►
From YouTube: DevoWorm (2021, Meeting 14): Biological Cell Separability, Cybernetic Logic, Learning and Memory
Description
Separability and Overlap in Cell Microscopy, Discussion About Version-controlled Books, Correlation Does Not Equal Causation, Bioelectricity as a Form of Organismal Coherence and Cybernetic Logic as a Form of Organismal Branch Proliferation, Learning and Memory in Developmental Systems. Attendees: R Tharun Gowda, Debojyoti Chakraborty, Ujjwal Singh, Jesse Parent, Richard Gordon, Susan Crawford-Young, Krishna Katyal, Vrutik Rabadia, Bradly Alicea, Mayukh Deb, Mainak Deb, Vash Yadi, Aswath Narayana, Aayush Kumar, Shruti Raj Vansh Singh, and Muhammed Abdullah
C
D
E
There's:
okay,
yeah:
here's
some
other
patterns.
D
Yeah,
those
are
nice.
I
I
do
have
some
some
shells
somewhere,
they
weren't
necessarily
collected
for
their
patterns,
so
that
as
well.
E
You
know
one
of
the
wolfram
patterns
and
can
we
show
that
the
time
course
matches
that
of
some
real
pattern
or
to
put
it
another
way,
can
you
go
backwards?
Can
you
see
the
pattern
and
figure
out
the
rules?
Oh
okay.
Okay,.
F
A
E
G
H
E
Patterns
are
made
by
by
the
automatic
rules
of
golfer,
and
they
some
of
them
look
similar
to
it.
But
the
question
is:
can
you
can
you
image
the
pattern
and
then
derive
the
rules
from
the
pattern
in
order
to
do
the
inverse
prophet.
G
H
E
So
so,
basically
you
want
to
go
from
from
the
images
to
three
dimensions,
to
unwinding
in
two
benches.
G
A
Yeah,
that
was
good.
Thank
you,
dick
yeah.
That's
another
problem
for
the
group.
We've
been
talking
about
that
for
a
couple
years
where
you
know
we
have
to
want
to
figure
out
how
to
image
the
whole
surface
of
the
shell,
and
you
know
turn
it
like.
This
is
kind
of
like
the
the
thing
we
have
with
the
ball
microscope
and
the
flipping
microscope,
where
you
have
these
images
that
you're
collecting
from
different
perspectives
and
you're,
trying
to
stitch
them
together,.
A
E
G
E
A
So
we
actually
we're
talking
about
morphozoic,
which
was
the
cellular
automata
approach
to
it,
which
is
to
take
like
a
cellular
automata
and
then
take
the
output
and
put
them
into
a
neural
network
and
classify
the
outputs,
but
that
didn't
really,
you
know,
get
us
exactly
where
we
wanted.
A
I
might
put
a
link
to
neural
cellular
automata
in
the
chat
and
that's
a
paper
we
went
over
a
while
back.
This
is
the
distilled
publication
on
by
mike
11
at
all
on
on.
You
know,
looking
at
using
a
neural
network
or
a
deep
network
at
the
beginning
of
the
process,
and
it
generates
patterns
that
then
you
put
into
a
ca
and
and
that's
that's
the
way
they
approach
it
so
deebo
says:
can
you
get
the
rule
from
the
pattern
and
noyoke
says?
Yes,
that's
what
happens
during
training.
A
E
E
You
know
computer
can
do
it
by
just
training
and
giving
correlations,
but
I
don't
think
humans.
Do
it
that
way.
A
We
can
interpret
the
rules
of
feature
visualization
and
there's
another.
So
myoka
has
a
thing
called
differentiable
morphogenesis,
which
I
guess
is
something
he
worked
on
in
one
of
his
projects.
He
he
did.
He
implemented
some
neural
cellular
automata
on
pi
torch.
A
If
you
go
to
the
I'll
share
my
screen,
if
you
go
to
the
link
in
the
chat,
it's
right
there,
so
this
is
his
work
on
it,
where
he's
taking
different
patterns
and
turning
them
into
other
patterns,
which
is
not
quite
the
same
thing
as
what
we're
talking
about.
But
this
is.
E
E
E
A
E
There's
lots
of
literature
on
on
saying:
oh,
here's
a
pretty
picture
of
a
snail
and
here's
the
pattern
that
looks
something
like
this,
but
there's
isn't
any
that
quantitatively
tries
to
relate
scale
patterns
to
a
set
of
rules.
A
So
that
might
be
something
interesting.
I
know
surety
actually
is
preparing
a
application
with
some
interesting
things
in
that
direction.
So
we
can
talk
about
that
more.
You
know,
anyways.
B
A
Welcome
to
the
meeting
I
know
a
lot
of
you
have
been
busy
this
weekend
submitting
puzzles.
B
A
Okay,
welcome
to
the
meeting
everyone.
I
know
a
lot
of
you
this
weekend,
we're
working
on
your
proposals,
and
so
those
are
at
the
deadline.
I
think,
is
either
today
or
early
tomorrow.
My
time
which
just
make
sure
you
check
the
time
if
you
haven't
submitted
already,
because
that's
in
that
the
time
change
the
time
difference
between
your
location
and
the
time
when
a
proposal
is
due
is
often
different,
and
so
you
want
to
make
sure
you
have
it
in
on
the
right
hour
of
the
right
day.
A
So
just
make
sure
that
everything-
and
I
sent
out
some
reminders
that
if
you
go
into
the
submission
portal,
for
example-
and
you
have
a
draft-
you
need
to
change
it
to
final
version
before
it
gets
accepted.
A
Little
things
like
that,
and
so
I
think,
if
you
know
that's
I've
reviewed
a
lot
of
people's
proposals
and
they
all
look
pretty
good.
So
good
luck
on
on
this
to
everyone
and,
unfortunately,
we'll
be
able
to
select
everyone-
or
you
know
we'll
have
like
maybe
one
or
two
people
at
the
end,
who
get
actually
selected
for
the
for
the
actual
program.
But
if
you
don't
get
selected,
I
definitely
welcome
you
to
continue
working
on
some
of
this
stuff.
A
It's
definitely
there's
been
a
there's,
been
a
lot
of
really
good
ideas
and
we
can
always
turn
it
into
something
else.
