►
From YouTube: DevoWorm (2023, Meeting #22):
Description
GSoC Updates (devolearn @ onrender.com, SAM model manuscript, L1 phase C. elegans training data, and FR-CNNs). Overview of SynBio 2023. Expansion microscopy of connectomes using a synchrotron. Follow-up on the Gray-Scott model, chemical pattern formation (autocatalysis), and collective behavior as discrete n-wise interactions (automata, graphs). Attendees: Sushmanth Reddy Mereddy, Jyothi Swaroop, Susan Crawford-Young, Himanshu Chougule, Bradly Alicea, and Richard Gordon.
B
C
B
Yeah
I
did
a
pretty
good
one.
Last
week,
I
haven't
done
anything,
but
this
week
this
was
quite
good
Bradley
did
you
saw
that
I
have
run
so
many
times,
hugging
face
spaces
because
actually,
two
days
back,
they
are
not
working.
There
is
a
burger,
so
it
tries
to
implement
them
again
and
rewrite
the
books.
B
C
That's
good
yeah.
How
are
you
oh.
C
D
D
C
B
Due
to
some
Library
issues,
so
I
changed
everything:
I
added
notebooks,
also
training
notebooks,
which
was
written
by
Mayu
and
I
changed
this
requirement.txt
and
yeah.
These
are
the
sorry
these
are
the
commits
I
have
kept
last
week
in
this
report.
Like
this,
all
models
are
have
stopped
working
everything,
this
nucleus,
segmented,
membrane,
segment
or
so
I
made
them
already
on.
My
work
for
the
last
couple
of
week
was.
B
B
B
C
B
Of
course,
this
is
the
code
actually
I
was
working
on;
almost
it
is
over,
I
should
just
need
to
add
the
data
set
and
start
training
it
and
little
bit
of
hyper
parameter.
Tuning
also
needed
for
this
thing,
and
your
this
is
a
kind
of
extra
work.
I
did
your
brandly,
IP
I
hope
you
remember
this
thing
made
by
my
neck.
Sorry
fast,
pass
G
songs,.
B
It
is
stopping,
and
there
is
a
problem
with
re-running
them
all
so
to
clear
that
I
have
took
a
server
which
was
on
render.
This
is
the
link
of
it.
I
hosted
them
on
render
and
they
are
working
actually,
but
they
need.
There
are
little
bugs
I
need
to
solve.
If
those
bugs
are
solved,
we
can
keep
this
working
like
as
fast
and
our
table
warm
organization
in
hugging
page
also
feature
I
mean
we
can
keep
them
running
and
Sam
model
oil
will
also
be
ready
by
next
week.
B
I
just
like
I
just
need
to
start
them
training.
Naturally,
that's
not
left.
These
are
the
notebooks.
Actually,
last
week,
I
haven't
worked,
but
not
past
week
before
past
week,
I
haven't
worked
that
much
so
I
didn't
took
that
as
a
first
week,
one
of
my
trees
of
credit
I
took
this
week
as
my
juice
of
Groupon
period
and
I'm
gonna
write
the
blog
about
Waterloo
I
work
at
work.
B
This
is
the
work
I
did,
maybe
by
next
week
segment.
Anything
model
will
be
ready
and
after
that,
I
will
start
working
on
the
devonet
model,
which
I
have
proposed
that
I
need
to
update
I.
Actually
I
wrote
the
code
for
it,
but
it
is
in
Python
3.7
version.
I
need
to
update
this
to
python
related
version.
B
B
C
D
That
sounds
great
yeah.
It
sounds
great.
You
know,
have
any
idea
why
the
hugging
face
repository,
isn't
working
or
wasn't
working
I
mean?
Is
it
like
a
problem
in
there
on
their
end.
C
B
I
need
to
update
those
all
dependencies
in
the
hugging
test
when
I
updated.
It
started
working
actually
two
days
back
after
that
I
hosted.
This
render
thing
also
I
need
to
make
some
small
changes
in
this.
Also
because,
when
I
click
here
run,
it
is
not
working,
see
I
need
to
rewrite
some
things:
yeah,
it
is
working
yeah,
it's
working,
sorry,
but
there
are
so
many
deprecated
things
in
here.
So
yeah
Bradley
I
will
add
this
repo.
After
changing
some
code
in
it,
I
will
add.
B
This
refer
into
the
main
developer
and
I
will
keep
this
link,
as
the
who
can
visualize
them.
I
can
see.
Hugging
face
will
be
like
another
platform
who
can
see
our
Transformer
spaces
models.
This
will
be
like
official
anyone
can
open
and
see.
This
will
not
stop
like
a
clean
face
thesis
in
the
two
days
after
three
days
after.
D
B
I
didn't
got
anything
this
message
about
this
deprecated.
Suddenly
I
opened
hugging
face
for
posting
segment.
Anything
one.
Next
time
when
I
saw
these
three
models
are
not
working
when
I
started
digging
out.
What
is
wrong
going
on
so
when
I
encountered
I
saw
like
these.
All
libraries
are
duplicated.
B
I,
don't
have
idea
to
house
to
start
a
paper.
I
mean
we
haven't
returned
to
know
anything
in
new
paper.
How
to
write?
Do
you
have
any
guidelines
or
some
kind
of
instructions?
We
could
work
on
that
in
order
to
write
as
abstract
it's
starting.
But
after
that,
what
structure
we
need
to
follow?
We
don't.
D
Have
an
idea
about
it:
well,
yeah,
so
the
paper
is
going
to
be
on
the
method
right,
mainly
yeah
yeah,
so.
D
D
You
know
and
and
I
can
help
you
with
the
significance
and
then
going
to
the
methods
just
have
like
you
know,
step
through
the
methods,
one
at
a
time.
You
know
what
it,
what
is
it
involved?
Does
it
involve?
You
know
these
libraries?
What
do
they
do?
How
do
they?
How
are
the
dependencies?
D
How
are
they
related,
and
you
know
that,
like
sometimes
you'll
put
in
like
a
demo
or
a
couple
figures
to
show
that
you
know
what
what
you're
talking
about
in
the
paper
so
oftentimes,
you
know,
I,
don't
know
if
you've
ever
seen.
What's
the
journal
that
used
to
run
distill,
distilled.pub
they've
written
some
nice
interactive
papers
where
they
Implement,
it's
basically
a
method,
a
coding
method
for
something,
but
they
do
it
like
where
they
put
the
demos
interactive
Demos
in
the
paper,
but
it's
written
like
a
paper.
B
B
C
D
Sounds
good
I
look
forward
to
it
yeah.
Thank
you
for
the
update
and
kind
of
explaining
everything
and
yeah.
It
happens.
A
lot
where
you
get
things
that
are
deprecated
and
like
I
know,
GitHub
will
send
alerts
sometimes
too
much
too
many
alerts
for
different
things.
But
that's
you
know.
That's
where,
like
sort
of
the
maintenance
part
comes
in
that's
why
we
kind
of
you
know
we
have
projects.
D
You
know
every
summer
we
try
to
get
people
into
to
work
on
the
platform
to
improve
the
platform,
but
also
to
maintain
it
make
sure
things
are.
You
know
working
because
you
get
a
lot
of
over
the
sort
of
we.
We
don't
have
like
a
dedicated
Workforce
on
on
the
platform.
D
So
when
you're
developing
open
source
software,
sometimes
you
have
to
you
know
just
rely
on
people
to
come
in
and
maybe
work
you
know
come
in
for
a
while
and
and
work
on
something
and
you'll
find
that
they're
broken
things
in
the
code
and
things
like
that.
Sometimes
things
will
break
just
at
random
if
a
dependency
is
hit
wrong.
So
it's
always
a
struggle
to
keep
that
up.
D
D
Yeah,
it's
usually
the
best
way
to
do.
It
is
to
robot
through
automation,
so
and
then
yeah
I
look
forward
to
seeing
GOC
as
well.
If
he,
you
know
what
he
can
do
on
that
and.
D
That's
wonderful,
yeah,
congratulations,
DLC,
but.
C
B
Which
I
found
like
really
interesting,
I'll
start
by
showing
that
for
us?
So
basically,
this
is
a
3D
nuclear,
instant
segmentation
data
set
of
similar
to
us
like
the
fluorescence
microscopy
volumes
of
C
elegans
so
like,
and
it
is
like
a
three.
B
It
is
a
this
paper
is
like
an
iteration
of
previous
papers
in
which
the
first
paper
also
about
the
data
set
and
in
the
next
paper
they
did
like
augmentation
on
it
like
data
augmentation
like
just
before
doing
the
segmentation
part,
and
in
the
this
paper
they
are
basically
finished
like
at
the
entire
data
set
and
given
like
the
directory
wise,
like
test
stream
test
screen
and
validation
and
data
for
it.
B
So
I
try
to
visualize
what
it
was
given
like
similar
to
the
four
which
was
written
last
week
and
basically
the
data
options.
So
it's
like
a
huge
line
and
it
has
the
entire
verb
and
that's
how
it
visualizes
like
these.
Are
you
know
the
individual
cells?
B
So
this
was
like
what
I
started
with,
and
this
is
basically
the
masks
for
the.
