►
From YouTube: DevoWorm (2023, #9): GSoC, Origins of the Embryos, Alternative Histories, Swarms and Microswimmers
Description
Semantic Segmentation vs. Instance Segmentation for embryo cell segmentation. Update on how to propose a GSoC project. Origins of the animal embryo: phylogeny and cell biology. Alternative histories of science. Swarms, microswimmers, and chemical gradient learning. Attendees: Richard Gordon, Sushmanth Reddy Mereddy, Kritika Verma, Shalini Vaggu, and Bradly Alicea
A
B
C
D
B
D
Because,
whatever
segmentation
we've
done
till
now
is
a
semantic
segmentation,
actually,
okay,
I
mean
this
really
means.
Like
every
cell
is
classified
into
one
class
I
mean
all
cells
will
be
like,
but
it
is
not
good
at
all,
because
semantic
segmentation
is
generally
happens
for
tissue
level.
When
the.
If
you
want
to
segment
tissues,
it
will
be
like
good
idea
to
implement
semantic
segmentation,
but
for
cell
level
we
need
to
implement
instant
segmentation
and
it
is
completely
related
to
GN
and
what
we
are
actually
applying
because
see
I
mean.
D
Can
you
see
here
Bradley
the
this
is
like
stage
one
of
demograph
yeah
implemented
by
jianglee,
and
you
can
see
here.
He
mentioned
the
model
works
well
for
images
where
cells
are
sparsely
populate,
but
not
for
cells.
Densely
populate
I
have
checked
the
actually
not
only
I
have
checked
up
till
32
cell
stage.
Development
is
working
pretty
good,
but
after
32
I
increase
the
tension,
I
mean
10
cents,
10
cents.
It
is
not
working
properly.
That's
the
main
issue.
D
I
mean
we
can
create
a
graph
neural
network
up
to
32
cell
stage,
but
at
the
end,
while
we
are
extracting
the
CSV
file,
is
there
right,
yeah
yeah,
whatever
in
stage
one?
We
are
extracting
a
CE
raw
data
CSV
file,
and
this
is
not
good
for
after
party
to
sell
stage
I
mean.
Can
you
see
the
cell
name
or
actually
not
predicted,
actually
correctly
and
the
time
and
the
points?
Also,
when
you
plot
on
a
3D
graph,
there
are
different
points
which
is
not
rejected.
D
This
is
the
segmented
image
not
original
image
there.
This
is
segmented
image
which
I
have
segmented
by
myself,
I
applied
for
it
see
I
mean
you.
Can
you
see
these
results
actually
see
here?
Centroids
were
not
extracted
properly
itself
or
not
even
segmented
properties,
but
here
every
cell
is
segmented.
Properly.
Yeah
and
I
have
gone
to
2019.
Gst
results.
Also
I.
Think
like
see
here
the
picture
of
three
cell
stage.
It
is
not
segmented
properly
and
cells,
I
mean
we
can't
extract
the
centroids
of
this
event.
D
Development
works
properly
than
2019
and
I
have
gone
through
2018
G-Shock,
and
this
is
like
some
lineage
population
and
got
what
it
is.
I
need
to
go
on
and
I
have
went
to
2017
segmentation
results.
This
is
2017
segmentation
result
see,
I
mean
here
also
the
self
were
not
segmented
properly.
We
can't
extract
and
2017
G-Shock
idea
mainly
concentrates
on
images
not
on
video
data.
B
D
A
D
Actually
you
mentioned
here
in
the
first
idea:
first,
work,
refining
means
to
segment
raw
data
and
incorporate
into
demograph
Pipeline,
and
you
mentioned,
will
require
working
refactoring
scene
and
models
and
understanding
biological
training
data
sets
so
see,
I
mean
we
can't
I
have
tried
to
improve
development
segmentation,
but
what
happening
is
worse.
Dolan
is
completely
built
on
a
machine
learning.
Algorithm
resonate
18
structure,
so
we
can't
segment
it
properly
using
development.
We
need
to
implement.
Do
you
know
unit
unit
structure?
It
is
mainly
used
for
foreign.
D
I
just
need
to
complete
work
with
that
extractive
CSV
file,
that's
I
mean
I,
can't
work
with
due
to
health
issues.
I
was
not
able
to
work
properly
yeah,
but
if
we
try
to
extract
centroids
and
another
thing,
also
in
demograph
stage,
two,
the
only
hasn't
used
the
CSV
file
which
he
was
getting
in
stage.
One
see
this
CSV
file
is
e,
Rod,
dot,
raw
data,
CSV
right
yeah
and
in
stage
two
as
an
input
file
he's
not
using
that
he's.
D
B
D
Yeah
I
mean
it
is
completely
related
to
graph
neural
network
only.
Actually,
we
are
completely
working
on
stage
one
of
demograph.
Clearly
we
are
incorporating
demograph
stage
one
pipeline
to
stage
two
actually
generally
worked
with
pretty
good.
At
this
one
stage.
Two
we
created
this
load
Embrace
using
k
m,
but
the
stage
one
is
little
messed
up
actually
yeah.
He
hasn't
given
the
CSV
file,
which
we
are
getting
from
the
stage
one
he
has
given
some
other
CSV
file.
When
you
see
raw
data
CSV-
and
here
the
input
was
different-
the
raw
data.
D
B
B
D
B
We
have
the
data,
sets
that
are
images,
and
you
know
so
I
mean
there
are
different
ways
that
you
can
train
this
see
we
use
work
on
C
elegans,
which
is
where
you
have
movies,
and
you
have
images,
but
we
also
have
things
that
are
like
other
types
of
data
that
we've
ex
like
extracted
and
the
cells
of
names,
so
there's
a
nomenclature
and
a
TR
and
a
lineage
tree.
So
we
have
all
these
things
that
we
can
train
the
model
with,
so
that
that's
where
I
think
some
of
this
stuff
comes
from.
A
D
B
D
D
Actually,
double
n
was
built
on
this
data
cell
tracking
challenge.
Data
set
and
I
was
using
the
same
data
set.
Actually
so
I
mean
we
can't
private
different
models.
We
will
use
that
data
and
extract
CSV
file,
whatever
we
want
and
we
can
that
incorporate
into
demograph
states
too
yeah
okay,
I
mean
this
idea
is
clear.
Right,
I
mean
this
is
related
to
GN.
Only
like
the
comment
you
have
mentioned,
yeah.
D
D
B
Mean
that's
why
I
like
to
see
the
proposal?
First,
like
you
know,
you
get
like
I,
usually
go
over
proposals
with
people,
and
so
anyone
who's
applying
to
gsaka
I
I
can
look
at
your
proposal
before
you
and
I
recommend
that
you,
let
me
look
at
it
before
you
turn
it
in,
because
it's
always
like
you
know,
it's
always
good
to
go
over
the
different.
You
know,
especially
the
schedule
and
the
description
and
everything
now
yeah
such
month.
B
I
think
you
did
a
pretty
good
job
of
describing
the
problem
and
in
general
and
then
the
the
schedule
I
think
has
laid
out
pretty
well
I'd
I'd
go
over
it
again
and
make
you
know,
make
sure
like
I
think
I
mentioned
before
you
know,
you
want
to
make
sure
that
you've
built
in
redundancies
for
if
something
doesn't
work.
If
you
can't
Implement
something,
what
are
you
going
to
do
for
that
week?
You
know
you
don't
want
to
sit
on
things.
D
Okay,
actually
I
implemented
this
thing
with
moles
in
real
sense,
I
haven't
worked
on
CL
again,
but
I
trained
the
model
on
most
embryo
cells
and
I,
just
segmented.
That
CL
against
data
I
mean
then
I
got
those
results.
I
mean
this.
Will
work
actually
I
mean
I.
Have
the
results?
I
have
I,
have
wrote
the
code
also,
but
I
need
to
extract
this
dot,
py
file
and
the
volume
and
sell
and
in
cell
names.
B
D
B
B
B
You
have
a
lot
of
cell
proliferation
and
you
know
it's
not
as
c
elegans
as
a
very
specific
way
that
it
divides
every
time
you
get
this
early
polarization.
You
get
other
things
that
are
going
on
with
Mouse
embryos
a
little
bit
different,
but
yeah.
It
would
be
yeah
I!
Think,
that's
good
that
you
can
pick
out
and
you're
making
that
distinction
between
instance
and
semantic
segmentation.
