,
gonna
record
his
presentation
on
the
free
energy
principle
and
we're
gonna
present
it
next
week,
sort
of
on
video.
B
Theory
of
the
brain
or
something
so
Jesse's
willing
to
wait
into
that
you're
gonna,
pretty
good
presentation,
looks
pretty
good,
so
doing
that,
and
so,
but
otherwise
have
a
good
week.
Remember
we
still
have
spots
open
for
presentations,
so
people
are
interested
in
presenting
yeah
wait
right
in
please.
Let
me
know
otherwise
of
a
good
week.