B
B
I'll,
try
to
run
my
model
on
this
as
well.
I'll
just
have
to
resize
the
images
and
do
a
few
Peaks
and
I
also
worked
also
simultaneously
on
the
on
the
cell
membrane
data
like
the
basic
pre-processing
object
and
kept
it
ready,
so
that
I
can
use
it
and
segmentation
data
file
and
also
I've
been
working
on
like
like
mass
star
CNN
model,
which
is
basically
like
it's
similar
to
oh,
something
called
as
a
faster
C.
It
basically
does
instant
segmentation
in
a
way
which
is
like
it.
B
It's
basically
a
concatenation
of
two
models.
You
could
say
it's
like
a
faster
CNN,
faster
rcnn
model
and,
along
with
that,
fully
connected
convolutional
neural
networks.
So
it's
an
fcm.
So
what
it
does
is
it's
like
it's.
It
does
pretty
much
everything
data
faster,
our
CNN
does,
except
of
the
region
of
interest,
is,
like
you
know,
faster,
our
CNN
that
uses
Roi
cooling,
but
in
this
it
uses
Roi
align,
which
basically,
the
changes,
the
slight
values.
B
It's
like
the
entire
thing,
and
due
to
this,
we
get
the
object,
detection
part
of
the
code,
and
after
that
we
use
the
NFC
ends.
So
to
get
the
segmentation,
so
this
is
how
instances
segmentation
is
done,
and
it
is
one
of
the
state
of
the
art
models
which
has
been
used
for
a
long
time.
Also.
I
have
to
pass
up
models
ready
on
which
is
like
on
having
faces
and
I'll
try
to
implement
that.
If
this
doesn't
work.
B
I
try
to
find
out
what
it
is
about
and
I
also
came
across
a
few
more
papers
and
interesting
stuff
which
I've
just
kept
it
in
this
file.
I'll
just
share
it
with
you
ever
find
interesting.
B
So
in
the
so
hopefully
in
this
week,
I'll
try
to
complete
the
instant
segmentation
model
and
the
subsequent
weeks
I'll
do
the.
C
D
That
sounds
good,
so
this
new
data
set
that
you
found.
Could
you
pull
up
the
zenoto
Hub
or
the
zenodo
stop
again.
D
Yeah
there
it
is
3D
nucleus
and
segmentation.
Fluorescence
Mike
cross
could
be
volumes,
so
this
is
28
volumes
of
C
elegans,
L1
stage
and
corresponding
Stacks.
So
this
is
where
the
worm
is
hatched,
but
it's
still
like
a
larva,
so
it
has
the
anatomy
of
the
of
the
worm.
Basically,
you
know
from
head
to
tail
it's
unfurled
now
and
then
28
raw
images
and
corresponding
masks
of
average
Dimension,
which
is
the
thing
that
they
Define
here
pixel
size.
It's
pretty
fine
range
then
microscopy.
D
This
is
the
citation
yeah
I,
don't
think
I've
I,
don't
know
I'm,
probably
familiar
with
it,
but
I.
Don't
remember
right
now,
then
these.
So
these
are
nuclei
masks
so
for
the
whole
worm.
So
it
gives
you
some
anatomical
resolution.
I,
guess
you
know
what
cells
are
at
the
head.
What
cells
are
at
the
tail.
D
Yeah
well,
I
mean
you
know,
some
of
them
I
think
might
be
from
judging
from
the
name
here.
D
It
might
be
like
different
samples,
because
it
looks
like
they're
like
kind
of
similar
stages,
but
some
of
them
are
a
little
bit
different,
like
4l1
I
mean
they're,
they're,
different
I,
think
they're
different
markers
Maybe
like
elt3
hlh1,
and
it's
just
where
they
pick
a
certain
Gene
to
express
and
and
get
the
fluorescence
data
out
of
it.
So
you
know
you
have
to
pick
a
gene,
that's
expressed,
maybe
in
certain
cells
or
in
all
cells.
D
So
you
know
that
it'll
it'll
fluoresce,
when
you,
when
you
put
it
under
a
microscope
and
so
that
you
know
they'll,
do
that.
Sometimes
that's
what
the
first
part
of
this
is
and
then
L1
is
the
develop,
is
the
stage
in
in
development.
So
that's
the
first
stage
after
it
hatches
and
then
the
number
after
I
guess
is
the
sample
number
I,
don't
yeah.
The
paper
will
probably
have
the
all
the
details
on
what
those
mean
so.
A
B
D
A
A
D
Yeah,
so
it
can,
it
can
identify
single
cells
or
because
they're
yeah
they
get
it
they
get.
The
fluorescent
signal
is
usually
pretty
clean
if,
if
there
isn't
a
lot
around
it,
but
sometimes
you
get
like
noise
and
now
it's
it's
something
that
just
happens
with,
because
you
know
you
have
tissues
underneath
sometimes
there's
what
they
call
autofluorescence,
where
you
get.
D
If
you
know
interactions
at
the
tissue,
sometimes
you
know
they're
very
close
together,
so
it's
important
to
make
sure
that
it's
training
on
single
cells
and
not
like
a
clump
of
what
will
be
interpreted
as
a
clump
of
cells,
because
you
have
this
fluorescent
signal
in
one
area,
so
I
mean
you
know,
it
looks
like
it's
probably
pretty
clean
I
mean
you
might
go
through
and
make
sure
that
it
looks
like
it's
picking
up
like
individual
cells
when
it
you
know,
goes
like.
D
D
It's
good
to
have
a
data
set
where
you
can
like
look
at
some
of
the
you
know
actually
have
real
features,
so
yeah
I
mean
I,
would
yeah
I
think
this
is
good,
so
it'd
probably
be
some
pre-processing
of
this,
though,
or
at
least
like
you
know,
making
sure
that
it's
well
prepped
and
then
you
can
train
your
model
that
you're
you're
trying
this
faster
rcnn
so
as
opposed
to
an
rcnn
which
basically
is
a
combinational
two
models
like
a
faster
RCA,
Network,
okay
and
then
you're
gonna
go
from
there.
D
C
D
So
yeah,
when
we
bring
all
this
together,
we'll
have
like-
and
we've
had
this
in
the
past,
where
we
kind
of
have
this
collection
of
different
models.
You
know
because
the
problem
here
is
that
we
want
to
segment
cells,
but
the
other
problem
is
that
you,
you
know:
there's
not
uniformity
in
cells
and
biological
cells.
You
have
a
lot.
You
know
a
lot
of
different
configurations
of
cells
in
a
in
an
embryo
or
in
a
developmental
phenotype,
and
you
know
sometimes,
depending
on
the
organism,
we
usually
train
it
with
C
elegans
data.
D
We've
tried
different
other
species
before
as
well,
but
this
is
largely
you
know.
We
use
see
all
against
data
sets,
and
that
has
some
limitations,
because
C
elegans
has
a
specific
developmental
progression.
It
has
certain
cell
types
with
certain
shapes,
so
you
know
if
you
want
to
go
say
to
a
drosophila
embryo,
you
may
or
may
not
be
able
to
use
the
software
very
well
to
segment
out
those
cells,
because
it's
a
different
process.
D
There's
a
lot
of
cellularization
there's
a
cellularization
stage
from
a
centitium
and
things
like
that.
So
there's
a
different
set
of
things
going
on
there,
but
yeah
this.
So
it's
nice
to
have
like
multiple
models
where
we're
kind
of
seeing
you
know
we're
doing
different
things,
and
then
we
want
to
be
able
to
put
those
together
and
eventually
maybe
you
know,
we
say
that
there's
a
you
know
a
pipeline
for
getting.
D
You
know
one
answer
and
if
that
doesn't
work
you
can
try
another
set
of
models
to
get
something
else,
and
so
we
give
like,
at
the
end
of
the
day
we
give
people
options
of
what
they
can
use
different
tools,
it's
kind
of
like
different
types
of
pool,
cues
or
something
you
know
when
you're
playing
pool
people
like
different
types
of
weights,
and
things
like
that.
So
it's
so
it's
good
to
have
multiple
models
for
multiple
tasks
or
different
specialized
tasks.
D
But
also
I
also
want
to
make
sure
we
don't
replicate
any
work
too.
So
I
think
right
now
we're
in
pretty
good
shape,
though
all
right
yeah,
thanks
for
the
updates,
looks
like
things
are
coming
along
pretty
well.
B
B
B
I
learned
like
to
integrate
ml
into
it,
using
by
implementing
on
US
models,
so
I
have
taken
so
I.
Have
this
going
to
write
the
code
by
somewhat
completed
the
code.
I
need
to
know
first
test
it
on
our
data
set,
so
I'll
use
the
last
year's
images
which
I
got
from
season
and
I'll
test.
If
it
somewhat
work
and
it's
a
positive
output,
then
I
can
use
even
use
a
new
data
set
to
get
a
better
route.
E
Yes,
I
just
sent
Bradley
more
images
and
I'm
going
to
put
some
up
on
sync
for
Redtube
or
deck
for
thank
you,
so
everybody
can
share
them.
Yeah.
The
second
egg
turned
out
fairly
well
and
well
so
the
first
egg,
not
so
well
so,
but
I'll
just.