D
B
A
A
A
B
B
A
D
B
Yeah
I
mean
it's:
it's
not
it
I,
don't
know
how
much
work
people
have
actually
done
on
that.
So
it's
kind
of
one
of
those
things
where
it's
we
had
so
I
I
have
to
go,
dig
up
the
things
that
I
passed
around
last
year,
but
I
I,
don't
know
how
useful
it'll
be
so
you
know
so
yeah
so
yeah
I
I've
been
yeah
keeping
just
if
you're
interested
in,
let's
see
if
you're
interested
in
learning
more
about
what
we're
doing
in
the
different
projects.
B
This
is
the
one.
This
is
the
Devo
learn
repository
of
course.
Actually
it's
the
devil
and
organization,
and
we
have
the
diva
learner
repository,
which
is
was
actually
built
in
2020
in
2021,
and
this
is
something
you
know.
If
you
want
to
get
familiar
with
what
we're
doing
in
the
organization,
you
can
go
through
the
materials
here
we
have
Diva
learn.
We
have
devograph,
which
is
here
and
then
you
know
those
are
sort
of
there
there's
documentation.
B
There
are
also
a
couple
of
different
tool
boxes
that
social
Montes
pulled
out
and
we're
making
new
repositories
for
so
we
have
the
nuclear
segment
nucleus
segmentor,
we
have
the
membrane
segment
or
and
the
lineage
population-
and
these
are
all
based
on
the
sort
of
semantic
segmentation
techniques
that
we're
using
currently
in
evil
learn.
Although
you
know,
if
social
month
is
successful
this
summer,
we
might
get
some
instant
segmentation.
B
B
These
are
not
focused
on
evil
learned
specifically,
but
these
are
just
like
all
the
stuff
that
we've
done
since
the
beginning
of
the
project
or
the
beginning
of
the
group
when
the
group
started
back
in
2014.,
so
in
2014
we
started
out
and
we've
been
working
on
a
lot
of
projects.
Since
then,
we
have
a
lot
of
repositories
in
here.
B
B
There's
a
gsoc
2018
and
a
gsoc
2017,
and
they
just
go
back
ways.
You
have
to
kind
of
go
through
this
to
look
for
it,
but
yeah.
So
2018
is
here
and
2017
is
here,
and
so,
if
you
go
back
to
those
projects,
you'll
be
able
to
see
what
people
were
doing
back
then
we've
done
a
lot
of
different
things.
We've
done
things
with
trying
to
predict
lineage
trees,
we've
done
things
with
image
segmentation
and
we've
done
other
kinds
of
projects,
so
those
things
may
be
useful
to
you.
B
They
may
not
be
useful
to
you,
depending
on
what
you
want
to
do,
that
your
pro
your
proposal
is
about
so
and
anything
from
like
20
20
20
to
1
2022.
Those
are
all
at
the
in
the
Divo
learn
organization,
but
before
that
we
have
things
in
this
Repository
github.com
tivoworm,
so
that
I
want
to
disclose
the
history
of
that.
In
case,
you
find
that
useful.
B
B
I'm
going
to
go
on
to
some
other
things
here,
share
my
screen
again,
so
I
was
able
to
finally
to
get
something
together
for
what
they
call
Darwin
day,
which
is
on
February,
14th
and
I've,
been
meaning
to
post
a
blog
post.
It's
usually
something
you
do
with
Evolution
and
evolutionary
biology
and
I've
done
a
lot
of
different
posts
over
the
years
since,
like
I
started
this
blog
in
2008
or
actually
2009
and
I've
done
a
post,
I
think
since
2012
almost
every
year.
B
B
The
street
kind
of
shows
how
the
origin
of
Life
relates
to
the
origin
of
embryoes,
so
yeah,
this
last
Universal
common
ancestor
at
about
four
billion
years,
and
you
get
these
different.
Two
different
things
called
oxygenation
events,
which
are
where
the
Earth's
atmosphere
increased
its
oxygen
content.
So
you
have
the
great
oxygenation
event
and
then
the
second
oxygenation
event
and
the
second
oxygenation
event
seems
to
be
the
thing
that
really
kind
of
pushed
us
towards
an
embryo
of
some
type.
B
We
don't
really
have
embryos,
yet
we
go
through
this
second
oxygenation
event.
We
go
through
the
Advent
of
photosynthesis
and
we
end
up
with
this
sort
of
gray
area.
Where
we
don't
know,
we
haven't
seen
much
in
the
fossil
record,
but
we
do
have
these
pieces
of
evidence
like
from
the
dashanto
assemblage
in
China.
There
are
some
other
examples
of
multicellular
organisms,
maybe
a
bit
earlier
than
that,
and
so
all
these.
E
B
That's
the
thing
I
get
into
in
the
post
that
there's
a
there's
evidence
like
you
know
they
have
evidence,
they've
written
it
up
and
they
say
well.
This
looks
like
an
embryo,
but
actually
what
they're
looking
at
is
you
have
these
different
things
from
the
yeah
from
the
from
the
fossil
record
that
sort
of
have
these
properties
of
an
embryo,
but
they
don't
have
all
the
properties.
B
So
one
example
is
this
by
Salem
brassiere,
which
is
this
billion
year
old
protist.
So
this
is
like
going
back
about
a
billion
years,
and
these
aren't
really
embryos.
I
mean
they
kind
of
maybe
look
like
embryos,
but
what
they're
doing
is
they
have
these
different
stages
and
they
have
these
different
stages
of
differentiation.
E
B
Yeah
it
could
be,
and
so
this
is
the
kind
of
thing
they
have.
You
know
it's
not.
We
can't
really
say
it's
an
embryo
because
it
doesn't
exhibit
all
the
properties
of
a
modern
embryo,
but
it's
starting
to
get
like
from
this
split
between
bacteria
and
eukaryotes,
which
is
well
it's
back
here
somewhere,
but
the
last
unit
you
know
last
eukaryotic
common
ancestor
is
here.
It
probably
retains
a
lot
of
them
a
lot
of
bacterial
properties
back
in
in
this
time
period.
B
So
you
know
you're
getting
starting
to
get
things
that
look
like
in
a
unique
to
maybe
the
modern
type
of
embryo
that
we
see
now
and
then
there's
also
cavisphera,
which
has
developmental
stages.
But
you
know
these
aren't
really
the
developmental
stages
that
we
see
in
any
extant
embryo
necessarily
could.
B
A
B
All
right
there
we
go
yeah,
and
so
this
is
the
cavisphera.
This
shows
the
relationship,
the
we
have
the
last
metazone
ancestor
back
here,
and
we
have
you
know.
So
it's
it's
in
that
sort
of
phylogeny,
but
you
have
this
these
different
stages.
That
kind
of
look
maybe
like
it's
going
through
development,
but
it
doesn't
necessarily
look
like
a
modern
membrane.
B
Then
the
dashonto
embryos
are.
This
is
I
think
this
is
a
picture
of
one
of
the
deshanto
embryos.
So
I
mean
this
is
like
again.
You
don't
have
a
lot
of
the
things
that
you
associate
with
the
modern
embryo,
their
cell
migration,
their
unit
difference
is
you
have
these
cells
living
in
association
they're
in
this
sort
of
mass
environment?
So
you
have,
you
know,
basically
an
egg.
So
if
you
ask
the
question:
what
came
first,
the
chicken
or
the
egg?
Definitely
the
egg-
and
this
goes
back
to
this
time.
B
If
you
go
back
to
the
Cambrian,
which
is
about
550
million
years
ago
or
so-
and
this
is
the
tambourine
explosion
that
people
talk
about
when
they
talk
about
the
explosion
of
different
types
of
animals
that
exist,
you,
you
begin
to
see
sort
of
the
origins
of
eggs.
You
begin
to
see
the
origins
of
egg-laying
behaviors,
so
they
found,
for
example,
fossilized
Burrows,
and
this
is
about
70
to
80
million
years
before
the
Cambrian
explosion.
You
see
like
where
they're
they.
B
So
these
are,
these
are
pancreasstation
arthropod
species.
Basically,
so
these
are
not
really
I
mean
these
are
kind
of
in
within
the
Cambrian
explosion,
but
it's
kind
of
coming
out
of
that
period,
And
just
before
that
period.
So
actually
what
I
found
an
interesting
paper
on
this
by
the
end
of
the
Cambrian
explosion?
You
start
to
get
different
life
history,
trade-offs
related
to
clutch
size,
which
means
that
you
know
these
organisms
are.