Maybe
another.
You
know
a
gsoc
project
next
year
or
some
other
thing.
You
know
if
you
are
doing
work
on
something
that's
really
relevant
to
what
we're
doing.
We
can
pick
up
on
a
theme
and
pull
it
forward.
So
there's
definitely
all
sorts
of
opportunities
for
continuing
with
that.
So
and
then
we'll
be
reviewing
the
applications
in
the
next
couple
weeks
and
well.
A
I
don't
know
when
the
final
decisions
are
due,
I
think,
in
about
a
month
and
then
we'll
announce
those
and
then
the
community
period
will
start
after
that.
So
that'll
be
you
know
something
that
we
do
here.
We
kind
of
stress
the
we
kind
of
stress
the
open
room
foundation
and
the
things
that
are
going
on
here.
So
I
think,
if
you've
been
attending
the
meetings,
you
kind
of
know
what
the
community
is
all
about.
So
so
I
guess
we
have.
I
wanted
to
open
the
floor
to
anyone
who
wanted
to
present.
A
F
A
Be
very
long,
it
can
be
very
you
know,
just
seating,
your
pants,
whatever
so
diva.
Why?
Don't
you
you?
Are
you
ready
to
go
or.
H
Okay,
am
I
I'm
going
to
sell
my
speed.
H
Okay,
so
today
I'm
going
to
present
a
new
technique
and
the
approaches
to
detect
and
count
the
diatoms
using
some
rc
energy,
yellow
based
methods,
which
I
get
the
difference
from
the
counting
blood
cells
like
there
are
some
paper
which
demonstrate
how
you
can
count
white
and
other
platelets
everything
using
this
kind
of
techniques.
H
I
H
H
It
works
so
for
this
I
think
the
most
good
one,
which
is.
H
Movement
of
the
of
a
vascular,
colony
and
yeah,
so
like
that
way,
we
can
understand
that
and
also
create
some.
H
Is
like
currently,
we
have
the
version
three
pre-trained
models
which
extract
the
phenotypic.
H
Shape
of
the
cells
and
also
but
they
don't
have
any
count
method.
As
far
I
see
that
yeah,
and
so
let's
go
to
the
next
slide.
H
Now,
while
I
am
preparing
my
proposal,
so
I
was
searching
for
new
methods
to
detect
the
and
count
the
diatoms
so
from
there.
I
got
this
new
idea
of,
like
some
researchers,
detecting
blood
cells
using
faster
or
cnn
and
yellow
based
techniques,
as
I
got
that
from
another
it's
a
core
base
which
is
open.
H
H
And
there
are
some
like
some
experimental
result
of
the
like
as
per
the
paper,
and
they
they
also
use
transfer,
learning
using
rxnet
or
dtg16,
which
we
also
can
use
for
our
problem,
like
just
changing
the
data,
and
yes
and
after
that,
these
are
the
like.
These
are
the
some
photo.
Some
images
are
generated,
and
this
is
the
link
from
where
I
got
the
code
and
but
sometimes
it
the
model
do
something
wrong,
like
overlaps
or
counter
same
cell
twice.
So
for
that.
H
We
are
also
want
to
take
the
movement
of
the
diatoms,
so
we
need
to
some
tweaking
in
the
model
to
improve
the
detecting
and
counting
the
diatoms
easy
from
here.
Anyone
have
any
question.
H
H
A
Up
yeah,
so
you
can
look
through
them.
Anyone
have
any
questions
about
any
specific,
slides
or.
A
A
All
right,
yeah
he's
get.
I
guess
he's
gotta
talk
to
someone
for
a
minute,
but
yeah.
That
was
an
interesting
talk.
I
I'm
interested
in
his
well.
That's
the
book
that
we
took.
We
talked
about
putting
together
a
documentation
book.
A
A
A
Deebo,
please
define
your
abbreviation,
so
yeah,
that's
important
when
you
present
on
something
to
put
it
the
abbreviations.
You
know
in
full.
Did
you
succeed,
meaning
I
guess
like
on
the
data.
H
I
think
everything
is
in
the
like
the
open
source
project
like
yesterday.
I
found
this
project,
so
it's
a
week.
I
created
the
presentation
very
quickly
so,
like
I
don't
get
the
time
to
get
the
time
to
make.
You
know
my
local
system
so.
A
Yeah
and
then
we're
working
on
this
we're
working
in
a
book
actually
on
the
mathematical
bio
biology
of
diatoms.
So
there
might
be
an
opportunity
to
contribute
to
that
so
yeah
and
then
this
is
the
link
to
debose
notebook.
So
could
you.
A
I
wanted
to
look
at
some
of
the
one
of
the
couple
of
things
so
that
where
you
did
the
thing
with
the
k
ns
and
the
yeah
that
one
that
slide
six.
A
Well,
the
next
one,
maybe
yeah,
so
that
one
is
okay,
so
sometimes
these
models
do
some
wrong
predictions
yeah
yeah
yeah.
To
avoid
this,
these
can
k-nearest
neighbor
an
intersection
over
union-based
verification.
So
this
is
a
verification
system
for
finding
like
the
overall.
H
A
H
H
A
And
then
the
slide
above
it
is,
what
is
it
the
number
five.
G
A
Yeah
that
one,
so
that's:
okay,
that's
where
you're
doing
transfer,
learner
and
custom
data
using
alexnet
or
vgg16.
Okay,.
H
Like
they
are.
A
H
Like
the
approach
I
found
from
the
research
paper,
there
is
like
there
is
no
implementation
or
no
code
is
open
source,
but
there
they
have
the
regressor
with
the
classifier
so
like
fast,
it
will
classify
the
like.
First,
it
will.
H
Cells
in
blasters
and
after
that,
the
repressor
I
think
regressor-
will
count
the
numbers
of
the
black
stuff.
Okay,.
A
A
Some
information
about
her
so
if
you're
interested
this
is
something
you
can
check
into,
I
mean
I'll,
probably
probably
we'll
probably
want
to
talk
about
it,
some
more,
but
you
know
just
to
give
an
idea
of
like
what
you
know
how
we
might
move
this
stuff
forward
and.