A
Put
all
of
the
images
up
Frozen
you
want.
You
want
me
to
create
a
URL,
so
anybody
can
do
it.
E
I'm
sure
sure
I
made
them
to
share
so.
Okay
I'll
do
that.
Okay,
I
just
need
to
get
them
there.
I've
been
I,
have
a
large
meeting
tomorrow
with
my
advisors.
C
D
It'd
be
great
if
Hari
Krishna
could
get
it
his
hands
on
the
new
data.
So
if
you,
if
you
could
share
that
dick
share
the
link
to
Hari
Krishna
I,
think
well.
A
A
A
A
Okay,
fine,
send
me
Susan,
send
me
an
email
message
when
you've
done
it
so
because
so
sync
does
not
send
me
doesn't
tell
me
about
changes.
Okay,
all
right!
Yes,
okay,.
D
D
He
can
keep
working
on
the
thing,
but
yeah
I
mean
you
know:
it'd
be
nice
to
have
like
a
update
on
that
just
to
see
for
the
new
people
too,
who
you
know
people
haven't
been
here
last
year
if
they
could
see
what
what
it
is
that
you
did
last
year
and
how
you
want
to
improve
it,
that'd
be
good
I
mean
it
doesn't
have
to
be
today,
of
course,
but
it
can
be
like
in
the
coming
weeks,
presentation.
C
C
D
A
D
Right
so
yeah.
A
There's
only
one
thing
to
keep
in
mind
and
that
is
I've
got
over
a
terabyte,
but
anybody
who
wants
to
use
it
use
sync
for
free
is
limited
to
five
gigabytes.
D
D
Well,
yeah:
until
he's
getting
around
these
limits,
that's
the
yeah
yeah
all
right!
This
is
near
Ohio
Krishna.
Do
you
have
anything
else.
E
C
E
C
D
Good
so
I
guess
we'll
move
on
to
some
things
here
now,
I
I
promised
that
I
would
talk
about
this
Workshop
I
went
to
I
think
it
was
two
weeks
ago
now
and
so
I
can't
share
any
of
the
slides
or
anything
you
know
like
screenshots,
but
I
can
show
like
the
agenda
so
I'll
do
that.
Let
me
share
my
screen,
so
this
is
send
cell
2023.
This
was
a
virtual
conference
has
hosted
in
Minneapolis
Minnesota
at
the
University
of
Minnesota,
but
it's
like
you
know
it
was
online.
D
It
was
on
Zoom,
so
access
to
the
lectures,
so
they
basically
this
is
about
like
building
synthetic
cells
and
building
the
parts
of
synthetic
cells.
So
why
would
you
want
to
build
a
synthetic
cell?
Well,
you
know
you
want
to
build
a
synthetic
cell,
maybe
to
synthesize
biomaterials,
or
you
want
to
build
a
synthetic
cell
for
modeling
like
a
disease
state
or
something
like
that.
So
the
conference
was
interesting.
The
keynote
speakers
kind
of
got
into
like
why
you
would
build
a
synthetic
cell.
D
Some
of
the
recent
progress
that's
been
made
on
it
and
then
the
panels
that
they
had
or
the
the
the
sections
you
know
that
are
not
Keynotes.
Those
were
more
about
like
different
parts
of
how
to
do
synthetic
cells
and
build
them.
So
that's
really
interesting
conference.
D
If
you've
heard
of
the
Jay
Craig
Venter
Institute
they're
in
in
San
Diego
in
the
United
States,
and
they
build
a
lot
of
these
like
they
build
a
lot
of
like
genomic
tools.
Jay
Craig
Venter
ran
one
of
the
genome
human
genome
sequencing
projects
in
the
90s.
He
founded
this
institute
and
they've
been
working
on
not
only
building
genomic
tools
but
building
these
synthetic
cells.
So
they've
been
some
really
interesting
progress
in
this
area,
so
this
is
kind
of
from
the
conference
here.
A
D
D
Yeah
yeah
I
will,
let's
see
yeah
buildasell.org
there
we
go.
D
So
we
had,
let's
see
if
we
can
find
the
program,
so
they
had
these
sessions
and
they
were
kind
of
going.
They
had
like
people
worked
on
these
different
areas.
They
had
it
broken
down
into
these
cell
free
translation
systems,
the
cytoplasm
of
synthetic
cells.
They
had
two
sessions
on
that
and
some
posters
they
had
beyond
the
bench:
education,
Outreach,
ethics
and
biosafety.
This
was
set
one
session
of
the
poster.
D
They
had
something
called
containers
which
I
was
really
interested
in,
and
this
is
where
you
build
containers
like
lipids
and
and
what
they
call
coaster,
coaster,
Vates
and
microfluidics
microfluidics
being
test
systems.
You
can
test
your
containers
in
liposomes
being
like
vesicles,
that
you
can
pack
things
into
co-acerbates
or
these
sort
of
Aggregates
that
we've
talked
about
in
the
meetings.
So
these
are
things
that
have
to
do
with
building
a
container
for
yourself,
there's
a
lot
of
stuff
that
intersected
with
early
life.
D
So
we've
talked
about
early
life,
the
origins
of
life-
and
you
know
so.
Some
of
this
is
like
engineering
cells
for
the
you
know,
applications.
But
some
of
this
was
like
looking
at
how
life
might
have
started.
You
know
thinking
about
like
how
maybe
life
started
to
organize,
and
so
the
container
sessions
were
really
sort
of
focused
on
that
intersection.
So
it's
really
interesting
stuff.
There
were
five
sessions
on
containers
and
then
posters.
D
The
first
session
for
containers
was,
you
know
they
had
the
something
in
synthetic
organelles.
They
had
things
about,
like
you
know,
designing
synthetic
cells
for
to
protect
from
antibiotic
antibiotic
resistant
bacteria,
they've,
combining
models
of
membranes
and
intracellular
condensates
non-lamellar
lipid
compartments
for
artificial
cell
development.
D
D
Let
me
see
program
in
the
containers
sessions.
Then
there
was
building
blocks,
which
is
where
they
looked
at
metabolism,
gene
replication,
membrane,
synthesis
and
ribosome
assembly,
and
things
like
that.
D
So
these
are
like
the
building
blocks
of
a
cell,
like
what
kinds
of
things
need
to
be
there
for
a
cell
and
again,
this
is
mostly
like
for
building
cells
for
applications,
but
some
of
it
also
was
this
sort
of
idea
of
early
life
and
how
you
build
a
cell
or
or
some
sort
of
vesicle,
and
so
they
had
some
interesting
sessions,
so
Lynn
Rothschild,
who
works
at
Nasa,
talked
about
extremophiles
and
how
those
can
Aid
in
building
cells.
D
There's
this
project
to
build
a
cell
project
where
they
were
looking
at
building
synthetic
extremophiles,
which
are
bacterial
cells,
mostly
that
live
in
extreme
environments,
like
really
hot
environments,
were
really
oxygen.
Poor
environments,
things
like
that.
D
I,
don't
I,
don't
think
so,
but
they
I
think
it
was
like
one
of
these
student
organizations
where
they
were
built,
trying
to
get
like
a
hands-on
experience
with
like
synthetic
biology.
So
they
were
building
cells.
They
were
just
doing
these
experiments
where
they
were
trying
to
build
different
types
of
extreme
files.
You
know
and
they
call
them
Health
cells
which
are
like
they
have
they
live
in
hell.
You
know.
D
Then
there
was
this
artificial
cells
from
soft
matter
to
cell-like
behavior.
So
there
was
a
soft
matter
component
to
this,
which
is
that
you
know
you
have
these
cells
that
have
behaviors
and
they
behave
according
to
the
principles
of
soft
matter.
So
you
have
all
sorts
of
things
that
you
can
look
at
there.
D
D
They
had
a
lot
on
like
cytoskeleton
and
actin,
and
things
like
that
and
their
role
in
some
of
these
synthetic
cells
says
all
stuff
that
you
know
we're
very
familiar
with
from
the
meetings
and
they
had
an
applications
section.
Is
there
anything
useful
in
money
making
like
medicine
bioengineering
Etc?
So
basically,
if
you
can
commercialize
it
there,
that's
what
that
session
was
the
cytoskeleton
session,
which
was
protein
DNA
and
other
structures.
D
They
had
some
interesting
talks
on
like
protein
and
DNA,
like
engineering
like
structure,
secondary
structure
for
DNA
and
proteins,
and
things
like
that.
So
there's
interesting
stuff.
There
I
didn't
really
attend
any
of
these,
but
and
then
minimal,
sell
on
live
chassis,
which
is
Michael,
Michael,
Plaza,
plasma
and
others.
So
this
is
just
kind
of
putting
what
they
call
minimal
Sal,
which
is
what
are
the
minimal
parts
that
you
can
put
together
in
a
cell
and
then
having
this
in
a
like
a
living
cell.
D
So
the
session
was,
let's
see
if
they
had.
D
I'm
not
sure
this
is
the
these.
Are
the
posters
for
minimal
cells
on
live
chassis,
so
yeah
I,
don't
think
they
did
they.
