►
From YouTube: DevoWorm (2021, Meeting 22): GSoC Update #2, Developmental Networks, Bio Models, NKS, and BNN-ANNs.
Description
Update on Summer of Code activities (Mainak Deb, Improving DevoLearn), review of Networks 2021 submissions, Katatrepsis of Insect Embryos. Papers on the natural history of model organism development (Zebra Finch), 15 years after NKS, simulating primate visual cortex using CNNs, and compound imaging techniques. Attendees: Akshay Nair, Tom Portegys, Jesse Parent, Richard Gordon, Bradly Alicea, Ujjwal Singh, Assaf Wodeslavsky, Shruti Raj Vansh Singh, and Mainak Deb.
A
Good
all
right
well
welcome
to
the
meeting.
I
don't
know
who
else
is
gonna
show
up.
I
I
have
a
couple
things
this
year,
we're
gonna
do
a
gsoc
update.
So
what
do
we
yeah?
What
does
anyone
have?
Oh
tom?
How
are
you
so?
Does
anyone
have
anything
you'd
like
to
present
before
we
get
started?
I
mean,
except
for
my
knock
who's
gonna,
give
a
update.
B
A
C
B
D
B
B
D
C
D
B
B
D
B
B
And
I'll
be
working
on
this
later
on,
but
this
is
where
it
stands
right
now
and
apart
from
all
this
yeah,
this
is
another
thing,
I'll
open
the
image
to
this
okay,
so
this
is,
I
was
actually
trying
out
a
library
called
radio
which
is
basically
like
it's
it
like.
It
makes
it
very
easy
to
create
a
gui
for
an
existing
python
model.
B
E
A
Did
you
did
you?
You
got
the
link
that
I
sent
you
to
the
open
devo
cell,
which
was
from
two
years
ago,
and
we
had
a
student
vinay
vargas
who
put
he
had
he'd
done
some
work,
not
the
same
project,
but
he
put
it
up
on
a
you
know.
I
think
I
can't
remember
where
he's
hosting
it
heroku
app,
I
think-
and
so
that's
that's
yeah,
but
that's
not
ideal
for
a
number
of
reasons.
So
maybe
we
could
take
that
and
and
integrate
a
lot
of
that
into
a
new.
A
I
know
that
oswald
did
some
integration
work
last
summer
as
well,
but
I
think
that
should
all
be
like
sort
of
part
of
it.
You
know
part
of
sort
of
integration,
but
that's
that's
something
we
can
do.
I
think,
maybe
later
in
the
summer
as
we
get
as
we
get
done
with
some
of
the
machine
learning
aspects,
you
know
so
yeah.
A
A
A
A
Okay,
yeah
so
go
ahead
and
share
the
link
for
that
and
so
yeah.
If
there
you
know,
if
there
there's,
no
one
else
wants
to
say
anything.
We
can
go
to
the
next
step
here
or
the
next
thing
and
on
the
agenda,
which
is
to
that
we
have
the
network
networks
2021
conference
coming
up
so
there's
some
move,
there's
some
movement
on
that
front.
Let
me
share
my
screen.
A
A
A
For
life,
so
these
two
videos
are
to
be
submitted
as
talks
for
the
or
they've
already
been
accepted.
They
need
to
be
uploaded
as
talks
for
another.
So
networks
is
this
is
the
main
track
here.
This
talk
it's
about
15
minutes
and
then
this
euler
cycles
is
for
the
topo
nets
workshop
and
that's
about
10
minutes
long
and
so
that'll
be
going
live
in
the
next
couple
weeks,
just
so
that
people
are
looking
forward
to
that.
A
So
this
is
where
it
is
so
far
we're
putting
in
some
images
and
text
from
the
abstract
and
the
references,
and
so
it's
kind
of
being
structured
right
now,
not
really
sure.
Where
we're
I
mean,
we
need
to
go
back
to
the
so
there's
this
paper,
the
a
ns
bnn's
paper
that
I
worked
on
with
krishna
and
jesse,
and
I'm
going
to
add
some
of
that
material
in
here
as
well.
A
So
again,
this
is
the
neuromorphogenetic
patterns
and
the
theory
of
deep
learning.
This
is
sort
of
like
it's
sort
of
like
it's
sort
of
like
a
comparison
between
artificial
and
biological
neural
networks,
but
there's
also
this
aspect
of
like
the
brain
and
modeling
the
brain
as
well,
and
so
you
can
see
that
you
know
when
you
zoom
in
the
text
is
a
lot
bigger.
So
the
text
I
think,
is
about
the
right
size,
but
you
know
we
need
to
put
in
like
some
images
in
each
of
these
boxes,
so
this
network
depth
box.