A
Oh
well,
we
could.
I,
I
have
our
contact
information
too,
so
we
can
talk
about
it
later,
but
my
oak
said,
as
far
as
I
remember,
the
classifier
determines
which
classes
are
present
in
their
image
and
the
regressor
estimates
their
positions,
width
and
height,
so
the
classifier
actually
determines
the
sort
of
what
category
they
belong
to.
Yeah.
F
H
And
badly,
have
you
take
a
look
at
the
draft?
I
send
you.
A
Yeah
I
looked
it
over
looks,
I
mean
it
looks
good,
but
I
have
to
evaluate
it
a
little
bit
more,
but
it
looks
pretty
good.
F
A
A
Solution,
though,
we'll
have
to
work
with
it
a
little
bit
more.
A
A
B
A
H
A
G
A
So
does
anyone
else
have
anything
they
want
to
present
today?
Is.
A
A
Okay,
so
I
usually
says
yeah
the
regressor
tries
to
learn
the
numeric
values
for
the
bounding
boxes
and
the
vvg
backbone
tries
to
classify
the
inbox
object
inside
the
bounding
box.
Jesse
says:
git
book
looks
cool,
as
you
all
says,
regressor
can
be
used
to
classify
only
when
classes
are
continuous,
which
is
not
the
case
here.
I
believe
so.
There's
a
continuous
versus
discrete.
H
Because
the
because
the
moment,
the
diatom
colony
has
a
number
right-
that's
a
discrete
property
of
the
diagram
colony.
So
from
that,
we
can
figure
out
some
equation
or
something
like
that.
Like
some
model,
if
there
exists,
we
can
use
that.
So
that's
why
we
can
tweak
the
model
as
power
as
per
our
need.
A
Okay,
all
right
yeah
I
mean
there
are
different
ways.
You
can
do
it.
I
know
people
have
come
up
with
different
tricks
and
sometimes
they're
in
the
different
fields.
Statistics.
Literature,
like
you,
know
the
the
psychologists
and
economists
come
up
with
all
sorts
of
interesting
ways
around
the
sort
of
the
assumptions
of
these
statistical
models
because
they
have
different
types
of
data
from
you
know.
What
a
statistician
might
assume
is
the
you
know,
object.
A
F
A
Data,
so
so
malik
says.
Sadly,
good
book
does
not
support
themes
unless
we
use
the
paid
version.
A
great
free
alternative
is
hugo,
so
good
book
actually
has
a
paywall
on
it.
At
some
point,
read
the
docs
also
be
a
good
alternative
yeah
that
read.
The
docs,
of
course,
is
basically
very
similar
and
maya
put
a
link
in
the
chat
for
hugo.
A
K
A
So
let's
see
brutic
is
here
and
the
verdict
yeah
usually
is
here.
Ayosh
joined
us
jesse's
been
here
my
krishna's
here,
vasha's
here
yeah.
So
let's
see
okay,
my
oak
says
they
support
direct
mark
down
yeah.
I
think
if
we
just
create
you
know,
if
we
have
some
good
content
in
markdown,
we
can
put
it
on
whatever
platform
we
want.
A
You
know,
so
I
guess
we'll
then
move
on
to
some
other
business.
If
no
one
else
wants
to
present
I'm
going
to
present
my
screen
again
and
not
open.
My
chat
again
was.
L
L
A
Okay,
let's
make
it
read
the
docs
after
gsoc
yeah
we
can
so
diva
learn,
read
the
docs
open
worm
has
a
read
the
docs
and
I
you
know
I
have
some
information
on
diva
worm
on
there
and
it's
a
little
dated
now,
of
course,
but
yeah.
We
should
make
one
for
divalern,
because
we've
not
like
that.
That
needs
to
be
picked
up
again.
A
A
This
is:
is
this
like
a
challenge
call
to
action?
One
of
these
grand
challenge
type
things
dick.
H
I
know
one
question
like
open:
one
is
signed
up
for
the
g
sword,
yeah.
A
Well,
yeah
gsod,
I
don't
know
that's
a
little
bit
different
this
year,
so
they're
they're
the
projects
are
already
submitted
and
they're
not
like.
I
don't
know
how
they're
doing
it
this
year,
it's
like
they've
got
a
group
of
people
they
already
selected
and
they're.
Doing,
like
I
put
in
a
different
I
didn't
put
in
this
project.
I
didn't
think
of
that
at
the
time.
So
you
know
that's
something
that
we
probably
wouldn't
be
able
to
do
much
with
this
year,
but
next.
H
L
A
L
I
L
A
L
Each
other,
so
there
are
some
good
space
or
models
field.
So
what
I
did
was
I
took
the
image
I
locally
enhanced
images
like
if
we
have
a
pixel,
I
added
the
densities
of
surrounding
pixels,
so
which
would
eventually
help
me
to
get
the
center
pixel
of
cell
to
the
highest
density.
So
that's
what
I
did
by
taking
and
candidate
region
and
taking
some
few
diameter
diameter
dimensions
which
would
be
possible,
like
we
have
maximum
80
pixels.
We
had
maximum
80
pixels
of
dimension
in
our
data,
so
I
took.
L
L
B
A
A
Well,
he's
yeah
he's
got
a
lot
of
slides
here
if
you're
looking
through
them
put
the
slides
in
the
chat
and
he's
talking
about
refining
the
centroid.
So
he's
done
a
lot
of
work
on
that
he's
got
output
which.
A
So
this
looks
like
an
interesting
algorithm.
I'm
not
sure
what
happened
to
rudick,
but
it
looks
really
interesting.
I
hope
he's
able
to
join
us
again.
I
don't
know
what
happened
here.
A
B
I
A
Good
work
for
dick
yeah.
I
A
A
The
current
algorithm
is
implemented
on
2d
images
due
to
which
there
is
a
loss
in
data
due
to
varying
pixel
intensity,
which
arises
due
to
different
z
coordinates.
So
I
think
maybe
it
was
verdict.
We
talked
about
the
z,
coordinates
or
maybe
josh
as
well,
that
you
have
this
stack
of
images
when
we're
using
the
raw
images.