They
had
tool
kits
for
genome
transplantation,
physiological
characteristics
of
a
genome
minimized.
One
of
the
things
they're
doing
at
the
Venture
Institute
is
they're
building
these
minimal
genomes,
where
they
knock
out
all
the
genes
that
aren't
necessary.
D
So
you
just
knock
out
one
gene
after
another
and
if
it
doesn't
kill
the
cell,
then
it's
not
essential,
and
so
you
just
get
a
map
of
like
all
the
essential
genes
that
you
need
and
that's
your
minimal
genome
and
then
you
can
put
genes
into
the
genome.
It's
a
circular
genome
for
a
bacteria,
and
you
can
put
genes
in
to
see
if
you
can
add
function
to
it.
D
So
you
know
you
can
characterize
the
minimal
genome
in
a
minimal
cell
you
can
put
in
genes
and
see
you
know
for
different
things
like
you
know,
synthesizing
different
chemicals
and
things
like
that,
and
then
you
can
do
all
sorts
of
other
things.
You
can
look
at
so
like
they.
In
this
case,
they
were
looking
at
cell
division
by
getting
rid
of
all
the
non-essential
genes
and
then
adding
in
new
genes,
so
they
were
talking
about
and
this
poster
they
have.
D
This
Craig
Venter
Institute
sent
sin
3.0,
which
is
this
model
of
a
synthetic
minimal
bacterial
cell,
and
this
is
derived
from
mycoplasma.
So
that's
why
they
talked
about
mycoplasma
and
then
this
this
genome
contains
only
473
genes,
they're
able
to
use
this
to
sort
of
study
the
basics
of
us.
What
you
know
what's
needed
to
build
a
cell
and
then
what
they
can
do,
then
is
add
in
genes
to
add
in
different
functions
like
cell
division,
so
for
a
minimal
Gene
genome
they
can't
they
don't
have
cell
division.
D
It's
just
sitting
there,
maybe
with
some
minimal
metabolism
and
things
like
that.
They
can
actually
take
these
minimal
genomes
start
adding
in
genes
that
we
know
are
involved
with
cell
division
and
we've
talked
about
Fitz
Z
in
some
of
the
meetings,
sep
F
and
some
other
genes.
So
you
add
those
in
and
then
you
can
restore
cell
division.
D
So
that's
really
interesting
that
you
can
like
get
it
down
to
that
minimal
genome
and
then
expand
it
out
a
little
bit
with
you
know
things
that
you've
identified
from
other
studies
that
are
involved
in
a
function
and
actually
add
that
function
in.
So
that's
a
good
way
to
look
at,
like
you
know,
basic
biological
questions
like
you
know,
how
do
you?
Maybe
how
do
you
get
cell
division
from
like
a
very
minimal
genome?
D
So
you
add
in
a
couple
genes-
and
you
know
it
works,
you
get,
you
can
restore
cell
division
or
you
can
restore
some
aspect
of
metabolism,
so
it's
kind
of
where
they
were
going
with
that
and
then,
of
course
they
did
a
lot
of
stuff
with
cytoskeleton
and
actin
molecules.
D
They
did
some
things
on
mechanical
Behavior,
a
positioning
of
ectomyosin
rings
and
vesicle
bleving,
so
they
were
able
to
look
at
some
of
these
sort
of
engineering,
minimal
sets
of
modules
instead
of
just
a
genome.
So
now,
if
you
want
to
have
like
self-division
in
a
Cell,
you
know
what
kinds
of
modules
do
you
need
in
the
cell?
So
do
you
need,
like
things
like
actin
molecules,
behaving
in
a
certain
way?
You
need
certain
functions
like
movement
or
you
know,
for
cell
division.
D
D
So
that's
that
was
that
conference
I
can't
share
any
of
the
screenshots
or
anything
because
it's
you
know
secret,
but
yeah.
A
Bradley
one
way
of
reducing
a
minimal
cell.
A
D
E
E
It's
as
far
as
they
know,
the
actors
are
acting
on
on
the
exterior
of
every
cell.
Okay,
the.
A
So
you
don't
know
I,
think
I,
think
in
pay
for
the
win
and
I
did.
E
Microtubian
rings
well:
I
just
I
have
a
paper
that
shows
that
acting
is
on
the
exterior
of
the
cell
result.
E
Well,
it
has,
it
has
to
be
in
attached
cells,
because
at
the
actual,
myosin
attachments
are
what
attached
cells
together
and
adherence
Junctions.
So
if
you've
gotten
in
the
hair
joint,
add
here
in
Junction,
it's
supposed
to
be
a
actin
myosin
complex
according
to
what
I
know:
okay
I'm,
not
a
biologist.
Okay,.
E
I
don't
know
trying
to
read
biology
papers
on
this
is
like
first
off
where's,
the
paper
about
it
and
second
off
what
what
are
they
talking
about
or.
E
A
E
But
I
also
have
a
paper
about
they
look
like
they're
cells,
intestinal
cells
and
the
ring
is
at
the
bottom.
E
Okay,
and
then
they
show
a
ring
at
the
top
I'll
find
the
paper
and
I'll
send
it
to
you
to
be
broken
yeah.
They
can
be
either
way
and
if
it's,
if
it's
a
Soft
Cell
it's
at
the
top
and
if
it's
a
non-elastic
cell,
it's
at
the
bottom.
E
E
Yes,
I
have
a
Bailey
with
me
this
morning
because
he
went
in
followed
somebody
here
and
now.
We
have
to
stay
here
with
me
because
he
wasn't
supposed
to
be
here.
Yeah.
D
Said
one
more
thing
to
mention
here:
this
is
actually
quite
interesting.
A
long
time
ago
we
talked
about
using
the
synchrotron
for
different
types
of
microscopy
and
so
I'm
going
to
put
this
in
the
chat.
D
So
this
is
feasibility
and
mapping
the
human
brain
with
expansion
x-ray
microscopy.
This
is
I,
mean
the
human
brain.
Is
this
one,
for
you
know
application,
but
basically
this
kind
of
goes
through
some
of
the
like
kind
of
thinking
through
how
this
would
be
used
for
doing
this
sort
of
Imaging.
So
this
is
Logan
Thrasher
Collins
he's
he
does
a
lot
of
stuff
in
in
biotech
and
Jesse
actually
knows
who
he
is
so
now.
D
Jesse
I
know
he's
not
here
today,
but
he'll
appreciate
this,
so
this
is
synchrotron
x-ray
microscopy.
So
synchrotron
is
actually
like
a
a
particle
accelerator
that
they
use
to
power
the
microscopy.
This
is
x-ray
microscopy
with
expansion
microscopy,
so
expansion
microscopy
is
where
they
put
in
proteins
into
the
sample
and
they
expand
the
sample.
So
it's
like
you
know
if
you
have
a
very
small
sample,
maybe
50
microns
in
size,
a
feature
that
you
want
to
look
at
you.
D
You
can
either
magnify
down
to
that
scale
and
hope
you
can
get
a
good
resolution
or
you
can
put
these
protein
proteins
in
the
sample
and
expand
the
proteins
in
size
and
it
blows
up
the
feature
size,
and
so
it
just
makes
it
bigger
so
that
you
can
see
it,
and
so
that's
what
they
do.
It's
a
relatively
new
technique
so
but
they're
they're
combining
these
two
things
and
they
want
to
look
at
whole
brain
human
brain
conicomics
at
the
nanoscale.
D
So
they
want
to
look
at
features
of
the
nanoscale
which
they're
going
to
be
too
small
to
see
with
conventional
microscopy
they're
coupling
the
synchrotron
X-ray
microscopy,
which
gives
resolution
and
then
the
expansion
microscopy,
which
gives
additional
resolution.
So
this
just
works
out
how
this
might
be
done.
I
thought
this
was
interesting
because
you've
talked
about
this
in
the
past
and
it's
very
much
a
proposal.
I
think
it's
there
isn't
really
any
there.
Any
images
yeah.
D
Well,
I
think
it's
probably
like
in
like
corpses
or
something
but
I
yeah
I,
don't
know
how
they
do
I.
Don't
how
they're
going
to
do
this,
but
so
yeah.
So
it
kind
of
works
on
some
of
the
technical
details
and
this
perspective
I
will
explore
balances
between
synchrotron
Optical
engineering
choices
and
expansion
Factor.
So
these
are
all
like
technical
details
proposed
methods
to
successfully
implement
this
in
the
context
of
human
brains
and
estimate
how
much
it
would
cost
to
image
the
human
brain
in
this
way.
So
this
is
yeah.
D
You
have
the
let's
see
so
I
guess
you
know
there
are
things
that
are
very
small
in
the
in
the
human
brain,
like
very
small
structures
like
synaptic
structures
or
other
types
of
like
extracellular
Matrix,
and
like
getting
a
really
good
set
of
images
that
show
those
things
and
how
they
interact.
D
So
you
know
we
have
like
a
lot
of
really
small
things
in
the
connectome,
like
you
know,
different
yeah
different,
like
ion
channels
and
and
synapses,
and
things
like
that
and
to
get
a
really
good
resolution
of
those
things
is
important.