A
A
A
A
We
have
this
mathematics,
a
diatoms
chapter
which
is
still
going
it's
very
slow,
but
it's
going
and
I
think
I
sent
links-
or
I
think
I
said
copies
to
dick
and
jesse
and
dick
sent
me
back
his
comments.
So
I'm
going
to
be
addressing
that
and
going
through
the
document
and
beefing
it
up
a
little
bit,
making
it
a
little
bit
more.
You
know
focused
towards
certain
things
that
we
can
do
so
it'll
be
good
and
jesse.
A
You
know,
I
know
you've
been
very
busy,
but
you
know
whenever
I
mean
it's
not
like
something
that's
imminent.
So
whenever
you
want
to
give
comments,
that's
fine!
This
open
worm
poster
we
finished
that
that
was
presented
last
week.
I
showed
everyone
what
we
were
doing
for
that
and
that's
almost
done
so
that'll
be
presented
at
the
international
c
elegans
conference.
A
A
A
A
A
You
know
closer
to
the
conference.
The
conference
is
in
the
fall
to
winter.
I
think
it's
in
december,
but
I
don't
know
when
the
announcements
are
going
to
be
so
we'll
keep
an
eye
on
that.
I
know
I
think
usually
was
trying
to
submit
something
to
under
ips
and
they
didn't.
It
wasn't
ready
to
go
so
you
know
that
might
be
a
venue
one
of
these
workshops.
A
Then
we
have
the
mathematics
of
divorm.
That's
I
don't
know.
I
was
talking
to
fahid
with
an
open
worm
about
this,
and
he
suggested
maybe
would
make
this
as
part
of
worm
book,
and
I
don't
know
how
worm
book
views
computational
work,
but
I
think
this
is
definitely
worth
going
back
and
pursuing.
I
know
I
presented
on
it
several
weeks
ago
when
I
showed
you
sort
of
in
this
poster
format.
A
We'd
obviously
want
to
maybe
write
a
little
bit
more
about
the
each
thing
in
in
that
diagram,
but
this
I
think
this
is
worth
pursuing
and
you're,
probably
gonna
have
to
wait
until
for
a
little
while
until
things
settle
down
with
you
know
some
of
the
other
things
going
on
with
some
of
the
summer
conferences
to
to
pick
that
up
again
so
but
that'll
be
you
know
if
anyone
wants
to,
you
know
contribute
to
that.
If
they
have
ideas
on
how
to
make
that
better.
That
would
be
great.
A
I
know
I
sent
tama
notification
about
that,
but
that
was
I
don't
know
if
that's
something
that
he
did
or
not,
but
that
anyways
it
grayed
that
one
out,
then
we
have
these
two
tests
of
william
williamson
symbiosis,
which
is
something
that
we
proposed
as
a
sort
of
a
genetic
or
a
bioinformatics
test
of
genetic
data.
So
you
know
this
would
involve
taking
genomes
from
a
couple
of
species
and,
looking
for
you
know,
blast
matches,
or
things
like
that.
So
this
is
still
pretty
rough.
A
We'd
have
to
talk
about
this
offline,
to
kind
of
get
a
sense
of
what
needs
to
be
done
for
that,
but
and
then
there
we
have
these
two
papers
on
diatom
movement,
the
molecular
level
simulations
and
the
diatom
movement,
smoother
jerky,
which
is
essentially
the
sort
of
the
higher
moments
of
movement.
So
those
are
those
are
related,
24
and
26.
Although
they're
separate
projects,
then
so
this
is
the
topo
nets
workshop
here
and
this
dynamics
days.
A
A
A
We
might
also
pick
up
on
some
of
these
themes
here,
like
the
boring
billion
again
or
you
know,
maybe
like
an
origin
of
embryos
type
thing,
and
then
we
have
other
things
that
are
kind
of
being
worked
on
over
time.
So
we'll
see
how
that
goes,
and
then
we
have
the
c
elegans
conference
and
the
networks
conference
to
report
on
to
see
what
people
say
about
the
submissions,
so
that'll
be
good
all
right.
So
that's
our
submissions.
A
So
again,
this
is
our
group
meetings
spot
on
github.
So
we
have
a
number
of
different
directories
here
which
have
different
things
that
are,
they
haven't
really
been
maintained
very
well
internally,
but
they're
little
stubs.
Just
to
you
know
if
there's
some
reference,
so
we
want
to
stick
in
there.
We
have
a
lot
of
things
like
this.