B
If
you
know
you
can
see
differences
in
terms
of
the
number
of
eggs
that
they
weigh
and
their
their
length
of
time
that
they're
spending
like
before
they
lay
their
eggs
and
the
amount
of
time
they
spend
laying
eggs.
So
there's
an
interesting
sort
of
already
an
adaptation
related
to
egg
laying
there,
and
so
basically
the
evidence,
though
too,
is
where
you
get
these
well-preserved
multicellular
structures.
B
So
you
know
you're
starting
to
get
these
geometric
structures
that
are
multicellular
they're,
starting
to
form
these
little
colonies
and
they're
not
really
reminiscent
of
any
of
the
species
of
the
time.
So
they
can't
link
these
embryos
to
anyone's
species.
They
can't
say
that
this
is
an
embryo
of
species
X.
They
can
only
say
that
this
is
an
embryo
or
an
embryo
like
thing,
and
that
happens
a
lot
in
paleo
like
Imperial
botany,
for
example,
you
know
you'll
find
different.
What's
that
I
used.
B
E
B
B
C
E
B
B
B
Yeah
yeah
kind
of
interesting
I
would
definitely
be
interesting,
so
yeah
like
in
napilio
botany.
They
have
this
problem
where
they
have
different
parts
of
the
tree
or
different
parts
of
the
plant.
It's
like
the
root
system,
the
leaves
the
trunk
and
they're
all
different
species,
because
they
can't
say
for
sure
that
this
belongs
to
species
X.
So
that's
really
interesting,
but
you
know
we're
basically
trying
to
do
is
they're
trying
to
find
the
place
we're
in
the
tree
of
life.
B
We
get
the
first
embryos
and
they're,
you
know,
so
we
we
still
don't
know.
After
all
this.
This
is
a
lot
of
a
lot
of
images,
but
we
still
don't
know
exactly
where
it
happened.
We
know
that
there
are
a
couple
of
different
traits,
though,
that
might
Define
that
that
sort
of
arise
in
embryogenesis
or
that
maybe
related
to
embryogenesis
that
have
to
sort
of
you
know
things
that
happen
in
these
organisms
that
require
an
embryo
perhaps
and.
B
B
Yes,
there
are
a
couple
of
things
that
we
see
in
some
of
these
early
organisms
that
suggest
that
there's
this
need
for
a
an
embryo.
We
have
multicellularity,
we
have
cell
type
and
tissue
differentiation,
so
we're
starting
to
get
different
types
of
tissues
emerge.
We
get
axial
polarity
in
the
organisms.
We
get
immunity
systems
immune
systems.
B
The
Cambrian
explosion
basically
and
they're,
putting
this
in
this
in
this
phylogeny
to
show
where
these
organisms,
so
you
go
from
like
a
circular
organism
that
has
a
sort
of
radial
symmetry
and
then
your
this
is
where
you're
starting
to
go
down
the
bioletarian
into
the
bioletarian
clade.
So
a
lot
of
the
examples
from
the
literature
talk
about
bioletarian
embryos
now
they're
also
radial
and
early
radial,
symmetrical
embryos.
That
is
a
different
literature
that
I
don't
really
focus
on
in
this
post,
but
they
do
exist
and
that's
kind
of
it's
its
own
little
area.
B
B
Interestingly,
these
radiosymmetrical
embryos
these
early
radial
symmetrical
embryos.
They
go
down
a
different
evolutionary
path
and
they
tend
towards
these
nerve
Nets
instead
of
a
central
nervous
system.
So,
like
you
know
today's
Hydra,
if
you're
familiar
with
the
Marine
organism,
Hydra,
they
have
what
they
call
a
nerve
net,
which
is
where
the
nervous
system
is
distributed,
Across
the
Skin
and
that
that's
actually
a
direct
relationship
to
the
difference
between
the
triproblastic
and
the
diphoblastic
origins
of
tissues.
So
you
have
this
three
layered
embryo
versus
the
two
layered
embryo
and.
E
E
B
B
B
So
that's
that's
interesting
and
then
you
know
from
in
the
download
bilateral
bioitarian
lineage.
You
get
things
like
a
through
gut,
which
means
you
have
a
mouth
and
an
anus
and
a
gut,
and
then
you
get
limited
body
regionalization,
so
you
get
different
structures,
start
to
get
different
structures
like
you
know,
muscle,
pads
and
and
other
things
like
you
know,
a
heart
or
some.
You
know
other
organs
that
are
within
the
you
know.
B
Maybe
in
some
region
of
the
body,
that's
different
from
other
regions
and
then
you
get
and
you
also
get
segmentation.
So
you
get
these
different
in
invertebrates.
We
see
a
lot
of
segmentation
and
that
relates
to
Hox
genes,
and
they
talk
about
the
origins
of
fox
genes
is
one
of
the
things
that
they're
using
to
sort
of
interpret
some
of
these
embryos.
B
A
B
So
yeah
talk
about
the
fossil
evidence.
I
talk
about
some
of
the
things
with
going
on
with
some
of
the
molecular
information
and
again,
the
molecular
information
is
something
that
you
can
take
and
you
can
look
at
like
Divergence
times
between
different
genes.
So
you
can,
you
know,
look
at
the
genes
in
in
one
organism
versus
another,
and
you
can
do
a
you
know,
an
analysis
of
their
Divergence
and
you
can
pin
their
Divergence
time
back
to
a
certain
point.
B
You
can
look
at
like
you
know,
Gene
duplications
or
you
can
look
at
like
the
origin
of
a
gene,
and
you
can
you
can
time
that
as
well
through
mutations
and
other
types
of
measures,
and
then
you
get,
you
can
build
a
reconstruct
a
tree
like
this.
So
that's
what
they're
doing
with
the
molecular
data?
It's
just
from
Modern
organisms,
mostly
and
they're,
able
to
you,
know
infer
a
lot
of
these
things
back
in
time.
A
B
B
Have
oh
yeah?
Well,
that's
a
different!
That's
a
different
period
of
time
like
so
some
of
the
things
other
things
I
have
in
here
are
you
know
talking
about
like
the
molecular
evidence
and
then
talking
about
the
emergence
of
nervous
systems
and
brains
which
I
always
find
interesting,
because
there's
a
kind
of
an
it's
it's
similarly
hand
wavy
in
some
of
the
things,
but
we
know
that,
like
synaptic
proteins
have,
you
know,
evolved
very
early
in
sponges
and
you
know
they've
been
around
longer
than
brains.
B
So
it's
interesting
how
those
things
have
kind
of
gotten
put
in
place,
but
the
other
thing
is
that
you
have.
These
remains
of
you
know
once
the
eggs
once
you
get
further
down
the
line
in
evolution
and
you
get
towards
the
Cretaceous
Period,
you
start
to
find
complex
organisms
in
eggs,
and
this
is
an
example
of
an
embryo
or
like
a
almost
hatched
dinosaur
from
I.
Think
the
Cretaceous-
and
this
is
showing
the
the
egg
of
a
dinosaur.
B
A
dinosaur
is
in
the
egg
and
it's
ready
to
hatch,
and
you
can
see
that
that
it's
kind
of
folded
up
you
know
the
legs
are
folded
out
the
arms,
and
so
they
were
looking
at
the
different
postures
of
this
dinosaur
and
how
you
know
how
the
body
folds
up
in
the
inside
the
egg-
and
this
is
kind
of
the
first
example
of
really
complex
organism
inside
an
egg
I
know.
It
seems
like
it's
a
kind
of
late
in
Evolution
for
that
that
demonstration
to
be
there,
but
that's
and
it's
a
fossil
record.
B
E
B
Yeah,
well,
they
have
like
these
things,
get
preserved
in
in
different
ways.
So
there's
a
lot
of
preservation,
sometimes
if
there's
a
a
volcano
or
if
there's
like
some
sort
of
lava
flow
or
something
you
know,
you'll
get
like
conditions
where
things
can
be
preserved
quickly,
where
sometimes
they
get
preserved
in
you
know
different
conditions
where
the
there's
no
Oxygen,
so
it
gets.
You
know
it's
just
the
right
set
of
conditions
for
it
to
be
preserved
and
not
you
know,
doesn't
disintegrate.
So
sometimes
you
get
those
kind
of
conditions.
B
A
lot
of
the
ancient
DNA
is
actually
from
long
bones
and
other
types
of
Bones.