I
guess,
but
you
know
again,
like
you'd
have
to
Target
these
structures
and
you'd
have
to
you
have
to
really
kind
of
get
into
the
tissue.
So
you
know
having
a
good
sense
of
like.
D
I
think
it's
tissue
slices,
so
he
talks
about
that.
This
is
a
promising
alternative
to
electron
microscopy,
which
is
another
method
that
they
can
use
to
look
very
at
very
small
structures,
and
then
this
modality
also
falls
short
when
considering
the
volume
of
the
human
brain,
particularly
after
expansion.
D
So
you
know
they're
doing
this
expansion
they're
doing
the
microscopy,
the
utilized
Eightfold
expansion
in
lattice,
light
sheet
microscopy
and
drosophila
Central
complex,
which
is
a
part
of
the
drosophila
brain,
these
three
colors
to
visualize
things
into
voxels
of
30
by
30
by
100
nanometers.
So
that's
a
pretty
small
voxel
area,
it's
just
the
area
in
which
they're
each
part
of
the
acquisition
at
you
know
you
have
so
it's
a
pretty
small
resolution,
but
even
accounting
for
the
Eightfold
expansion.
This
amounts
to
a
volume
of
less
than
0.5
millimeters
cubed.
D
So
these
are
very
small
volumes.
If
you
expand
it,
you
know
further,
you
can
get
a
bigger
volume,
but
it's
still
like
you
know
extremely
small
feature
size.
D
No
I
think
they
want
to
get
like
specific
places
in
the
brain
like
a
specific
locations
and
see
yeah
yeah
because
it
would
take
like
you
said.
You
know
in
this
example
of
the
drosophila
central
complex,
getting
this
kind
of
data.
If
you
you
know,
you
did
the
whole
brain
I
think
it
was
in
Mouse
brain
this
estimate
is,
you
would
take
the
876
years
to
acquire
the
entire
brain.
A
A
E
Is
sorry
there's
a
science
fiction
book
about
brain
uploading,
it's
called
Bob
or
the
Bob
verse.
A
C
A
Actually,
a
good
book:
oh
yeah,
okay,
Bradley,
have
you
tried
getting
AI
to
to
get
the
Google
money
instead
of
students.
D
D
A
D
Have
to
have
a
lot
of
infrastructure,
and
but
it's
all
particle
physics
in
here,
going
on
so
you're,
generating
a
you're
generating
electric
and
magnetic
fields
and
you're
applying
this
to
this.
This
problem
you
know,
and
then
this
is
the
actual
Imaging
device
where
you
have
a
sample
out
of
the
brain,
you're,
doing
I
guess
slices.
He
puts
a
brain
on
here,
but
that's
not
really
I,
don't
think
you're
gonna
get
it
out
of
like
just
a
brain
sitting
there
you'd
have
to
have
like,
and
then
you
have
the
insertion
invoice.
D
That's
the
X-ray
beam
being
generated
by
all
this
comes
through.
It
has
to
be
charged
up,
it's
like
super
concentrated
and
then
it
goes
through
the
insertion
device
to
this
monochromator
it
through
the
sample,
and
then
you
acquire
the
image
and
it's
then
that's
your
image,
and
so
that's
that's!
Okay,
because.
A
D
Mentions
tomography
and
last
part
here
so
then
he
recommends
the
methodologies,
and
you
know
one
of
the
hurdles
that
you
have
for
this
method
is
attaining
sufficient
feature.
Contrast
for
capturing
clear
images.
So
sometimes
you
have
to
clear
your
clarify
your
tissue,
so
you
have
a
lot
of
proteins
and
other
types
of
things
that
are,
you
know,
sort
of
in
The,
extracellular,
Matrix
kind
of
getting
that
removed.
So
you
can
actually
see
the
structures
you
want
because
they're
so
small,
so
getting
that
feature
contrast
is
really
important.
D
That's
where
you
get
like
some
well,
not
only
you
know
sort
of
inference
of
of
different
things
like
you
know,
but
you
also
have
to
have
like
all
sorts.
Well,
you
could
apply
a
lot
of
the
image
processing.
Technologies
we've
been
talking
about
in
the
meeting
to
this
problem.
You
know
finding
the
the
features
and
kind
of
extracting
them
from
a
lot
of
noise
in
the
image
so
yeah.
D
So
they
have
this
other
method
called
unclearing
microscopy
where
they
actually
use
chemical
means
to
like
clear
the
sample,
and
then
you
can
take
images
of
it.
So
that
helps
too.
But
you
know
having
some
combination
of
that,
so
this
is,
of
course,
the
expansion
part
where
they
expand.
So
this
is
a
brain.
They
expand
the
tissue,
then
they
stabilize
it.
They
clear
the
tissue
of
all
the
proteins
and
again
this
is
not
a
living
brain.
This
is
a
dead
brain,
but
they
have
to
you
know.
D
The
idea
is
they're
expanding
the
features
out
in
space,
it's
expanding
the
size
of
everything
and
then
they're
flaring
the
tissue
and
then
they're
stabilizing
and
imaging.
So
you
get
a
like.
This
is
a
motor
neuron
here
or
a
stellate
neuron,
I,
guess
and
then
it
gets
you
can
expand
it
you
stabilize
it.
Then
you
can
image
things
like
the
synapses.
You
know
things
like
different
ion
channels
on
the
sell
and
so
forth.
You
can
really
get
really
good
resolution
on
that.
So
how
fast
can
synchrotron
images
expanded
image,
expanded
brains?
D
We
talked
about
that.
It's
you
know
it's
going
to
be
a
slow
process
if
you
want
to
do
the
whole
brain,
but
but
we're
really
interested
in
these
very
small
features.
So
this
is,
you
know,
it
calculates
out
some
limits
for
the
number
of
projections
needed
to
completely
reconstruct
a
single
tomogram.
D
D
D
It
takes
time
and
you
know
so.
There
are
a
lot
of
technical
issues
to
overcome
aside
from
just
the
expansion
part
and
like
kind
of
getting
the
clearing
the
tissue,
so
he
offers
up
some.
You
know
Tech
a
lot
of
technical
details
here.
The
cost
is
pretty
high,
but
you
know
that's
that's
something
that
always
comes
with
imaging.
Okay,.
C
D
Dick
had
to
leave
looks
like
he
had
I
think
he
told
me
he
had
a
session
to
attend
so
and
hamanchu
had
to
leave
as
well.
So.
D
Very
good
Hare
Krishna
as
well.
That's
fine!
Thank
you
for
attending
and
thank
you
for
attending
people
were
left.
Do
we
have
any
comments
before
we
go
questions.
B
And
that's
it,
those
are
my
updates.
Maybe
we
will
start
I
will
share
the
talk
with
you
because.
D
D
D
Are
a
little
bit
different,
but
they
basically
both
use
a
couple
differential
equations
to
produce
an
output,
so
the
gray
Scott
model
was
first
proposed
in
the
Journal
of
physical
chemistry
in
1985,
and
the
title
of
the
paper
was
sustained
oscillations
and
other
exotic
patterns
of
behavior
in
isothermal
reactions.
D
So
this
is
actually
using
the
chemical
reaction
kinetics
and
looking
at
that
and
looking
for
things
like
sustained
oscillations
and
other
patterns.
So
last
time
we
talked
about
the
gray
Scott
model
in
the
context
of
microorganisms,
using
their
flagellum
to
produce
waves
through
a
medium
and
those
waves
interacting
and
producing
patterns.
D
There
are
other
things
like
the
Bella
sanjabatinsky
reaction,
where
the
bz
reaction,
which
occurs
during
a
set
of
chemical
reactions
and
produces
highly
ordered
patterns
such
as
spiral,
waves
and
other
types
of
things.
So
this
is
the
first
paper
that
proposed
the
gray
Scott
model
as
kind
of
walks
through
the
model
talks
about
chemical
reactions
in
open
systems,
so
we're
interested
in
in
biology
and
in
natural
systems
in
open
thermodynamic
systems.
So
these
are
things
like
streams
and
fires
living
organisms
in
single
cells.
D
Related
to
that
is
this
concept
of
a
dissipative
structure
or
a
dissipative
system,
which
is
where
you
have
an
open
system
and
the
spare
energy
or
the
products
of
the
energetic
flow
produce
patterns.
So
you
can
think
of
a
flowing
river
where
you
have
the
riverbed,
which
is
the
structure
that
results
from
the
flow
of
the
river,
the
flow
of
energy
through
that
system,
and
so
on
that
structure,
then
you
can
have
things
like
trees
and
other
types
of
things,
and
so
you
have
structure
where,
as
before,
you
didn't
have
any.
D
They
also
talk
about
autocatalysis
as
isothermal
feedback,
so
the
feature
common
to
nearly
all
isothermal
oscillatory
reactions
is
Auto,
catalysis
and
so
autocatalysis
for
sure,
not
only
by
the
bellasoft
javitinska
reaction,
which
we
talked
about
just
a
couple
minutes
ago,
a
wide
range
of
haywide
based
oscillators
and
the
arsenite
plus
iodine
reaction.
Those
are
three
things
that
sort
of
characterize
out
of
catalysis,
but
also
you
see
auto
catalysis
with
numerous
enzyme
systems
in
chain
branching
reactions
in
the
gas
phase.