M
A
I
think
you
were
talking
about
this
like
manipulating
the
pixels
and
yeah.
I
think
you're
at
this
side
here
number
eight.
Yes,.
L
L
L
L
L
L
These
are
the
useful
centers,
but
the
problem
arises
here
is
there
are
means
few
centers
detected
really
close
to
each
other,
like
for
a
single
cell?
There
will
be
a
multiple.
There
will
be
multiple
centers
detected,
so
we
have
to
keep
minimum.
L
Yeah,
so
we
got
this
output
now,
as
this
is
a
2d
image,
there
are
no
thread
planes
involved
in
it,
so
you
can
see
there
would
be
some
like
if
two
cells
are
overlapping
with
that,
you
have
converted
into
locally
enhanced
image.
But
if
the
cells
are
not
overlapped,
you
can
see
there
are
four
cells
nearby
each
other.
Still
we
are
getting
three
of
them
and
one
is
missed
due
to.
L
Thing
in
3d,
by
taking
the
z
coordinate,
also
are
the
images
obtained
from
epic
data
set
2
which
had
31
slices
in
that
coordinate,
so
it
worked
really
fine
like
it
gave
me
output
with
errors
of
like
10
to
15,
plus
minus
10,
to
15
pixels
and
in
z,
coordinate.
It
was
always
a
whole
number
because
I
went
slice
right,
so
this
was
what
I
was
doing.
I
have
been
optimizing
this
algorithm
on
2d
images
it
works.
L
A
L
F
A
Yeah,
I
think
thank
you
for
that.
That's
good,
I
think
yeah
verdict
and
ebo
have
some
things
in
common
there
in
terms
of
like
looking
at
like
how
you
deal
with
noise
and
overlap
and
that's
an
important
issue,
I
mean
it's
easy
to
think.
Well,
you
can
just
you
know,
follow
the
boundaries
between
cells,
and
sometimes
you
know
in
some
of
these
data
sets
you're
just
taking
a
slice.
You
know
a
slice
of
the
the
sample,
and
so
in
doing
so
you're
going
to
pick
up.
A
You
know
overlapping
cells
and
things
like
that,
and
especially
when
it's
not
something
you
know
well,
even
with
an
embryo.
You
have
that,
but
especially,
if
you
have
like
you
know
some
group
of
cells
or
in
a
colony
or
or
maybe
you
know
in
different
layers
that
are
kind
of
interdigitated
you're
gonna
have
to
deal
with
that
issue.
So
I
think
those
are
two
very
good
solutions
to
that.
If
you
could
share
your
your
notebooks
or
whatever
you
have
in
maybe
put
them
in
the
slack
as
well,
I
think
people
would
appreciate.
A
Okay,
thank
you
and
then
dick
put
some
more
references
on.
A
Oh
yeah,
no
problem
dick
put
some
references
to
idi
ac
idiak.
So
this
is
this
approach
and
so
there's
a
lot
of
stuff
on
diatom
identification
here.
Different
techniques
like
binary
masks
of
concentric
rings,
so
people
have
been
doing
this
for
a
long
time
coming
up
with
different
techniques
or
you
know.
E
A
Okay,
thank
you
so
yeah
check
that
out
if
you're
well,
I
would.
M
M
A
A
A
N
N
N
Examples
that,
like
the
say
when
the
sales
of
soft
drinks
increases,
therefore
the
debts
of
drowning,
also
increases.
Therefore
soft
drinks
cause
drowning,
but
that's
not
true.
The.
N
Temperature
and
in
summers
people
usually
tend
to
go
to
near
water,
go
for
you,
know
water
sports
and
go
for
swimming.
So
it's
you
know
going
it's
the
temperature
that
is
causing.
N
Yes,
yes
yeah,
so
one
this
is
in
the
upper.
It
shows
that
global
average
temperature
and
number
of
pirates.
So
if
one
can
see
this,
it
can
be
easily
shown
that
rising
temperature
is
kind
of
wiping
out
pirates,
but
that's
not
true,
it's
the
advancement
of
the
civilization,
so
it
has
nothing
to
do
with
the
global
warming.
It.
A
N
Do
with
the
you
know
how
the
civilization
is
progressing,
so
one
can
these
charts?
Maybe
you
know
depicting
some
relationships,
but
it's
a
third
variable
that
this
chart
are.
You
know
more
connected
to
like
files
and
global
temperatures.
Don't
have
any
correlation.
N
Technologies,
the
industries,
it's
the
global
trade
that
corresponds
to
increasing
temperature
and,
you
can
say,
falling
off
pirates.
So,
for
example,
there
are
so
many
things
that
usually
so
show
correlation,
but
they
are
not
the
causation.
A
Like
I
was
seeing
that
when
the.
N
In
a
city
named
kentucky
when
the
when
the
sales
of
boats
fishing
boat
increases,
there
were
also
more
marriages,
so
one
can
deduce
that
okay
selling
off
boats
and
buying
off
what
is
causing
marriages.
But
it's
not
true.
It's
actually
the
marriage
season
and
the
fishing
season
is,
you
know
overlapping.
N
So
it
is
the
third
variable.
N
Not
the
relationship
in
between
two
so
like
marriage
and
selling
of
boats
is,
you
know,
one
can
say
that
independent
from
each
other.
N
N
Some
human
interventions,
so
we
can,
you
know
we
can
justify
that.
The
correlation
is
just
you
can
say
happening
randomly
or
it
has
some.
A
Okay,
yeah
yeah,
there's
a
lot
of
literature
on
that
you
can
even
get
lost
in
it
in
terms
of
like
you
know
what
what's
a
neat
correlation
you
can
make,
or
you
know
what
requires
a
larger
scale
analysis
like
you
know,
if
you're
looking
at
the
intervention
of
different
variables-
or
maybe
you
have
like
you-
know
something-
that's
just
like
correlated
because
they're
by
chance
that
thanks
for
that,
that's
good
it
so
that
paper
yeah
we're
working
on
it's
actually
more
in
our
other
group,
but
we're
yeah
we'll
talk
about
more
about
that
later.
A
But
I
think
that
was
useful
for
this
group.
I
wanted
to
talk
about
some
things
shift
to
something.