So
there's
a
bone
matrix
that
you
know
gets
preserved
if
it
gets
preserved
quickly.
Enough
you'll
have
you
know
tissue
inside,
that's
where
you
can
get
DNA
out
of,
and
there
are
other
biochemicals
where
that's
also
the
case
where
you
can
get
things
from
fossilized
remains
and
fossilization
is
a
complex
process.
B
So
you
get
like
you
know.
The
bone
typically
gets
replaced
with
the
mineral
other
minerals.
So
you
know
you
get
the
shape
of
it
and
you
know
that
there
are
things
that
get
preserved
and
there
are
things
that
don't
so
yeah
it's
it's
a
difficult.
It's
it's!
You
know
it's
not
very
common
to
find
things
that
are
mostly
soft
tissue
in
the
fossil
record,
but
you
can
find
them
so
that's
kind
of
an.
B
So
one
last
thing
I
wanted
to
say
about
this-
was
that
you
have.
We
also,
you
know
so
they're
the
bilateral
embryos
or
the
early
bilateral
embryos,
the
early
radial
embryos
which
are
the
symmetry.
So
you
have
something
where
you
have
a
left
side
and
a
right
side
for
bilateral
embryos,
the
symmetrical
embryos
you
have
this
sort
of
circular
symmetry.
So
you
have
like
you
know
it's
almost
like
a
pinwheel
where
you
have
different
symmetrical
parts
and
then
you
have
plant
embryos
and
plant
embryos.
B
I,
don't
see
plants
on
here,
but
it
would
be
off
to
this
side.
Plant
embryos
are
actually
well
I,
don't
know
if
they're
any
good
fossil
plant
fossil
embryos,
but
oh
plants
have
embryos
just
like
animals,
except
that
they're
a
little
bit
different
they're
seeds
they
have,
but
they
have
a
development.
They
have
cells
that
you
know,
there's
an
emergence
from.
You
know,
cells
and
everything
like
that,
and
you
know
we
don't
really
know
much
about
that.
So
you
know
the
there's
a
one
of
the
papers.
I
read.
B
E
E
B
B
That's
good,
thank
you
yeah,
so
that
was
yeah.
That
was
a
nice
discussion
and
so
yeah
that's
available.
I
put
it
in
the
chat.
You
can
read
it
over
and
you
know
talk
about
this
more
later.
This
is
actually
an
image
of
if
you're
interested
in
in
stable
diffusion
and
in
generative
machine
learning
models.
I
created
this
image
from
stable,
diffusion,
I
typed
in
Darwin
as
an
embryo.
It
was
a
prompt,
and
it
gave
me
this.
E
E
A
B
Then
I
found
this
other
article
in
the
Nautilus.
This
is
by
Philip
bali's,
a
science
writer,
and
he
said,
if
not
Darwin
who-
and
this
is
an
alternate
history,
if
you're
interested
in
alternative
histories,
where
you
know
you
suppose
that
history
unfolded
in
a
different
way,
then
this
is
probably
for
you.
So
this
is
an
alternative
history
of
the
great
ideas
of
science,
so
he
mentions
Darwin,
but
he
talks
about
a
couple
of
different
scientists
here.
Basically,
the
idea
is
if
they
hadn't
existed
like
if
Einstein
had
never
existed.
B
E
A
B
Yeah,
so
that's
so
this
just
kind
of
goes
through
this,
so
I
mean
it's
really
speculative.
But
basically
you
know
it's
just
kind
of
an
interesting
exercise
and
thinking
about
this
I
said
you
know
these
material.
These
theories
and
and
discoveries
probably
would
have
materialized
anyway,
sooner
or
later
so
you
know
it's
it's
nice
to
Revere
people,
individual
figures,
but
sometimes
these
things
will
come
about
one
way
or
another,
and
so
it's
really
about
like
us,
Reading
literature
and
figuring
things
out
and
you
know
someone's
gonna.
B
Do
it,
and
sometimes
you
know
it's
just
the
person
is
in
the
right
place
at
the
right
time.
Sometimes
that
person
really
has
skills
that
make
that
advanced
or
you
know,
like
I,
think
Darwin
going
to
the
Galapagos
was.
You
know,
obviously
key
to
getting
down
the
theory,
because
people
have
been
talking
about
a
lot
of
ideas
and
evolution
and
how
that
happened.
But
you
know,
and
then
you
go
to
a
place
where
you
have
a
lot
of
tropical
species.
B
You
have
a
lot
of
what
they
call
adaptive
radiation,
meaning
you
have
these
different
islands
where
different
Phoenix,
you
know
different
species
occur
and
have
diverged
based
on
their.
You
know,
isolation.
Those
sorts
of
things
were
necessary,
I
think,
but
still
people
would
have
done
this
at
some
point.
So
this
isn't,
like
you
know,
didn't
necessarily
require
Darwin.
B
Yeah
so
I
mean
you
know,
let's
see
it's,
it's
definitely
a
nice
little
exercise
and
then
so
you
know
this
talks
about
the
great
man
view
of
History,
which
means
that
you've,
you
know
you
have
a
you
know,
hagiography
of
one
person
versus
like
this
process,
view
of
History.
Where
you
look
at
the
ideas-
and
you
say
you
know
those
ideas
were
driven
forward
and
you
know
there
may
be
different
ideas
that
emerge,
but
that
yeah,
that's
basically
the
difference
here.
B
B
So
you
know
there
are
differences
in
caperna,
copernican
theory
that
are
mathematical
and
you
know
have
its
own
flavor,
but
these
ideas
were
around
before
they
were
just
you
know,
put
put
forward
in
this
way,
not
by
this
person,
so
talks
about
the
laws
of
motion,
hugans
talks
about
James,
Clark,
Maxwell
and
special
relativity,
which
is
not
general
relativity.
It's
a
different.
You
know
different
way,
a
different
Theory.
B
But
again
you
know
this
is
stuff
that
was
around.
They
were
different.
Like
things
there
were
different
loose
ends
that
had
to
be
resolved.
So
in
this
case
you
know
you
had
an
attempt.
There
was
already
an
attempt
to
square
electromagnetic
Theory
with
relative
motion
before
Einstein
and
thought
that
the
Dutch
physicist
Lowrance
the
two
invoke
contraction
of
space
and
time
after
a
fashion
that
did
so
with
a
view
to
preserving
The
Ether,
which
was
it
a
concept
they
used
before
Einstein's
base
time.
B
So
they
had
different
concepts
and
some
of
the
concepts
were
wrong,
but
they
were
basically
very
similar
to
other
Concepts
that
become
you
know
that
are
actually
right,
maybe
or
more
predictive.
We
should
say
so.
The
ether
is
like
space-time,
except
The.
Ether
didn't
predict
things
very
well
and
space-time
dead,
so
we
discarded,
The,
Ether
and
Einstein
came
up
with
general
relativity
and
space-time
became
the
predominant
idea.
So
this
is
yeah.
B
So
what's
a
swarmulator,
they
have
straight
reads.
We
study
that
emerging
behaviors
of
a
population
of
swarming
coupled
oscillators
sub
swarmulators,
so
these
are
coupled
oscillators,
which
are
generally
two
agents
that
are
synchronized
or
have
some
sort
of
oscillatory
behavior,
and
so
we
have
a
population
of
these.
B
If
you're
familiar
with
swarms,
you
know
that
you
could
have
a
bunch
of
insects
or
a
bunch
of
other
types
of
organisms
or
agents
that
swarm
around
a
common
resource
or
form
a
common
swarm,
and
it's
just
a
body
of
collective
behavior
and
so
we'll
see
examples
of
this
later
previous
work
considered
the
simplest
idealized
case
identical
storm
relators
with
global
coupling.
So
this
is
where
everything
is
homogeneous
and
the
coupling
is
global.
So
that
means
there's
no
variation
within
this
one.
B
Here
we
expand
this
work
by
adding
more
realistic
features
such
as
local
coupling
non-identical
natural
frequencies
and
chirality,
so
the
frequencies
are
shifted
so
that
there's
heterogeneity
in
the
Swarm
the
coupling
is
local.
So
that
means
that
there's
heterogeneity
in
the
Swarm
and
corality
simply
means
that
the
shape
has
a
sort
of
handedness.
You
may
think
of
chorality
as
like
the
orientation
of
DNA
or
of
proteins
where
they're
right-handed.
So
that
means
that
they
turn
to
the
right
when
they
move,
and
so
this
is
something
that
you'll
see
in
swarms
as
well.