D
So
these
are
all
systems
that
are
chemical
in
nature,
that
exhibit
autocatalysis
and
then
not
just
you
know,
produce
reactions,
but
they
produce
structure,
they
produce
pattern
formation
and
then
they
talk
about
the
brescillator
and
the
Oregon
oregonator,
which
is
from
the
University
of
Oregon.
The
Brussel
later
is
another
version
of
that
those
are
complicated,
chemicals
of
organization
schemes,
and
so
they
kind
of
get
into
that
they
talk
about.
So
the
equations
that
they
propose
here,
which
we
talked
about
last
week,
are
based
on
the
Stoichiometry
of
the
system.
D
So
you
can
look
at
the
Stoichiometry
and
look
at
how
things
are
added
to
the
system
and
produce
reactions.
So
this
is
kind
of
how
they
walk
through
this.
They
walk
through
the
mathematics.
Then
they
walk
through
the
differential
equations
and
then
these
equations
can
be
coupled
to
produce
this
morphogenetic
system.
D
So
there's
a
lot
of
math
in
this
paper
and
this
is
sort
of
the
original
math
of
the
model,
and
so
then
they
walk
through
some
results.
Here
then,
in
terms
of
autocatalysis,
so
you
see
cubic
Auto
catalysis,
so
the
Catalyst
definite
lifetime
non-zero,
one
flow
of
B,
so
they're,
looking
at
different
stoichiometries
they're,
looking
at
different
flows
of
energy
and
then
they're
producing
this
pattern
of
Auto
catalysis,
which
is
the
pattern
formation.
The
second
paper
is
this
generative
complexity
of
the
gray
scout
model.
D
This
is
by
Andrew
adamatzky
who's
done
some
really
interesting
things
in
the
artificial
life
community.
So
this
is
from
2018
it's
a
more
recent
paper
and
this
extends
the
gray
Scott
reaction,
diffusion
system
to
a
model
of
generative
complexity.
So
the
abstract
reads
and
the
grace
got
reaction.
Diffusion
system,
one
reactant-
has
constantly
fit
into
the
system.
Another
reactant
is
reproduced
by
consuming
the
supplied,
reactant
and
also
converted
to
an
inert
product.
D
The
rate
of
feeding
one
reactant
into
the
system
and
the
rate
of
removing
another
reactant
from
the
system
determine
configurations
of
a
concentration
profile
or
concentration
profiles,
and
those
are
Stripes
spots
and
waves.
So
you
basically
have
these.
These
reactions
there's
a
Stoichiometry
which
means
that
there's
a
certain
amount
of
that
reactant
and
then
given
those
parameters,
we
can
get
this
sort
of
profile,
which
is
almost
like
a
distribution
or
an
ensemble
of
things,
and
those
ensembles
are
stripe
spots
and
waves.
D
So
that's
something
you
think
about
in
terms
of
information,
Theory
and
thermodynamic
entropy,
and
so
you
know
we
can
use
the
mathematics
of
that
those
those
phenomena
to
characterize
some
of
these
patterns.
D
As
it
stands,
we
don't
have
a
really
good
method
for
characterizing
Collective
Behavior
other
than
sort
of
the
qualitative
method
of
that
looks
really
interesting.
So
you
know
to
be
able
to
take
what
we're
generating
from
these
models.
And
of
course,
we
talked
about
cellular
automata
last
week
as
well
being
able
to
implement
these
types
of
patterns
in
Solo
or
automata,
and
then
analyzing.
Those
sewing
order,
automata
in
terms
of
the
different
classes
of
complexity
they
produce
now
Stephen
wolfram's
work
on
a
new
kind
of
science.
D
Sort
of
goes
in
that
direction
where
he
classifies
different
patterns,
classifies
the
rules
and
gives
us
a
sort
of
a
good
sort
of
typology
of
complexity.
But
there
are
other,
you
know,
there
are
other
ways
we
can
analyze
this
too,
and
one
of
the
ways
is
to
sort
of
quantify
different
variations
in
some
of
these
outputs.
So
you
know
striping
patterns,
you
know
what
happens
when
you
change
a
set
of
parameters.
Upstream,
do
we
get
a
quantitative
difference
in
striping
or
in
spotting
or
in
wave
patterns?
And
what
are
those
parameters?
C
D
That's
an
aside
from
this
paper
we
calculate
the
generative
complexity,
a
morphological
complexity
of
concentration
profiles
grown
for
a
point-wise
distribution
of
the
medium
of
the
gray
Scott
system
for
a
range
of
the
feeding
and
removal
rates.
The
morphological
complexity
is
evalued.
It
is
in
Shannon
entropy,
since
in
diversity.
Approximation
of
lumpul,
zip
complexity
and
expressivity,
which
is
Shannon
entropy,
divided
by
space
filling
so
I,
didn't
realize
that
they
were
going
to
get
into
this
when
I
started.
D
The
abstract
I
just
mentioned
that
we
should
use
something
like
Shannon
entropy
or
you
know
the
mathematics
of
thermodynamic
entropy,
and
indeed
we
get
this
Shannon
and
Simpson
diversity
approach
or
Shannon,
entropy
and
Simpson
diversity
approach,
Shannon,
entropy,
being
the
information
content
of
something
the
H
value,
Simpson
diversity
being
very
similar,
but
also
having
like
a
be
allowing
you
to
look
at
the
diversity
or
the
unpredictability
of
things.
So
there's
a
there
are
different
measures
that
they're
using
here
to
characterize
some
of
this
morphological
complexity,
but
still
I,
think
my.
D
My
point
stands
about
like
being
able
to
characterize
some
of
these
outputs
quantitatively
and
classify
them
it's
kind
of
hard
to
do,
and
there
isn't
a
lot
of
work
on
this.
We
analyze
behavior
of
the
systems
at
the
highest
values
of
degenerative
morphological
complexity
and
show
that
the
gray
Scott
systems
expressing
the
highest
levels
of
the
complexity
are
composed
of
the
wave
fragments
similar
to
wave
fragments
in
some
some
excitable
media.
D
You
know
like
the
bz
reaction,
as
we
mentioned
earlier,
that
allow
you
to
generate
these
kind
of
things
and
these
kind
of
patterns,
and
so
there's
sub-excitable
media,
which
is
where
it's
not
as
excitable,
so
excitable
means
it
produces
outputs
that
are
chain
always
changing.
There
are
a
number
of
neural
models
for
this.
There
are
a
number
of
chemical
models
for
this,
so
you
get
these
kind
of
phenomena
you
get
some
excitable
media,
excitable
media,
and
then
you
get
these
different
patterns.
D
That
result,
then,
actually
he
brings
in
it
that
ties
in
this
idea
of
Conway's
Game
of
Life,
anticipative,
solitons
and
gliders,
which
we
talked
about
in
terms
of
The
Game
of
Life.
Last
time,
also
in
terms
of
the
gray
Scott
model
and
what
people
are
doing
with
soil
or
automata
and
then
with
the
the
sort
of
linear
models
that
we
talked
about
last
week.
D
So
that's
this
paper,
then
there's
this
game
of
life
in
the
epigenetic
principle.
This
is
in
Frontiers
and
cellular
and
infection
microbiology,
and
this
paper
uses
cellular
automata
to
explore
Conway's
Game
of
Life
and
epigenetic
principles
or
what
they
call
epigenetic
principles
so
we'll
go
through
the
abstract
here
once
again.
D
D
D
We
look
for
similarities
and
differences
between
two
epigenetic
models
by
Turing.
An
Edelman
so
Turing
proposed
an
epigenetic
model
as
in
addition,
it's
different
than
the
reaction.
Diffusion
model
and
Edelman
also
proposed
an
epigenetic
model,
and
so
these
models
are
realized
in
Game
of
Life
objects.
D
So
remember,
the
game
of
life
is
a
cellular
automaton
and
it
has.
You
know
these
different
patterns
that
form
as
a
cellular
automata
does,
but
it
usually
is
something
that's
moving.
So
you
have
these
things
like
gliders,
glider
guns,
different
patterns
of
activated
cells
that
often
move
they
either
move
stationary
in
a
stationary
manner
or
they
move
across
the
board.
D
The
cellular
automata
grid,
and
so
you
get
these
patterns
that
change
and
people
have
tried
to
classify
them
in
different
ways
and
they
result
from
the
zero-sum
game
or
the
zero
player
game,
and
so
the
Game
of
Life.
They
Implement
these
models
in
The
Game
of
Life.
They
show
the
value
of
computer
simulations
to
experiment
with
and
propose
generalizations
a
broader
scope
with
novel
testable
predictions.
D
We
use
the
game
to
explore
issues
in
simple
symbiopolysis
and
Evo
gevo,
which
is
the
evolution
of
development,
where
we
explore
fractal
hypothesis
that
self-similarity
exists
at
different
levels.
So
you
have
cells,
organisms
and
ecological
communities,
there's
this
complexity
of
each
and
that
complexity
itself
similar.
So
you
have
similar
patterns
of
cells,
similar
patterns
of
organisms
and
similar
patterns
of
ecological
communities.