Is
kind
of
old
this
periodicity
folder,
maybe
that
should
be
archived
out
early
life
in
the
origins
of
development
just
stuff.
A
That's
come
up
in
the
meetings
that
we're
trying
to
you
know
it's
either
maybe
passed
or
maybe
can
be
followed
up
on.
Then
we
have
these
open
papers
again.
So
we
have
the
developmental
modeling
papers
which
this
one
here,
for
example,
is
that
was
a
abstract
that
was
submitted
and
now
it's
being
presented
on,
and
then
you
know
after
that,
maybe
we
want
to
write
a
paper
on
it,
so
that
would
be
a
good
opportunity.
A
People
want
to
pick
up
on
that
and
you
know
flesh
out
of
paper.
We
have
this
theoretical
types
of
embryo
developmental
networks-
that's
kind
of
related
to
this.
Fourth
one,
but
that's
you
know
these
things
kind
of
merge
and
split
apart
as
the
as
they
get
developed,
so
game
theory
for
developmental
processes.
This
is
abstract,
that's
been
submitted
and
there's
a.
I
think
this
doc
actually
has
more
to
it
like
there's
more
of
a
fleshed
out
outline,
but
it
still
needs
a
lot
of
work.
A
This
is,
of
course,
the
abstract
that
we're
submitting
as
a
poster
to
the
net
neuro
satellite,
so
this,
of
course,
can
be
made
into
a
paper
and
we
have
the
a
ns
and
bnn's
paper
that
kind
of
goes
along
with
this,
so
this
was
actually
put
here
previous
to
that
paper
being
written.
So
this
needs
to
be
updated.
A
Then
we
have
some
things
on
behavioral
emergence.
This
networks
thing
is
a
duplicate
of
this
evil
learn
is
this
is
something
that's
already
been
said
or
already
been
presented
on.
So
that's,
there's.
Well,
there's
supposed
to
be
a
diva
learned
paper
that
we
haven't
really
worked
on
and-
and
you
know,
that's
something
that
we
can
work
on
later.
A
The
end
of
the
gsoc
period,
we'll
we'll
have
more
to
put
into
that
paper,
but
that's
still
ongoing,
and
then
this
basilarian
non-neuronal
cognition
work,
which
is
still
kind
of
in
a
state
of
being
updated
and
then
some
other
things
under
here
as
well.
So
this
this
this
needs
to
be
updated
here,
but
the
open
papers
are
still
very
much
open
and
I
know
people
have
ideas
for
open
papers,
but
basically
the
idea
of
an
open
paper
is
we
have
this.
A
We
come
up
with
a
title.
We
come
up
with
a
document
we
put
the
link
out
and
then
people
can
either
issue
a
pull
request
to
contribute
some
text
or
some
references,
or
they
have
a
google
doc
that
they
can
make
comments
on,
and
so
then
they
can
contribute
in
that
way
and
then
we
can
have
like
collective
action
on
some
of
these
papers.
A
So
this
is
our
task
board.
Here
we
have
to
do
old
action
items
in
progress
finished
and
then
things
that
are
off
the
radar.
So
this
off
the
radar.
Of
course,
we
have
a
lot
of
things
that
are
sort
of
were
on
the
board
and
then
got
forgotten
about
a
number
of
different
paper
reviews
and
lectures
and
papers
repositories
that
you
know
never
really
made
it
into
the
stream
of
our
work.
A
A
A
Something
that
I
don't
know
if
we'll
do
in
the
same
way
we
would
like.
So
this
networks,
2021
conference,
is
coming
up
at
the
beginning
of
this
next
month,
so
the
satellites
are
kind
of
at
the
end
of
this
month.
The
actual
conference
is
maybe,
like
the
you
know,
between
the
1st
and
10th
of
july,
since
it's
online,
it's
a
little
bit
flexible,
but
I
was
thinking
about
having
some
sort
of
group
presence
where
we
have.
A
A
Some
sort
of
place
where
we
can-
I
mean
they're,
going
to
be
a
lot
of
presentations
here.
There
are
actually
three
presentations
that
have
you
know
the
diva
warm
logo
on
it,
so
people
will
be
become
aware
of
it,
but
I
mean
you
know
other
than
that
we
can
do
something
like
a
social
media
campaign
around
it
or
a
series
of
blog
short
blog
posts
or
something
to
sort
of
draw
people
into
the
group
or
make
them
aware
of
the
group.
So
that's
something!
A
A
A
I
think
that
we're
in
a
good
place
on
that-
and
this
isn't
just
this
evil
and
software,
but
this
is
the
diva
learning
platform
so
now
that
there
have
been
a
couple
presentations
on
it
and
there
have
been
a
couple
of
gsoc
projects
devoted
to
it.