Actually,
let
me
go
through
the
a
little
chat
here.
Another
example:
all
deaths
are
due
to
covid.
Therefore
we
are
otherwise
immortal
usually
says
how
about
all
deaths
because
of
covet
premise
is
not
correct.
I
guess
in
this
case
dick
says,
because
anyone
who
dies
is
now
attributed
to
covenant.
Politicians
thank
you
for
the
update.
Krishna
gives
a
thumbs
up
so
yeah
yeah.
A
A
So
this
is
a
paper
in
the
biosystems
journal.
It's
in
one
of
the
special
issues,
it's
in
this
differentiation
waves,
special
issue-
and
this
is
on
something
called
janus
face
logic.
A
A
Dick
has
a
huge
book
on
this
topic
from
1999
and
so,
but
instead
of
going
through
the
whole
book,
you
can
read
this
tutorial
and
it
gives
you
some
idea
of
what
those
phenomena
are,
and
so
the
abstracts,
as
differentiation
waves,
offer
a
different
perspective
on
causality
and
embryogenesis
from
that
of
molecular
developmental
biology.
A
So
you
know
back
going
back
to
krsna's
thing
about
causation
and
correlation.
We
also
deal
with
causality
and,
of
course,
causality
is
a
minefield
in
and
of
itself.
But
in
this
case
you
know
we
we
look
at
like
the
embryo
and
we
have
a
lot
of
the
genetic
networks
mapped
out
and
things
like
that.
But
we
still
don't
understand,
maybe
the
causality
at
the
at
the
level
of
the
cell,
and
so
this
gives
us
a
framework
to
think
about
that
and
so
rob.
I
think
this
sounds
like
rob's
phrasing
here.
A
A
So
I
have
some
things
in
the
chat.
Oh
thank
you.
Deebo
for
attending
dick
put
a
link
to
his
book
in
in
the
yeah
1760
pages.
He
also
has
a
shorter
book
here,
but
this
of
course
and
jesse
says
looking
ahead
and
behind.
So
that's
the
janus
face
analogy,
but
the
important
thing
here
is:
it's
a
binary
opposition
so
with
the
with
local
and
global
top-down
bottom-up
dynamics.
This
is
the
janus
phase
approach.
A
It's
not
reductionist,
it's
distinct
from
emergence
and
outlines
the
process
theoretically.
So
this
this
paper
goes
through
a
lot
of
this
stuff
kind
of
describes
it
in
and
if
you're
not
familiar
with
some
of
these
concepts,
you
know
it
gives
a
it
gives
a
little
bit
of
background
into
it.
A
Although
you'll
still
have
to
look
things
up,
I
encourage,
if
you
read
the
paper,
to
look
things
up
and
he
gonna
they
walk
through
a
typical
what
we
call
a
differentiation
tree,
which
is
where
you
start
with
a
zygote,
and
you
get
these
differentiation
events
where
you
get
the
differentiation
of
tissues,
sometimes
cells,
that
divide
and
then
differentiate,
diff
into
different
functional
lineages,
and
that's
what
you
this
process
here
and
these
coins
are
the
janus
faces.
A
So
at
every
step
in
this
process,
you
get
this
binary
faced
decision,
so
you
you
know,
you
have
a
switch
that
happens
here,
and
so
you
know
I
mean
I'm
not
going
to
go
deeply
into
the
paper,
but
I'm
just
going
to
say
that
this
is
something
it's
pretty.
It's
pretty
digestible.
A
We
have
some
a
number
of
consequences
for
evolution
here,
which
is
talking
about.
You
know
different
things
in
in
development
that
this
model
of
development
has
a
consequence
on.
So,
for
example,
there
are
things
like
heterocrony
or
the
evolution
of
neoteny
or
metamorphosis,
which
is
when
you
have
one
organism
that
transforms
from
one
phenotype
to
another
in
development
like
a
caterpillar
into
a
butterfly,
you
get
different
types
of
asymmetric
cell
divisions,
and
so
there
are
all
these
different
things
that
it
explains.
A
So
I
think
that's
I'm
going
to
put
the
link
to
the
drive
here
again
for
the
with
the
papers
in
it
and
I'll
put
that
into
the
chat
and
that
that
contains
this
paper.
Thank
you
as
well
for
attending
yes,
says,
make
sure
you
submit
your
final
proposals
before
the
deadline
and
once
again
and
a
reminder,
and
then
dick
says
that
those
gold
little
circles
were
roman
coins
with
the
janna's
face
on
them,
so
they
used
to
have
coins
with
the
janice
face
on
them.
A
I
wanted
to
go
to
some
other
things.
I
have
a
couple
other
things
to
present
on.
I
wanted
to
get
into
or
kind
of
highlight
them.
Susan
sent
me
this
link
to
this
talk
by
michael
levin,
who
is
a
very
prolific
developmental,
biologist,
computational
biologist,
and
this
is
a
ted
talk
he
gave
back.
I
don't
know
how
long
ago
I
think
it's
fairly
recent,
but
this
is
on
a
lot
of
his
work
on
looking
at
bioelectricity
and
so
michael
levin
works
on
these
model.
Organisms
called
flatworms
and
you
can
see
them
here.
A
A
You
can
train
the
flatworm,
then
you
can
obliterate
everything,
but
a
single
cell
that
single
cell
will
turn
back.
It
will
regenerate
back
into
that
flatworm
that
actually
can
retain
the
memories
of
the
flatworm,
the
original
flatworm.
So
this
is
really
fast,
a
fascinating
model
system
and
he
studies
bioelectricity.
So
he
studies
the
bioelectric
communication
between
cells-
and
this
is
basically
very
similar
to
what
neurons
do
when
they
have
an
action
potential.
A
So
action
potentials
are
a
type
of
bioelectric
communication
but
they're
very
distinct
signatures
that
come
from
these
neuronal
cells,
but
you
also
observe
electrical
activity
in
non-normal
cells
like
in
skin
cells
or
in
liver
cells
or
any
other
kind
of
cell.
You
can
think
of.
They
have
some
electrical
potential,
but
it's
not
like
a
action
potential
where
you
have
a
spike.