B
This
more
realistic
model
generates
a
variety
of
new
behaviors,
including
lattices
of
vortices
meeting
clusters
and
interacting
phase
waves.
So
these
are
dynamics
that
you'll
usually
see
like
in
cardiac
muscle
or
you'll,
see
this
in
Squirrel,
what
they
call
scroll
waves
or
other
things
and
we'll
see
examples
again
later,
similar
behaviors
are
found
across
natural
and
artificial
micro
scale,
Collective
systems,
including
social
slime,
molds
spermatism
of
Vortex
arrays
and
quink
rulers,
I'm,
not
sure
where
clink
rollers
are
actually.
But
these
are
just
some
examples.
B
Our
results
indicate
a
wide
range
of
future
use
cases,
both
the
egg,
characterization
and
understanding
of
natural
swarms
into
design,
complex
interactions
and
Collective
systems
from
soft
and
active
matter
to
micro,
robotics.
So,
there's
a
lot
of
potential
use
for
this
in
bio-inspired
robotics,
but
also
in
soft
and
active
matter,
and
then
understanding
biological
swarms.
B
So
we
have
synchronization,
which
they
Define
a
self-organization
in
time
and
swarming
self-organization
in
space.
So
they
have
this
self-organization.
You
know
variable
both
in
time
and
in
space,
and
this
is
how
they
Define
swarms.
So
swarms
can
be
synchronized.
Swarms
are
actually
the
spatial
component.
Synchronization
is
the
temporal
component.
B
So
you
see
this
sort
of
synchronization
aspect
occurring
in
flashing
fireflies
firing,
heart
cells,
liking,
neurons
and
chorusing
frogs,
and
it's
basically
where
they
synchronize
their
signals.
Like
horsing
frogs
you'll
hear
like
a
DNA
frogs
Frog
song
as
opposed
to
a
bunch
of
independent
frog
songs.
All
overlapping
you'll
also
see
fireflies
synchronize
their
flashes
on
response
to
different
opportunities
in
the
environment.
B
Swarming,
on
the
other
hand,
is
seen
in
cell
collectives
that
migrate
in
response
to
external
signals,
flocks
of
birds
that
seamlessly
change.
Collective
flight
Direction
school
is
a
fish
that
coalesce
and
move
together
to
the
collective
Advantage.
So
these
are
all
things
that
are
spatially
organized
and
we
can
understand
them
as
spatial
phenomena.
B
B
Synchronization
research
has
been
broadly
speaking,
focused
on
oscillators,
which
may
synchronize
in
time,
but
now
move
around
in
space.
On
the
other
hand,
swarming
research
has
done
the
reverse.
It
is
studied
units
moving
through
space,
that
synchronize
spatially
dependent
variables
like
orientation,
but
not
internal
phase
variables
in
time
as
such,
the
interplay
of
synchronization
and
swarming
defines
a
new
kind
of
collective
Dynamics,
about
which
little
is
known.
B
So
this
is
kind
of
the
idea
of
the
swarmulator.
It's
combining
these
two
things
and
putting
together
a
mathematical
model
to
you
know,
get
us
further
down
the
road
of
understanding
these.
So
small
leaders
have
a
spatial
position,
X
of
I
and
an
internal
phase
which
evolve
according
to
equations
one
and
twos.
B
So
this
is
an
example
here
figure
one.
This
is
the
space
phase
order
parameter.
So
these
are.
They
have
five
emerging
behaviors
that
they
used
in
the
original
swarmulator
study
which
we'll
talk
about,
but
they
can
replicate
two
of
those
five
behaviors
in
this
example.
So
you
have
a
is
the
aesthetic
synchronization
where
the
parameter
s
is
near
one
or
near
zero,
and
then
the
static
phase
wave,
which.
B
Is
where
s
is
near
one,
so
you
can
see
that
between
zero
and
one
there's
the
synchronization,
you
have
variability
here.
You
have
this
hole
that
forms,
and
so
you
have
this
difference
in
synchronization
and
shows
a
difference
in
the
Swarm
and
so
there's
ways
that
you
can
analyze
this.
So
one
of
the
things
about
a
lot
of
collective
Behavior
research
is
people.
There
hasn't
been
a
lot
of
work
done
on
sort
of
analyzing.
These
shapes
that
form
from
Collective
Behavior.
B
So
they
talk
about
non-chiral
swarmolators,
they
talk
about
other
types
of
behaviors,
which
leads
us
to
figure
two,
which
is
where
non-chiral
swarmulator
is
no
natural
frequency
spread.
This
is
showing
some
of
these
phase
a
phase
diagram
of
different
phases
of
swarmulator.
So,
for
example,
you
have
this
a
example
where
you
have
disorder.
B
B
Then
you
have
synchronization,
which
is
here
where
everything
is
synchronized,
and
then
you
have
synchronized
expansion
where
the
Swarm
expands,
but
it's
in
the
same
sort
of
form
as
this
synchronized
example-
and
this
is
the
S
parameter,
so
the
s-parameter
is
high-
is
for
the
phase
wave
and
lowest
for
the
disorder.
Actually,
all
of
the
other
examples.
B
They
have
two
different
states
that
are
anti-phase,
so
you
have
one
set
of
oscillations
peaking
at
one
time
in
another
set
of
oscillations
being
in
another
time
and
that
results
in
this
sort
of
bipolar
segregation
within
the
Swarm
then
finally
get
radio
oscillation,
which
is
where
you
get
oscillations
that
are
radial.
They
go
around
in
the
circle
and
they
take
different
forms.
So
this
is
again
the
S
parameter,
it's
highest
for
phase
waves
and
bouncing
clusters
and
minimized
for
disorder,
anti-phase
and
Radial
oscillation.
B
B
Figure
three
is:
are
the
order
parameters
for
non-chiral
swarm
relators?
This
is
for
Disorder,
so
you
have
active
phase
waves,
splintered
phase
waves
and
synchronization
at
the
very
top
for
this
K
parameter
radial.
Oscillation
show
different
pattern.
Disorder
actually
shows
a
different
pattern,
for
this
is
for
the
J
parameter.
This
is
for
the
K
parameter.
J
equals
one
k
equals
one.
This
is
J
equals
one
for
disorder
and
then
J
equals
negative
one
for
disorder.
You
can
see
that
as
you
change
these
parameters,
they
change
as
well.
B
The
the
results
change
for
these
different
things,
all
right,
that's
a
little
bit
harder
to
unpack
if
you've
ever
read
the
paper,
but
this
is
actually
a
little
bit
better
example
of
what's
going
on
here
too,
you
have
again
these
phase
diagrams
and
the
S
parameter,
so
the
colored,
the
colored
gradients
or
the
S
parameter
changes,
and
then
these
are
the
Swarm
examples
of
the
Swarms
here.
So
you
can
see
that
you
have
all
sorts
of
different
behavior
for
swarms.
B
You
get
these
vortices,
which
are
where
the
the
agents
circle
around
a
central
point.
You
get
dense
revolving
clusters
where
they
revolve
around
a
very
dense
packings.
You
get
sparse
revolving
clusters,
so
they
revolve
around
a
central
point
but
they're
much
less
densely
packed,
and
then
you
get
vortices,
which
are
where
you
get
these
circular
movements.
So
these
are
very
similar
things,
but
they
all
they
exhibit
different.
They
they
inhabit
different
parts
of
the
face
space
so
for
the
S
parameter.
B
It's
lower
for
these
kind
of
clusters
here
for
these
kind
of
vortices,
for
these
kind
of
vortices
BS
parameters
higher
for
dense
revolving
clusters,
some
of
them
have
a
highest
parameter
value
and
then
for
some
dense
revolving
clusters.
Transitioning
sparse
revolving
cluster
thes
parameters
lower.
So
you
can
see
that
there's
it's
very
the
the
time
Evolution
and
the
spatial
level.
The
time
evolution
is
connected
to
the
spatial
Evolution,
and
it
has
a
lot
of
interesting
Dynamics.
B
In
this
case
you
even
have
a
an
even
more
complicated
phase
portrait.
You
have
these
double
revolving
clusters,
where
you
get
two
centers
of
mass
and
you
get
clusters
revolving
around
those
points.
You
gets
sparse
oscillating
clusters
where
things
oscillate
around,
but
they're
much.
The
population
is
much
disperser
and
then
you
get
these
vortices
again
for
high
values
of
PS
parameter
and
then
for
low
values
of
the
S
parameter.