D
So
you
get
these,
you
know
sets
of
interactions
that
occur
at
different
scales,
and
this
may
explain
why
you,
maybe
you
see
like
in
single
cell
communities
or
the
patterns
that
get
recapitulated
to
the
level
of
organisms,
complex
organisms,
interacting
and
then
ecological
communities
in
the
way
they're
organized
the
cell.
Similarity
is
the
result
of
homologous
interactions
of
two
is
processes
two
as
processes
modeled
in
The
Game
of
Life.
So
these
are
pairwise
processes,
modeled
in
on
the
soil
or
Autumn.
D
One
thing
they
mentioned
here
is
Conway
made
connections
with
Biology
part
of
his
purpose,
bringing
out
analogies
with
the
rise
fall
and
alter
alternations
of
a
society
of
living
organisms.
So
this
is
from
Gardner,
so
he
Conway
was
interested
in
biology
as
well
as
sort
of
this
gay
Mass
game,
theoretic
aspect,
and
so
this
purpose
explains
the
interest
that
this
came
from
biologists,
since
it
explicitly
aims
to
model
a
basic
process
in
biology.
The
evolution
of
ecological
communities,
but
you
can
also
apply
it
to
other
parts
of
biology.
D
You
can
apply
it
to
development,
to
early
life,
to
General
biological
complexity,
and
so
this
has
become
a
main
thing
in
looking
at
biological
chaos,
a
main
tool
by
which
we
can
look
at
these
things.
So
he
also
talks
about
some
of
these
distinct
emergent
forms,
so
there's
still
life
configurations
which
are
stable
over
many
iterations
and
he
shows
an
example
of
still
life
here,
where
these
are
things
that
form
and
that
remain
stable.
They
can
remain
in
their
fixed
form,
so
they
don't
really
move
around,
but
they
also
don't
disappear,
they're,
not
itinerant.
D
Then
there
are
these
blinkers
which
are
patterns
that
oscillate
with
a
fixed
period.
So
these
are
things
that
don't
are
also
stable,
but
they're,
not
stationary.
They
are,
you
know,
moving
around
I,
guess
in
a
sense
they're
turning
on
and
off,
and
so
it's
this
blinker,
where
it's
a
oscillation
that
can
be
measured.
It's
a
regular
pattern
over
time.
D
Oscillators
are
stable
over
a
cycle
returning
to
an
initial
State.
Such
Cycles
can
be
short
blinkers
or
very
long.
So
you
can
have
blinkers
or
longer
oscillations,
and
so
then
there's
this
other
set
of
things
called
movers
and
gliders
which
move
across
the
grid.
These
are
emergent
forms,
so
they
move
they're
itinerant
in
terms
of
their
presence.
You
can
see
them
and
then
they
disappear,
so
a
glider
might
Glide
across
the
cellular
or
automata
grid
and
disappear.
D
But
you
know
these
are
the
things
that
are
really
capture
the
attention
of
biologists,
since
they
look
biological
but
they're,
not
of
course
not
Quantified
as
biological.
So
you
know,
like
I,
said
from
the
last
paper:
we
have
these
quantitative
tools,
but
we
still
need
more
quantitative
tools,
I
think
more
rigor
in
terms
of
characterizing.
These
patterns
there's
a
lot
of
rigor
building
in
in
terms
of
describing
how
these
patterns
are
built,
but
not
a
lot
of
rigor
in
terms
of
analyzing
these
patterns
and
the
emergence
that
produces
them.
D
So
Conway
uses
a
lot
of
Concepts
from
ecology,
and
so
this
leads
to
their
paper
where
they
talk
about
these
ecological
relationships.
So
Turing
has
a
model
of
development
and
it's
related
to
his
modelomorphogenesis
reaction
to
Fusion
morphogenesis.
D
It
isn't
the
same
thing,
but
it
has
a
number
of
different
features,
so
his
concept
of
morphogens,
which
are
these
particles
or
molecules
that
he
proposes
your
theoretical
constructs.
So
they
can
be
anything
in
nature.
That
does
this
sort
of
thing,
but
they
basically
morphogens
are
foreign
producers
they're
these
fundamental
particles
of
his
reaction,
effusion
model,
you
have
gradients
of
morphogens.
D
D
There's
a
very
analogous
thing
where
you
have
this
common
basic
pattern,
which
is
maybe
defined
as
a
blinker,
and
it
can
be
understood
as
involved
in
catalysts,
so
the
blinker
begins
with
a
row
of
three
cells
following
the
rules
and
the
first
iteration
of
the
two
two
of
the
three
cells
dies
on
either
side
of
the
central
cell,
but
two
new
cells
grow
above
and
beyond
the
central
cell.
So
we
basically
have
this
template
for
interaction.
You
have
this
blinker,
where
parts
of
it
turn
off.
New
parts
are
added.
D
You
have
this
serious
kernel
of
a
blinker
that
has
you
know
the
things
the
cells
around.
It
have
a
frequency
and
it
basically
is
recognized
as
a
pattern,
but
it
you
know,
so
you
could
think
of
this
as
like
having
a
genetics
to
it,
where
it
turns
on
a
central
cell
and
that's
persistent
and
then
the
cells
around
it
are
variable
in
terms
of
their
pattern
and
that
pattern
is
driven
over
time.
So
the
interactions
over
time
give
you
this
frequency
of
the
neighboring
cells,
and
so
you
could
Analyze
That
as
a
common.
D
D
D
You
know
multiple
sets
of
interactions
all
at
one
time,
so
this
is
actually
kind
of
what
we
were
talking
about
with
some
of
these
patterns
in
The
Game
of
Life,
and
we
can
maybe
replicate
some
of
these.
It's
it's
hard
to
say
so.
One
of
the
things
I
want
to
point
out
about
trains
model
I
find
interesting
is
that
we
have
this
sort
of
X
and
Y.
D
So
we
have
X
look
at
Y
and
say
this
is
like
a
bunch
of
morphogens
in
a
concentration
gradient.
So
we
have
this
morphogen.
These
morphogen
patches
a
little
diffuse
in
either
direction,
so
it'll
start
to
overlap,
so
X
is
moving
in
this
direction.
Y
is
moving
in
this
direction.
This
is
the
region
of
overlap,
is
where
you
get
the
pattern
to
get
these
nice
striping
patterns
from
this
pairwise
interaction
of
X
and
Y.
So
X
is
here.
Y
is
here
x
is
here:
Y
is
here
they
diffuse
into
one
another.
D
They
form
these
regions
of
interaction,
these
regions
of
interaction.
Instead
of
being
great,
you
know,
gradated,
there's
a
there
should
be
a
gradation
like
basically
a
white
stripe
of
black
stripe
and
then
a
gray
interstitial
in
between,
but
that's
actually
not
what
you
see
usually
see
a
very
sharp
boundary
between
the
black
and
the
white.
Let's
say
that
x
equals
White
equals
black,
and
you
would
expect
from
those
two
to
see
a
gray
interstitial.
It
kind
of
grades
from
different
Shades
of
Gray,
but
you
don't
actually
see
that
you
see
a
sharp
boundary.
D
That
is,
you
know.
So
the
idea
is
that
this,
these
gradients
kind
of
intersect
and
interact
to
form
this
boundary
and
it
gets
reinforced
over
sets
of
interactions.
D
I'm
going
to
be
here,
n
would
be
here
and
we
move
out
from
there
and
we
get
these
concentric
circles
that
have
the
same
pattern
of
interactions.
So
as
we
had
before
with
the
overlap,
we
have
the
same
set
of
overlaps
and
you
could
draw
a
Venn
diagram
to
show
these
overlaps
and
the
sort
of
the
exclusivity
of
each.
So
you
know
X
and
M
might
overlap
x,
y
and
M
might
overlap.
There
might
be
a
regional
or
lab
that
we
can
Define
as
being
unique.
D
Now,
what's
interesting
about
this
system,
is
it's
not
just
the
pairwise
interaction?
It's
a
four
wise
interaction,
and
so
four
wise
interaction
is
going
to
behave
quite
differently
than
a
pairwise
interaction.
D
D
It
might
be
itinerant,
it
might
move
around
and
what
we
can
do
with
f4wise
relationship
is.
We
can
actually
put
it
on
a
cellular
automata,
but
we
can
model
it
a
little
bit
differently.
We
can
have
cells,
the
cell
here
and
its
neighboring
cells
in
a
neighborhood
and
then
from
that
neighborhood
we
can
basically
modify
the
rules,
the
interaction
rules
so
X
X
and
Y.
D
There's
an
interactional
here
between
this
cell
and
the
cell
X
and
M
there's
an
interaction
between
the
cell
and
the
cell,
and
you
notice
that
M
and
Y
don't
really
come
into
contact,
they're,
sort
of
catty
corner
from
one
another,
so
they're
distinct
but
they're
in
the
same
neighborhood
and
they
share
a
connection
with
x.
So
what
that
means
is
that
they're
part
of
the
same
pattern,
but
they
don't
necessarily
have
a
boundary
or
when
they
do
have
a
boundary.
It's
not
clear
what
the
rules
should
be
to
mediate
it.