A
It's
just
a
you
know
some
sort
of
electrical
signal
and
so
that
electrical
signal
you
can
study
that
and
see
if
it
has
an
effect
and
he's
finding
that
it
does
have
some
organizing
effect
on
the
phenotype.
A
So
you
know
if
you
have
a
bunch
of
developmental
cells,
that
differentiate
and
they
have
this
electric
potential-
even
stem
cells,
of
an
electric
potential-
that
electric
potential
was
used
to
coordinate
the
cells
and
to
form
features
of
the
phenotype
so
he's
he
does
a
lot
of
work
with
that
and
it's
a
really
interesting
area-
and
he
talks
about
this
in
this
talk,
but
he
also
talks
about
this.
This
thing
called
xenobots,
which
are
this
new
model
system
that
he's
developed
with
some
collaborators,
and
this
is
a
type
of
single
cell.
A
A
You
should
check
this
out.
Another
thing
that
I
wanted
to
talk
about
was
this
article,
so
we've
talked
about
cellular
automata
and
we've
talked
about
like
how
they
we
might
be
able
to
use
them
to
find.
You
know,
discover
pattern:
how
pattern
formation
works
and
things
like
that
in
this
article.
This
is
an
interesting
article
I
found
on
the
media.
A
I
think
it's
yeah,
it's
just
a
medium
article,
but
it's
on
making
stem
cells
using
ai,
so
it's
kind
of
an
interesting
foray
into
sort
of
cellular
regeneration
and
then
building
computational
models
of
things
like
regeneration
and
then
kind
of
going
through.
So
you
know
like
they
show
a
lot
of
examples
of
regeneration
and
say,
like
amphibians
and
salamanders,
where
the
tail
gets
cut
off
and
it
regenerates
a
tail-
and
you
know
you're
trying
to
boil
this
down
to
like
a
set
of
principles,
and
so
this
is
something
called
autonomy
of
cells.
A
A
And
then
it
gets
into
cellular
automaton
and
talking
about
like
how
cellular
automata
can
regenerate.
So
you
can
change
from
a
seven
to
a
two
given
some
differential
activity
in
the
on
the
grid
of
the
cellular
automata,
then
it
kind
of
goes
through.
So
this
is
like
a
tutorial,
but
it
kind
of
brings
up
some
really
interesting
issues
talks
about
neural
cellular
automata.
So
you
have
this
combination
of
the
cellular,
automata
grid
and
neural
networks,
and
so
this
is
a
good
tutorial.
A
You
know
I
recommend
like
if
you're
interested
in
this
area
to
read
through
this
discusses
a
lot
of
the
issues,
and
I
don't
know
if
it
links
to
a
lot
of
these
distilled
articles
that
we've
been
following,
but.
A
Conscious
and
eminence
digits,
which
is
kind
of
interesting,
the
uses
were
conscious,
but
I'm
not
a
big
fan
of
like
using
the
word
consciousness
when
you
mean
like
something:
that's
cognitive,
but
anyways
yeah.
So
this
kind
of
goes
through
a
lot
of
this.
Why
is
this
useful?
So
this
is
like
self-organizing,
mnist
digits
and
then
here's
some
resources
which
are
interesting.
So
here's
a
link
to
that
and
let's
see
we
have
in
the
chat
here
dick
says,
but
what
deep
knowledge
does
a
flat
worm
learn?
A
Well,
they
don't
really
learn
any
deep
knowledge.
I
think
they
know
like
like
basic
like
associative
memory
and
things
like
that.
It's
not,
but
you
can
tell
because
you,
the
they
learn
associations.
I
I'm
not
really
sure
I
I
I
haven't
read
the
paper
in
years,
so
the
electrical
signal
is
often
due
to
calcium
channels.
On
the
surface
of
cells,
susan.
N
A
In
the
cells
dick
says,
lionel
jaffe
found
huge
currents
in
chick
neural
plate
which
may
correlate
with
differentiation
wave
for
the
neural
plate
unexplored
area.
Susan
salamander
tales
do
not
always
grow
back
exactly
the
same
way
when
they
regenerate
that's
true
and
there's.
A
lot
of
there
are
a
lot
of
environmental
cues
and
we
talked
about
environment
a
couple
weeks
ago.
What
does
it
mean,
and
one
of
the
things
I
guess
it
means
is
that
it's
an
organizing
signal,
that's
still
pretty.
F
A
But
I
think
that's
that's
something
to
talk
about
too
in
more
detail,
because
that's
true,
they
don't
grow
back
the
same
way
they
just
make
do
with
what
they
have.
You
know
the
cues
that
they
have
and
then
krishna
says
my
problems
are
like
flatworm.
I
try
to
destroy
them.
They
multiply.
A
A
One
big
update
on
this
is
that
networks
2021
this
abstract
has
been
accepted,
so
this
is
going
to
be
a
presentation
at
networks
2021.
This
is
on
embryo
networks
plus
connectomes,
and
we
can
talk
more
about
this
later,
as
this
is
being
developed
if
you're
interested
in
collaborating
on
this
network's
abstract-
and
this
is,
I
think,
the
abstract
link
here.
Let
me
know,
and
we
can
work
on
the
presentation
you
can
get.
You
know
mentioned
in
the
presentation
so
yeah,
so
I
think
that's
it
for
that.
A
I
just
want
to
mention
that
and
then,
if
we
go
back
to
the
papers,
I
wanted
to
talk
a
little
bit
about
a
couple
more
p,
one
one
or
two
more
papers
before
we
go
today.
This
is
thematically
correct
here.
So
the
first
one
is
this:
learning
in
synaptic
plasticity
in
3d,
bioengineer
neural
tissues-
and
this
is
by
a
number
of
people,
including
again,
michael
levin,
and
so
the
abstract
on
this
paper
reads.
A
The
neuroscientists
have
historically
relied
upon
measurements
of
established
nervous
systems.
Contemporary
advances
in
bioengineering
have
made
it
possible
to
design
and
build
artificial
neural
tissues
with
which
to
investigate
normative
and
disease
states.