Those
data
aren't
shown,
but
they
would
be
a
little
bit
different.
B
Probably,
but
the
point
being
is
that
there's
a
lot
of
variation
in
the
phase
portrait
and
so
they've
been
looking
at
the
global
level.
So
far,
they've
been
looking
at
the
order.
Parameters
and
changes
in
the
order
parameters
exhibit
changes
globally
in
this
structures,
and
so
we
can
see
again
that
you
can
change
the
order
parameters
and
you
can
find
this
is
actually
figure
six.
B
These
are
frequency
coupled
Clairol
swarmolators,
so
these
are
again
phase
portraits
where
you
get
these
different
types
of
behaviors
and
depending
on
the
S
value,
you
can
classify
different
parts
of
this
phase.
Actually,
it's
the
S
parameter
and
then
the
J
parameter
and
the
K
parameter.
So
this
is
K
versus
J
and
that
that
tells
you
your
sort
of
position
and
phase
space
and
then
the
S
parameter
is
the
color.
So
again
we
have
these.
B
You
know
different
types
of
patterns,
single
phase,
wave,
double
phase
wave,
the
double
phase
wave
is
where
you
have
two
centers
of
mass.
The
single
phase
wave
is
where
you
have
a
single
Center
of
mass
and
where
they
oscillate
around
that
Center,
and
then
you
have
these
different
oscillating
clusters.
So
you
have
these
dense
clusters
and
these
sparse
clusters
and
then
for
this
one
here
you
have
it's
even
more
complicated.
You
have
these
double
vortices
that
emerge
at
high
levels
of
J
for
moderate
levels
of
K.
B
You
get
these
double
clusters
for
high
levels
of
K
and
most
levels
of
J
and
again
most
of
these
examples
are,
with
a
relatively
low
to
moderate
s,
parameter
value
same
for
double
Vortex
single
vortices,
however,
have
a
much
higher
s
parameter
value
and
that's
for
low
values
of
J
and
moderate
values
of
K.
So
you
get
this
kind
of
single
Vortex,
and
then
you
get
the
this
type
of
disorder,
which
is
just
nothing.
B
That's
makes
any
coherent
sense
and
that's
at
very
low
values
of
J
and
K
and
very
low
values
of
the
S
parameter.
So
you
can
see
all
that
variation
over
time,
and
this
is
an
interesting
way
to
analyze.
Swarms
again
like
there
are
no
really
good
methods
for
doing
this.
This
is
probably
the
best
method,
I've
seen
or
one
of
the
better
methods
I've
seen
for
doing
this
mapping
this
to
a
phase
space
and
then
interpreting
the
Clusters.
B
You
know
different
patterns
that
get
formed,
but
they're,
probably
even
better
ways
to
do
that,
to
do
it
than
that,
but
I
have
not
seen
good
papers
on
this.
So
that's
that
paper,
and
actually
one
more
thing
I
will
show
here-
is
that
you
have
these
correlation
functions.
So
you
have
these
phase
phase
correlation
functions
where
you
show
different
patterns
of
the
Swarm
and
these
different
values
here
in
these
graph.
So
if
you
want
to
read
this
paper,
I
would
go
ahead
and
dive
into
it.
B
B
B
Similarly,
if
you
have
highly
packed
like
a
capillary
filled
with
a
flow
and
you're
going
with
the
flow-
and
there
are
a
lot
of
you-
there
was-
the
Reynolds
number
will
be
higher.
If
there's
there
are
fewer
of
you,
the
Winner's
number
will
be
lower
if
you're
in
a
turbulent
flow,
as
opposed
to
a
laminar
flow.
That
Reynolds
number
will
also
be
higher.
So
it
depends
on
a
number
of
things.
It
depends
on
the
turbulence
of
the
flow.
It
depends
on
the
shape
of
the
body
and
depends
on
the
density.
B
So
these
are
all
things
that
we
have
to
think
about
when
we
think
about
flows
and
microswimmers,
and
so
the
Jeffrey
equations,
a
simple
set
of
ordinary
differential
equations,
provide
a
useful
building
block
and
modeling
analyzing
and
understanding
these
Dynamics
in
a
flow
current,
and
so
particularly
when
incorporating
the
impact
of
the
swimmer's
shape.
Since
the
equations
contain
a
shape
parameter,
it's
a
single
scalar.
B
The
particle
orientation
forms
a
closed
orbit
when
situated
in
a
simple
shear,
and
this
non-uniform
periodic
rotational
motion
referred
to
as
Jeffrey's
orbits
is
a
constant
is
due
to
a
constant
emotion
in
the
nonlinear
equation.
So
it's
a
non-uno
Jeffries
orbit
as
a
non-uniform
periodic
rotational
motion.
B
So
if
we
think
about
something
like
an
organism,
an
organism
has,
you
know,
maybe
a
different
shape.
Maybe
the
shape
changes
over
time,
maybe
over
development
you
go
from
a
spherical
shape
or
an
oblong
shape
to
something
that's
asymmetric,
and
if
you,
even
if
you
don't,
even
if
you
remain
like
a
single
cell
organism,
you
can
still
experience
deformations.
Sometimes
those
are
due
to
stresses
and
strains
of
the
environment.
Sometimes
those
are
due
to
the
membrane,
the
rigidity
of
the
membrane.
B
B
So
I
go
through
some
of
the
micro
swimmer
hydrodynamic
governing
equations.
We
use
the
navier
Stokes
equations
inspiration
here.
Then
we
have
the
equations
of
motion
and
a
swimmer
immersed.
In
a
fluid,
we
have
the
motion
of
a
deforming
object.
We
introduced
two
reference
frames:
a
laboratory,
fixed
frame
and
the
swimmer
fixed
frame
so
you're
having
you
have
the
participant
and
the
Observer,
the
linear,
angular
momentum
conservation
should
hold
so
at
low
Reynolds
numbers.
All
inertia
terms
are
negligible,
which
is
always
usually
the
case,
but
a
higher
revelance
numbers.
B
You
have
to
account
for
inertial
motion,
and
so
let
me
go
through
this.
We
go
through
the
flow
that
satisfies
The,
Stokes
equations.
The
total
fluid
velocity
is
the
sum
of
the
background
flow
and
the
distributed
flow
is
the
presence
of
swimmer
use.
Sub
D
that
is,
U
equals
this
many
situations
in
nature
in
the
laboratory.
B
Where
you
get
too
small
to
really
be
governed
by
the
Reynolds
number,
there
isn't
enough
term.
You
know
the
effects
of
turbia
ones
aren't
as
great
the
effects
of
the
swimmer's
body
isn't
as
great
so
those
those
equations
break
down
around
that
size
limit,
so
we're
above
that
size
limit,
but
we're
still
pretty
small,
so
we're
smaller
than
you
know,
most
multi-cell
organisms.
I
guess
maybe
some
multi
support
ministers
are
that
small.
B
But
you
you
get
my
point
so,
even
in
tribute
in
the
turbulent
ocean,
the
smallest
length
scale,
the
vortices
are
estimated
at
around
10
to
30
millimeters,
which
is
sufficiently
large
to
approximate
the
local
background
flow
to
be
linear
in
position.
So
we
see
that,
like
you
know,
sometimes
the
length
scale
is
smaller
than
some
of
these
vortices
that
we're
dealing
with.
B
In
fact,
so
we
actually
go
through
these
equations
again
we,
you
know,
we
talked
we
introduce
a
lagrangian,
then
we
focus
on
the
kinematic
problem,
one
which
solves
the
swimming
velocity,
U
H,
and
then
the
swimming
trajectory
under
a
given
shape
gate.
So
that
means
how
the
shape
moves
through
the
environment.
Now,
if
we
need
to
solve
the
shape
gate
as
an
unknown
function
by
solving
a
coupling
problem
with
the
material
elasticity,
this
is
called
a
lasterohydrodynamics,
and
so
we
can
combine
different
things.
B
We
can
combine
hydrodynamics
with
the
last
material
elasticity,
as
I
said
before,
we
have
to
account
for
deformation
in
the
shape.
So
we
can
see
that
this
is
the
swimmer.
The
swimmer
might
deform
significantly
in
the
flow
or
it
might
not.
We
have
to
account
for
those
with
our
equations,
and
we
can,
of
course,
can
do
this
in
our
modeling.
That's
just
a
matter
of
what
we
want
to
assume.