D
So
there
are
all
these
interactions
where
in
time
where
things
get
turned
on
and
off
so
this
might
these
two
cells
might
get
turned
on
a
t
equals
two
to
four.
These
two
might
get
turned
on
at
T
equals
four
to
six,
and
this
might
recur
at
like
six
to
eight
and
so
on
and
so
forth.
I
didn't
get
eight
in
there,
but
the.
D
That
there's
a
lot
of
sort
of
complex
pattern
formation,
especially
over
time,
and
so
these
sorts
of
things
these
these
things
that
are
greater
than
pairwise
comparisons,
drive
a
lot
of
this.
We
can
also
model
this
as
a
set
of
patterns
in
a
cellular
or
Automotive.
D
So
one
last
thing
I
want
to
talk
about
with
this
paper
is
edelman's
work
on
top
of
biology,
you've
wrote
a
book
in
I
think
in
the
late
80s
on
top
of
biology
and
I,
don't
know
how
much
people
pay
attention
to
it
anymore,
but
basically
there's
this
Topo
aspect,
which
are
these
spatial
relationships
and
the
importance
of
place.
So
if
we
go
back
to
our
cellular
automata,
there
is
the
spatial
aspect.
It's
very
explicit.
We
have
a
spatial
aspect
in
this
model
too,
so
these
these
variables
are
not
just
floating
around
nowhere.
D
D
So
we
have
this
aspect
of
Topo,
which
is
what
Edelman
defines
is
the
importance
of
place
in
other
spatial
relationships,
including
the
Newtonian
and
mechanical
processes
and
visaged
by
Turing
Edelman
proposed
five
primary
processes
involved
in
development,
three
of
which
you
call
driving
processes
and
two
he
called
regulatory
processes,
so
these
correspond
Loosely
the
distinction
between
genetic
and
epigenetic
processes.
So
he
defines
epigenetics
genetics,
of
course,
being
DNA.
Epigenetics
are
modifications
of
that
DNA
or
the
products
of
that
DNA,
but
in
edelman's
theory
they
are
all
epigenetic.
D
So
the
driving
processes
are
milieu
dependents,
a
milieu
dependence
means
it's
dependent
on
the
environment
or
the
sort
of
the
environmental
State,
and
so
that's
that
that
is
sort
of
a
different
definition
of
epigenetic.
But
that's
how
I
don't
know
deals
with
it
if
you
reframe
these
processes
in
terms
of
The
Game
of
Life,
there's
cell
division,
cell
death
cell
movement
cell
adhesion
and
differentiation
in
induction
and
all
these
can
be
formulated
in
the
day
of
life.
D
They're
consistent
with
edelman's
model
they're,
also
consistent
with
turning's
model
and
with
this
game
of
life,
implemented
implementable
in
The
Game
of
Life,
so
I'm
gonna
I'm
not
going
to
talk
anymore
about
that
paper.
You
can
explore
it
on
your
own,
so
I'm
going
to
move
on
to
another
topic
that
we
talked
about
last
week
and
that
is
swarm
intelligence
and
networks
and
sort
of
these
swarm
Intelligence
on
networks.
D
So
I
mentioned
that
we
talk
about
cellular
automata,
we
have
the
cellular
automata
grids
and
partially
these
things
can
be
sort
of
discrete
versions
of
these
gradient
systems.
So
you
can
have
these
discrete
versions
of
the
gradient
systems
where
the
two
parameters
interact
and
their
gradients.
You
can
basically
have
rules
between
cells
that
manages
interactions
now,
let's
suppose
that
we
want
to
put
those
in
a
network.
So
instead
of
having
a
grid
like
this,
we
have
individual
cells.
D
We
have
a
network
which
is
discrete
in
again
discrete,
but
it
also
isn't
really.
It
can
be
spatially
distinct,
but
it
doesn't
have
to
be,
but
it
basically
has
these
nodes
that
interact
and
their
interaction
stream
determines
their
connectivity,
and
it's
a
different
way
of
describing
this,
so
we
were
automatic
except
that
the
cells
don't
live
or
die
either
connected
or
not
connected
city
of
different
topologies.
That
form
you
have
different
shapes
like
cyclical
complexes,
and
you
can
do
like
all
sorts
of
top
logical
analysis.
D
D
Now
we
talked
about
in
one
case
where
networks
might
have
this
sort
of
collective
behavior,
and
so
imagine
that
these
networks,
whatever
their
topology,
have
a
collective
Behavior
inherent
in
their
topology.
In
other
words,
on
their
surface,
you
have
like
nodes
that
will
behave
in
a
certain
way
together
and
in
some
cases
that's
what
a
neural
network
is
it's
basically,
if
you're
connected,
if
you're
highly
connected.
If
there's
a
strong
affinity
for
connection,
you
behave
together
as
opposed
to
a
part.
So
that's
where
we
get
into
the
Swarm
intelligence.
C
D
Thought
yeah
I
think
people
were
doing
this
in
in
like
communication
networks,
and
indeed
that's
where
the
literature
is
so
there
are
three
paper
well.
This
is
a
article
from
LinkedIn
believe
it
or
not.
This
is
on
Swarm
intelligence,
helping
with
network
optimization
and
Dynamics
and
dynamic
and
complex
environments,
and
this
is
more
of
like
about
engineering.
So
this
is
where
we
want
to
optimize
the
network
topology.
D
We
want
to
improve
performance,
efficiency
and
reliability,
so
this
is
mainly
for
communication
networks,
like
cell
phone
networks
or
other
types
of
you
know,
Transportation
Networks,
but
so
people
are
interested
in
this
and,
of
course,
swarm
intelligence
is
something
we
find
in
nature.
We
find
this
amongst
collectively,
collectively
behaving
organisms
such
as
fishes
bees,
birds
and
ants.
We
talked
about
this
last
week
as
well.
D
There's
this
Paradigm
called
swarm
intelligence,
which
is
a
programming
Paradigm
that
takes
inspiration
from
this
and
builds
like
these
swarms
that
can
be
evaluated
for
their
their
sort
of
efficiency
and
their
robustness
instability.
So,
like
you
know,
we
go
back
to
the
game
of
life
where
we
have
these
patterns
that
are
stable.
Some
that
are,
you,
know,
itinerant
or
fleeting,
and
we
can
interpret
reinterpret
this
in
the
language
of
swarm
intelligence
and
swarm.
D
Intelligence
on
Networks
facilities
are
collective
behaviors
that
operate
on
a
discrete
Network,
and
so
we
can
use
swarm
intelligence,
algorithms
to
sort
of
evaluate
not
only
the
efficiency
of
networks
but
other
things,
such
as
the
robustness
of
the
networks,
the
other
other
attributes
of
the
networks.
That
may
not
be
immediately
apparent
just
by
doing
using
a
regular,
optimization
algorithm.
D
So
they
talk
about
ACO
PSO,
which
is
particle
swarm,
optimization
bco,
which
is
B
Colony
optimization.
So
there
are
a
lot
of
different
methods
to
these
kind
of
algorithms
and
they're
all
inspired
by
natural
processes.
So
they
talk
a
little
bit
more
about
Network,
optimization
how
swarm
intelligence
can
improve
it.
Can
it
can
allow
for
people
to
build
networks
that
adjust
to
changing
Network
conditions,
demands
and
constraints,
enable
cooperation
between
human
operators
and
algorithms
result
in
better
coordination,
communication
and
collaboration?
D
So
that's
what
swarm
intelligence
does
in
a
lot
of
these
communication
networks?
We
can
apply
algorithmic
approaches
to
this.
We
can
also
use
swarm
intelligence
to
look
at
routing
and
communication
networks.
This
is
a
paper
on
this
concept,
so
you're
applying
a
swarm
intelligence
algorithm
to
improve
routing
in
these
networks,
and
this
is
in
a
communication
Network,
where
you
want
to
Route
messages
as
quickly
as
possible,
and
so
a
lot
of
these
networks
are
ad
hoc.
D
They
come
into
existence
and
you
want
to
improve
the
performance
on
these
networks,
so
this
is
a
perfect
sort
of
Paradigm.
For
this
you
have
these
itinerant
Network
topologies
that
you
want
to
optimize
for
on
the
Fly,
and
certainly
swarm
intelligence
gives
you
that,
so
this
is
again
from
the
communications.
Literature
there's
a
third
one
here:
the
Swarm
intelligence
for
Next
Generation
networks.
D
This
is
a
review
from
the
Journal
of
network
and
Computer
Applications,
and
they
talk
about
these
next
Generation
networks,
with
communication
devices
being
networked
and
using
swarm
intelligence
and
as
well
as
Game,
Theory
and
convex
optimization
to
deal
with
some
of
these
problems.
So
this
is
really
interesting
work.
This
is
the
literature
where
it
is
now
it
doesn't
mean
that
you
can't
apply
these
to
other
problems.
Other
sets
of
problems
and
even
back
into
computational
biology
and
biology.
D
So
this
is
something
that's
an
active
area
and
I
wanted
to
follow
up
on
a
lot
of
that
work
that
we
talked
about
last
week.
So
thank
you
for
paying
attention
and
I
hope.
You
learned
something.