A
They
don't
have
to
be
bona
fide
action
potentials,
but
they're
things
that
show
up
and
you
can
record
and
that
the
regular
electrical
responses
of
the
cells,
a
return
of
the
evoked
response
following
rest
indicates:
learning
was
transient
and
partially
reversible,
applied
pattern
current
as
masked
or
distributed
pulse
trains
induced
differential
expression
of
immediate
early
genes.
A
So
what
they
did
was
they
took
current
and
they
delivered
it
in
different
patterns,
so
they
either
deliver
it
in
a
bunch
of
trials
almost
simultaneously
or
they
distribute
it
over
time
as
a
series
of
pulses,
like
maybe
once
every
two
seconds
and
then
they're
able
to
do
this
and
look
at
the
effects
and
in
both
cases
the
induced
differential
expression
of
immediate
early
genes,
which
actually
are
like
a
early
marker
of
synaptic
plasticity.
A
They're
genes
that
respond
to
stress
it's
like
the
first
set
of
genes
that
respond
to
stress
and
people
have
studied
immediate
early
genes
in
a
lot
of
different
areas
of
gene
expression,
but
also
it's
a
signature,
plasticity
and
memory
formation,
and
so
this
is
the
first
demonstration
of
learning
learning
response
in
a
bioengineered
neural
tissue
in
vitro.
So
this
is
not
the
again.
This
is
not
the
flatworm
context,
but
this
is
actually
something
much
less
organized.
A
A
Last
night
we
had
a
discussion
about
the
the
neural
link,
brain
machine
interface,
the
one
where
they're
just
delivering
information
to
the
organism
via
like
a
remote
controller
like
a
wireless
setup
and
then
the
organism
is
behaving,
and
so
we
had
a
discussion
about
the
role
of
maybe
something
like
embodiment
or
the
role
of
like
what
the
st
the
in
stimulus
input
should
take.
You
know,
is
it
just
okay
to
deliver
like,
or
is
it
just
enough
to
deliver
electrical
potentials
or
some
sort
of
signal
to
the
cells?
A
Or
do
you
need
the
entire
body,
whether
it
be
a
flatworm
body
or
a
human
body
or
a
primate
body
to
to
facilitate
the
proper
response
to
that?
Well,
in
this
case,
they're
not
really
looking
at
behavioral
outputs
they're
looking
at
electrical
outputs,
so
I
just
brought
that
up
as
an
aside
to
tell
you
that
this
is
a
very
simple
and
contrived
model
system,
so
they
kind
of
walk
through
this
protocol.
They
walk
through
how
they
record
the
electrical
activity.
A
Then
they
they
show
this
example
of
how
they
build
this
culture.
This
bioengineered
cortical
tissue.
It's
a
ring
and
you
get
these
different
silk
fibers
that
are
served
as
a
scaffolding
in
a
hydrogel
and
then
these
you
know
they
have
neurons
that
sort
of
build
connections
along
those
scaffolds
and
then
they're
able
to
apply
a
current
and
actually
do
these
experiments.
A
So
again,
this
is
a
very
contrived
model
system
and
they
walk
through
a
lot
of
the
results.
They
show
the
mast
and
distributed
patterns
of
stimulus
input.
So
this
is
where
you
just
deliver.
60
pulses
all
kind
of
in
a
rapid
sequence,
and
in
this
case
you
have
a
distributed
pattern
where
you
deliver
20
pulses
and
then
wait
a
minute
and
then
20
pulses.
You
wait
a
minute,
20
more
pulses.
So
you
have
this.
A
With
that
memory
of
these
pulses-
and
you
know,
there's
some-
maybe
some
anticipatory
response,
whereas
in
a
masked
pattern,
delivery
of
the
stimulus
there's
no
chance
for
the
system
to
build
any
sort
of
anticipatory
response,
and
so
it's
just
kind
of
being
hit
constantly
with
this
with
this
pulsed
electrical
activity.
So.
A
Expect
to
see
differences
in
maybe
some
sort
of
non-associative
learning,
so
this
is
a
this.
Is
these
are
the
references,
so
this
is
in
this
folder
that
I
shared
with
you
with
you
earlier
and
then.
Finally,
I'm
going
to
talk
about
this
paper
here
again.
This
is
just
a
thematic
thing.
In
my
part,
this
is
where
they
actually
these.
They
use
these
sea
slugs
called
the
plagia.
This
is
a
model
organism.
A
They're
doing
it
in
a
dish
as
well
here
so
they're
taking
yeah
well
they're
doing
it
in
a
in
a
c
slug
and
they're
doing
it
in
culture
and
so
they're
able
to
show
that
the
rna
is
responsible
for
transferring
the
memory
instead
of
something
like
learning
and
or
like
yeah
experiential
learning,
where
you
know
their
they're,
interacting
with
their
environment.
So
the
significance
statement
in
this
paper.
They
have
a
significant
statement.
A
It
is
generally
accepted
that
long-term
memory
is
encoded
as
alterations
in
synaptic
strength,
and
this
is
what
we
use
in
our
in
our
neural
networks
as
a
as
sort
of
a
rule
of
faith.
An
article
of
faith.
An
alternative
model,
however,
proposes
that
long-term
memory
is
encoded
by
epigenetic
changes,
so
what
they
call
non-coding
rnas
can
mediate.
Epigenetic
modifications,
therefore,
rna
from
a
trained
animal
might
be
capable
of
producing
learning-like
behavioral
change
in
an
untrained
animal,
and
so
rna
is
just
basically
these
molecules
of
expressed
genes.
A
They
come
from
genes
that
are
expressed
and
they're
turned
into
rna,
and
then
you
have
this
concentration
of
rna
in
a
in
some
sort
of
buffer
that
they,
you
know,
can
transfer
to
another
organism's
cells.
So
it's
it's
a
very
you
know
it's.
You
know
they're.
Basically,
this
is
the
thing
that's
encoding.
You
know
this
is
a
thing
that
they're
looking
at
instead
of
looking
at
synaptic
strength.
This
is
the
influence
and
so
so
they're
saying
that
this
may
produce
learning
like
behavioral
change
in
an
untrained
animal.