B
So
then
we
have
a
lot
of
other
cool
math
or
swing,
there's
a
swimming
formula,
and
this
is
an
example
of
a
slightly
deforming
microswimmer.
So
this
is
a
single
cell
organism,
I
believe
I,
don't
know
what
it's
supposed
to
be.
It's
just
a
generic
organism
with
cilia,
but
you
have
these
deformation
points
on
the
surface
where
things
get
deformed
and,
of
course
that
affects
the
Reynolds
number
and
affects
how
it
interacts
with
the
flow
and
all
this.
B
B
Now
now
they're
talking
about
a
swarmer
referring
to
a
mathematical
model
of
a
rigid
body
with
a
surface
slip
velocity
instead
of
an
active
affirmation.
So
we
can
actually
approximate
the
deformation
by
using
surface
slip
velocity
due
to
the
model
Simplicity.
The
squirmware
has
been
extensively
studied
in
the
past
few
decades
to
elicitate
Universal
features
of
micro
swimmers
and
their
hydrodynamic
interactions,
as
well
as
to
model
the
behaviors
of
specific
microorganisms
such
as
ciliates
and
volvox
and
vulva.
Actually,
that's
what
this
kind
of
looks
like
is
volvox.
B
B
Very
similar
to
them,
so
there's
a
lot
of
math
here
and
then
you
get
into
the
Jeffrey's
equation
and
you
get
into
some
other
things
of
looking
at
the
effects
of
Shear
on
this
micro,
swimmer
or
at
least
on
a
sphere,
and
then
you
can
use
that
to
understand.
What's
going
on
in
the
micro
swimmer,
you
have
hydrodynamic
symmetry.
B
So
you
have
these
other
things
you
have
to
consider,
and
so
you
can
see
the
parallels
between
this
paper
and
the
last
paper.
They're
theoretical
extensions
for
more
General
shapes
as
promised.
You
can
look
at
helices
and
helicoid
objects.
You
can
look
at
corkscrews
which
are
movements
of
bacterial
flagellum,
so
the
flagellum
who's
in
a
corkscrew
motion
to
move
the
bacteria.
This
is,
of
course,
going
to
affect
the
surrounding
medium,
the
surrounding
liquid
medium
and
create
a
wake
and
create
inertia.
B
B
A
B
An
example
of
the
spermatozoa
and
the
swimming
trajectory
of
a
sperm
head-
and
you
know
this
is
something
that
is
sort
of
driven
by
Brownian
motion.
So
you
know
we
can
track
that
and
look
at
its
Dynamics,
but
it
also
interacts
with
its
liquid
environment,
and
so
you
can
introduce
a
share
to
the
sort
of
micro
swimmer
and
see
the
effects.
B
B
Drift
torque
as
it
turns,
and
that's
that
those
are
the
parameters
you
need
to
model
so
I'm
going
to
go
to
the
last
paper
here
in
the
interest
of
time,
and
this
paper
has
to
do
with
the
chemo
reception
and
chemotaxis
of
the
three
sphere
swimmer,
and
so
this
is
where
we're
considering
chemoreception
and
chemotaxis.
So
chemoreception
is
where
you're
taking
in
chemical
information
into
an
organism.
B
B
So
the
last
paper
we
talked
about
a
lot
of
these
types
of
models,
the
hydrodynamics,
the
Reynolds
number
and
then
there's
also
the
pickling
number,
which
is
another
parameter.
That
involves
that
we
need
to
consider
this
is
a
numerical
simulation
of
the
problem
with
a
finite
element
code
based
on
the
Fenix
Library.
B
So
then
we
also
consider
the
swimmer
as
a
learning
agent
moving
inside
a
fluid
that
has
a
concentration
gradient.
The
outcomes
of
key
learning
processes
show
that
learning
Locomotion
with
the
displacement
is
reward
is
significantly
easier
than
learning
chemotaxis,
with
the
increase
of
solu
fluxes
reward.
So
this
is
interesting.
What
they're
doing
here
is
they're,
taking
the
swimmer
and
they're
looking
at
these,
the
hydrodynamics
of
it,
but
then
they're
also
considering
the
swimmer
as
a
learner
of
chemical
information.
B
So
it's
learning
about
the
chemical
gradient
and
it's
producing
an
outcome
or
like
a
taxes
and
so
they're
able
to
use
Q
learning,
which
is
something
that
they
use
in
reinforcement,
learning
and
other
related
techniques.
It's
a
process.
So
it's!
It's
very
analogous
to
reinforcement,
learning,
it's
actually
I,
think
just
a
form
of
reinforcement
learning,
but
it's
not
quite
the
same
learning.
B
Locomotion
is
something
that
you
can
do
with
a
reward,
so
you
learn
based
on
repeated
rewards,
so
you
can
move
the
learner
or
move
the
the
swimmer
in
a
certain
direction
behaviorally
by
having
a
reward
that
is
sort
of
iterative
that
you
can
acquire
or
reward
at
every
step
and
eventually
you'll
learn.
This
is,
is
the
correct
Behavior,
so
we
can
use
displacement
or
we
can
use
cellular
flux,
and
it
turns
out
that
it's
easier
to
use
displacement
than
solid
flows.
B
The
chemotaxis
problem,
even
at
low
p
e-
and
this
is
the
pickling
number-
has
a
varying
environment
that
renders
learning
more
difficult
further
learning.
The
learning
difficulty
increases
severely
with
the
piclay
number.
The
results
demonstrate
the
challenges
that
natural
and
artificial
swimmers
need
to
overcome
to
migrate
efficiently
when
exposed
to
chemical
and
homogeneities.
B
So
the
play
number
has
to
do
with
chemical
gradients,
and
the
Reynolds
number
has
to
do
with
the
hydrodynamics,
and
so
we
need
to
again
understand
these
homogeneities
and
heterogeneities
and
then
chemical
homogeneities.
We
often
see
these
in
chemical
gradients.
Chemical
gradients
are
necessarily
linear,
and
so
they,
but
they
need
to
learn
how
to
migrate
according
to
these
signals
and
according
to
the
hydrodynamics.
B
So
this
is
you
know,
looking
at
these
type
of
things,
there
are,
a
lot
of
you
know
uses
for
understanding
this
understanding,
these
small.
You
know
these
small
systems.
You
know
it's
important
in
embryo
development,
in
immune
responses
and
in
understanding
things
like
algal
blooms
and
cell
migration
and
the
microbial
life
cycle,
so
assuming
Strategies
employed
by
natural
and
synthetic
microorganisms
are
highly
diverse,
again
understanding
the
basic
mechanisms-
theoretical
Studies,
have
frequently
relied
in
simplified
models
and
then,
of
course,
then
they
bring
up
the
squirmer,
which
is
this
model
of
spheroid
particles.
B
With
a
given
split
velocity
over
its
surface,
this
was
introduced
by
white
Helen
Blake
and
they
also
say
it's
done
in
Samuel.
Another
simple
early
model
is
the
tailor
sheet,
and
this
is
my
Childress,
a
flexible
sheet
that
swims
by
deforming
transversely
according
to
a
moving
wave,
and
so
they
addressed
in
this
study,
a
three
Superior
swimmer
model
introduced
by
najafi
and
golsanian.
B
So,
let's
see
if
I
can
find
a
picture
of
this
swimmer,
so
this
is
basically
it
this
three
sphere:
swimmer
you
have!
This
is
from
from
najafi
and
golsanian,
so
you
can
see
that
you
have
three
spheres
they're
joined
by
these
two
rods
and
their
variable
length,
so
you
have
l1t
and
l2t.
These
are
different
lengths
between
the
different
spheres,
so
the
first
sphere
is
way
out
in
the
front.
B
And
so
assuming
no
external
forces
acting
on
the
swimmer,
the
hydrodynamic
problem
reads
given
x,
0,
L1,
T
and
l2t
for
t
or
times
larger
than
zero,
find
x,
t,
u
x,
t
and
pxt
such
that
it's
subject
to
these
constraints,
the
gradient
of
the
gradient.
So
what
we're
trying
to
do
is
we're
trying
to
find
the
optimal
solution
in.
B
So,
given
the
initial
concentration
field
for
times
larger
than
zero,
find
cxt
satisfying
these
factors,
so
this
is
again
the
gradient,
we're
looking
at
the
boundary
conditions
that
are
far
away
from
the
swimmer,
so
we
need
to
optimize
for
that,
and
so
you
deal
with
the
salt,
the
solute,
the
diffusivity
of
the
solute,
and
so
this
is
a
moving
radiant.