A
So
here
it
is
demonstrated
that
the
memory
for
long-term
sensitization,
which
is
the
model
that
they
use
in
in
applause,
the
smallisk
and
it's
just
like
a
a
it's
like,
I
think,
a
gill
sensitization,
where
they
learn
different
patterns
of
of
opening
and
closing
of
the
of
the
valve
or
the
you
know.
So
it's
not
like
this.
A
This
huge
set
of
thoughts-
or
this
consciousness-
it's
just
this
very
simple,
behavioral
adaptation
that
they
have,
and
so,
but
nevertheless
it
can
be
successfully
transferred
by
injecting
rna
from
sensitized
into
naive
animals.
Moreover,
a
specific
cellular
alteration
underlies
sensitization
in
applausia
sensory
neuron
hyperexcitability-
can
be
reproduced
by
exposing
sensory
neurons
in
vitro,
so
that
was
that
dish
that
they
showed
in
the
figure
to
rna
from
trained
animals.
The
results
provide
support
for
a
non-synaptic,
epigenetic
model
of
memory,
storage
and
appligia.
So
this
again
is
a
model
organism
for
learning
and
memory.
A
Aplasia
californica,
which
is
a
you
know,
it's
a
common
model
organism
they
have.
You
know
they
basically
take
they
train
this
trained.
You
know
they
train
one
of
them
in
this
habituate
it
basically,
so
they
expose
it
to
some
stimulus
and
they,
you
know,
observe
this
habituation
and
then
they're
able
to
take.
You
know
they
test
the
control
animal
first
and
they
test
the
train
animal.
A
Then
they
make
the
transfer
of
rna
to
the
to
the
controller,
the
naive
animal
and
it
exhibits
the
same
behavior
without
actually
experiencing
this
training
firsthand,
and
so
they
can
become
habituated
without
the
experience.
But
then
you
can
also
do
this
in
a
culture
with
just
with
neurons
that
are
disembodied
and
you
get
a
similar
result.
So
I
think
this
is
a
really
interesting
paper.
A
Again,
they
do
a
lot
of
electrophysiology,
which
you
know
it's
that's
going
to
be
a
barrier
to
understanding
exactly
what
they
did,
but
I
think
you
can
get
the
idea,
and
so
they
show
some
results
here
and
the
references.
So
I
would
that's
going
to
be
in
that
folder
as
well.
A
So
now
I
have
a
bunch
of
things
in
the
chat
here,
but
we'll
go
through
them
before
we
wrap
up
so
dick
has
the
jaffy
and
stern
citation
there.
Strong
electrical
currents
leave
the
primitive
streak
of
chick
embryos.
So
that's
the
science
paper
from
1979
verdict
says
I
get
a
really
annoying
network
issue
today.
I
need
to
leave.
Thank
you
for
attending
verdict
and
thank
you
for
your
presentation
catch
up
on
youtube.
A
Thank
you
see
you
next
week,
josh
also
had
to
leave
and
thank
you
for
your
presentations
for
dick
and
debo
jesse
says
yeah.
I
think
there's
a
point
to
be
made
somewhere
about
the
sort
of
embodied
radical
body.
Cognition
stuff
is
something
being
used
as
an
analog
or
a
transference,
or
is
it
doing
anything
beyond
that?
Moving
with
the
cursor
with
your
mind,
just
being
remapped
from
other
movements,
which
is
you
know
it's
an
interesting
question,
I
I
just
showed
you
two
papers
where
they're
sort
of
addressing
the
idea
of
like
you
know.
A
How
can
you
you
know
you
have
this
thing
called
learning
and
memory,
which
is
actually
something
that
maybe
is
a
broader
class
of
things,
behaviors
and
cells.
You
know
maybe
electrophysiological
behaviors,
and
so
we
don't
really.
You
know
we
don't.
We
assume,
we
know
what
learning
and
memory
is
it's.
You
know,
and
then
we
say:
well,
it's
learning
in
memory.
Well,
I'm
showing
you
an
example
from
cell
culture
where
they're
doing
basic
sort
of
a
so
maybe
some
non-associative
learning
things,
and
maybe
from
this
from
this
marine
organism,
where
they're
doing
this
habituation
behavior.
A
Okay,
thank
you
thurin
for
attending,
but
then
you
know,
that's
not
really.
The
same
thing
is
like
maybe
learning
how
to
control
something
with
your
mind.
You
know
in
the
the
neural
link
work
where
they
have
this.
I
think
it's
a
primate
or
could
be
a
rat.
You
know
it
to
control
something
with
your
mind
at
that.
You
know
and
say
a
mammalian
brain,
and
so
that's
quite
a
different
thing,
but
nevertheless
I
think
we
can
learn
things
about
that.
It's
really
interesting
how
you
know
we
kind
of
think.
A
We
know
what
memory
is
and
we
know
the
role
of
experience,
but
we
do.
We
really
know
that.
So
again,
that's
something
yeah
dick
says
you
are
who
you
eat
and
which
is
interesting
because
I
think
they've
done
other
experiments
where
they've
looked
at
like
different
things
that
or
smaller
organisms
have
consumed
in
the
environment.
In
fact,
I
think
it
was
flatworms
where
they
consume
things
in
the
environment.
A
They
can
actually
learn
from
that
because
they're
consuming,
I
guess
you
know
the
rnas
of
other
things,
and
I
don't
really
know,
but
there
there
is
that
I
I
can't
remember
what
paper
that
was,
but
I
remember
reading
a
paper
about
that.
One
time
and
again
this
isn't
like
really
complex
learning.
This
is
like
a
very
simple
form
of
learning,
but
nevertheless
you
know
worm
runners
digest.
Okay,.
A
Well,
thank
you,
yeah.
That
was
a
good
one.
I
have
to
look
that
up.
Actually,
I
don't
really
remember
that,
but
yeah
I
mean
I
wasn't
alive
then
so
I
don't
know,
but
I
mean
so
yeah.
Thank
you
for
attending.
Anyone
else
have
anything
they
want
to
add
before
we
go.
A
All
right:
well,
thanks
everyone
for
attending
and
next
week,
we'll
see
if
anyone
has
anything
present
next
week,
it'd
be
welcome
to
present.
I
think
it
went
the
way
of
the
journal.
Reproducible
results.
Oh
yeah.