Basically
so
the
chemical
gradient
diffuses
over
time.
We
need
to
account
for
that,
and
so
we
I
think
that's
I,
think
that's
enough
for
this
paper.
B
This
is
the
swimming
gate
for
three
sphere:
swimmers,
the
sequences
from
top
to
bottom,
and
follows
enumeration
introduced
in
section
5.1
for
States
and
actions,
so
they're,
actually
looking
at
how
this
thing
is
doing
using
Q
learning
to
learn.
But
this
is
basically
the
configuration
here
s
equals
four
S
equals
two
and
then
a
equals
one
equals
two.
B
You
you'll
have
to
read
the
paper
to
understand
that
a
little
bit
better
how
they
have
to
set
up.
But
basically
you
can
see
that
the
swimmers,
the
shape
of
the
swimmers,
is
a
little
bit
different
for
each
of
these
differences
in
parameter
for
States
and
actions.
You
can
see
that
you
have
a
state
here
and
then
you
have
a
different
action
so,
and
this
finally
gets
into
the
Q
learning
and
they
talk
about
the
reinforcement,
learning
framework
and
some
of
the
things
that
they're.
B
So
we
Define
that
the
agent
must
move
exactly
one
arm
in
each
time,
interval
from
the
fully
extended
position
of
the
fully
contracted
position
or
vice
versa.
The
only
decision
to
be
taken
at
each
discrete
instant
T
is
whether
to
move
arm
one
action,
one
or
arm
two
action.
Two
at
each
of
the
four
possible
states.
There
are
just
two
possible
actions.
This
kinematic
restriction
on
the
agent
rules
out
the
possibility
of
the
stormwater
remaining
still
in
any
position.
B
D
A
D
A
B
I
think
that's
great
definitely
work
on
the
proposal.
First,
that
should
be
your
first
priority
and
then
you
can
work
on
this
other
stuff.
So
word
about
that.
The
gsoc
proposals
open
on
March
20th,
so
that's
coming
up,
but
that's
just
the
beginning
of
that
period.
There's
a
deadline
of
I
think
in
April,
where
the
submissions
have
to
be
made.
What
was
that.
B
4Th,
okay,
so
that
so
at
the
road
to
this
deadline
is
if
you're
interested
in
developing
a
proposal.
You
know
sushma
has
already
gotten
his
draft
together
and
it
looks
good
so
far
you
know
you're
gonna
have
to
probably
make
some
iterations
of
it.
Where
you,
you
know
you
have
to
like
kind
of
you
want
to.
Let
me
see
it
before
you
submit
it.
Definitely
and
I
can
give
you
pointers
and
how
we
can
refine
it,
and
so
I
would
do
that
as
soon
as
possible.
B
I
wouldn't
wait
until
March
20th,
but
what
people
tend
to
do
is
they
tend
to
have
this
period
where
they're,
interacting
with
their
Mentor,
which
would
be
me
and
I,
encourage
you
to
interact,
and
you
know,
maybe
make
some
contributions
before
the
deadline.
So,
if
you're
interested
in
you
know,
if
you
want
to
explore
the
org,
that's
fine
I
can
give
you
I
can
send
you
things
if
we
usually
do
this
in
the
slack.
B
So
if
you're
in
the
diva
worm
channel
on
the
slack,
that's
good,
we
also
have
a
Devo
learn.
Channel
evil
learn
Devo
graph
Channel,
which
is
also
something
somewhere
where
we
usually
post
a
lot
on
just
on
devil
or
an
indivo
graph,
so
that
you'll
want
to
join
that
channel
as
well,
and
so
when
we
get
closer
to
the
deadline,
you
know
I'll
be
sending
out
materials,
maybe
more
on
some
of
these
topics.
B
It's
basically
where
you
have
an
introduction
to
yourself,
and
then
you
describe
your
approach
to
the
problem,
so
you
lay
out
what
you're
going
to
do
and
how,
if
it
works
or
not,
sometimes
that'll
require
you
to
do
a
little
demo
and
put
screenshots
in
the
the
proposal,
and
sometimes
it
just
requires
you
to
be
very
clear
on
the
outline
like
what
is
this
thing
I
want
to
do?
How
does
it
relate
to
what
has
been
done
in
the
org?
B
E
B
That's
that's
really
where
you
want
to
go
and
I
think
sushma
just
introduced
I
think
has
a
pretty
good
start
on
this.
But
if
you
want
to
do
this,
if
you
want
to
be
really
I
think
have
a
really
unique
proposal.
I
would
go
through
all
the
materials
and
see
what
we
have
and
you
know
kind
of
just
describe.
You
know
how
you
know
you
might
tie
this
into
some
of
the
other
organizational
things
so,
like
you
know,
we
do
we're
doing
image
segmentation.
C
B
Then,
of
course,
we
have
the
graph
neural
network
stuff,
but
we
also
have
the
lineage
tree
stuff
and
we
have
other
types
of
things
that
we've
done
in
the
past:
it
both
within
gsoc
and
then
in
the
diva
GitHub
repositories.
We
also
have
things
that
aren't
gsoc
related
and
those
may
or
may
not
be
relevant,
but
you
know
they
may
tie
into
those
and
then
you'll
need
a
schedule
which
is
where.
B
B
You
know
kind
of
think
about
this
as
like
a
schedule
where
you
go
through
your
work,
but
you
have
a
contingency
plan,
and
so
the
contingency
plan
is
where,
if
you're,
you
know,
you
have
this
path
to
success
right.
You
want
to
do
all
these
different
things
and
steps,
but
what,
if
something
doesn't
happen,
the
way
you
want
it
to
what?
If
something
doesn't
Implement
correctly
or
what?
If
it
takes
you
a
while
to
implement
something,
you'll
want
to
sort
of
try
to
implement
things
very
early,
maybe
even
during
our
proposal
period.
B
But
you
know,
sometimes
you
don't
know
when
you
get
into
something:
we've
had
problems
with
like
people
running
out
of
computational
resources
or
having
problems
with
bugs
and
that's
fine,
that's
perfectly
fine!
That's
what
the
project
the
program
is
about.
You
want
to
be
able
to
overcome
those.
So
in
your
proposal
schedule
you
want
to
have
like
your
schedule.
You
know
this
things
are
implemented.
B
So
if
you're
trying
to
implement
a
toolbox,
for
example,
and
the
Toolbox
isn't
implementing
you,
can
you
can
stop
doing
that
for
a
while
move
to
something
else
and
then
try
to
solve
that
toolbox
problem?
And
then
once
you
get
that
fixed,
then
you
can
Implement
that.
But
you
haven't
wasted
a
bunch
of
time
because
it's
a
very
tight
schedule.
It's
like
you,
know,
12
weeks
or
something.
So
you
can't
really.
B
You
can't
really
take
time
off
to
do
to
figure
it
out
and
that's
one
of
the
challenges
of
gsoc
is
that
they're
trying
to
teach
people
how
to
you
know,
work
within
a
very
tight
schedule
and
get
things
done,
and
so
that's
that's
what
my
my
advice
to
people
I
also
suggest
that
you
know.
If
you're,
you
know
you
don't
have
to
stick
exactly
to
that
schedule.
We
can
rescope
the
project,
but
it's
good
to
have
it
pretty
close
to
what
you're
going
to
do
because
rescoping
it
is
a
problem.
B
A
D
B
I
think
probably
I've
had
people
in
the
past,
maybe
eight
to
ten,
because
you're
gonna
have
like
references.
You're
gonna.
Have
your
schedule
you're
going
to
have
your
your
stuff
that
you,
you
know
your
images
I,
think
images
are
always
avoid
all
the
successful
ones
I've
had
for
the
most
part,
I've
had
images.
You
know
just
where
people
will
try
out
something.
They'll
write
a
little
notebook
and
they'll
have
them.
B
B
B
E
E
E
A
E
There
we
go
okay,
our
prokaryotes,
they
look
like
bacteria
but
they're,
very
different
from
bacteria
chemicals.
Well,
some
of
them
have
flat
polygons,
so
I
and
they
may
go
back
to
the
origin
of
life.
So
I'm
working
on
the
hypothesis
is
that
origin
of
life
was
a
transition
from
what
are
called
shape,
drops
all
polygonal
droplets,
they're
flat
so
to
the
original
archaea.