I
think
we
can
write
a
pretty
good
paper,
there's
this
axolotl
montaging
again.
That
was
something
from
the
g-suck
projects,
but
we
haven't
made
any
progress
on
that.
Yet
you
know
we'll
be
talking
about
that.
A
I
guess
in
future
meetings
the
complexity
measures
are
falling
off
the
radar.
I
think,
but
I
think
that's
something
that
we
can
still
talk
about-
that
actually,
the
mathematics
of
evil
worm
is
part
of
that,
but
we
can,
I
think,
that's
actually
a
good
sort
of
way
forward
on
this,
because
we're
kind
of
getting
these
complexity,
measures
and-
and
you
know,
kind
of
assessing-
where
we
are
with
that
and
then
the
mathematics
of
diva
worm.
A
That's
89,
that's
linked
to
three,
I
think
pretty
closely,
so
those
probably
will
move
together
and
then
hold.
We
have
a
bunch
of
things
with
axolotl
axolotl,
embryo
animations
and
segmentation
things
like
the
bibliography
for
the
gigapixel
technique.
So
I
think
dick
has
introduced
a
number
of
things
that
he's
been
interested
in
or
he's
offered
sort
of
people.
You
know
projects
that
people
can
work
on.
A
So
there
are
a
bunch
of
these
in
this
list
here
with
this
lagrangian
embryo
pattern,
resumption
after
self-repairing
seashells
all
these
different
issues
that
are
related
to
things
that
he
kind
of
has
proposed
in
previous
meetings.
So
I'm
not
sure
exactly
what
week
these
were
mentioned
in,
but
I
think,
if
you
go
through
the
description
of
the
different
videos,
you
can
find
it.
A
This
work
on
morphogenesis
and
deep
learning
paper.
I
think
this
is
actually
something
that
goes
with
this
net
neuro
poster.
So
that's
those
will
probably
move
in
some
relation
to
one
another.
We
have
some
to
do's.
We
have
the
boring
billion
boring
billions
paper.
We
have
build
a
collection
of
work
for
the
kindle
book.
A
A
A
A
A
A
It's
called
catatrypsis,
it
happens
in
some
insect
species
and
it
is
a
rather
mysterious
process
overall,
a
cheated
domestic
assembrio,
which
is
the
species
genus
and
species
captured
on
a
zeiss
micro
v16.
While
I
was
assistant
teaching
at
embryo
21,
this
is
from
a
summer
school
held
at
the
one
of
the
national
labs,
so
this
is
something
that
was
captured
during
this
period,
the
soap
this
summer
school
period,
and
so
this
is
what
he's
referring
to.
A
So
this
is
actually
something
that
was
not
something
you
just
observed.
This
was
initially
documented
in
orthoterra
by
eleanor
silfer
in
1932,
and
this
is
the
citation
here
for
those
curious
about
such
embryonic
movements
within
the
egg.
The
authoritative
review
on
the
subject
is
this
link
here
by
kristen
filio,
and
so
I
think
I
have
the
paper
here.
A
This
is
not
surprising
as
the
model
system
drosophila
melanogaster
has
an
extremely
reduced
extra
embryonic
component.
The
amniocerosa,
in
contrast,
most
insects,
retain
the
ancestral
complement
of
two
distinct
extreme
branch,
membranes,
amnion
and
serosa.
So
this
is
citrosofall
is,
of
course,
we're
familiar
with
that
model
organism,
but
they
don't
really
even
even
as
an
insect.
They
don't
really
exhibit
this
behavior,
and
so
in
a
lot
of
insects.
They
have
this
ancestral
set
of
membranes
that
are,
you
know,
from
some
common
ancestor
and
the
amnion
and
serosa.
A
A
A
So
this
is
something
that
I
don't
know
why
exactly,
but
this
the
terminology
and
awareness
of
this
these
systems
are
not
currently
really
being
studied.
There
isn't
like
been
a
lot
of
development
in
the
area,
as
a
number
of
recent
studies
have
identified
essential
components
of
these
events,
your
rna
interference
of
developmental
genes.
A
You
know
what
the
gene
is
expressing
and
just
simply
knock
it
down,
so
that's
no
longer
being
expressed.
So
then
you
can
look
at
what
happens
when
those
genes
are
shut
off,
and
so
this
is
what
they're
doing
here
as
well
as
ectopic
hormonal
treatments,
so
in
insect
development,
it's
very
heavily
dependent
on
hormonal
shifts
and
other
things.
So,
there's
a
lot
of
things
you
can
do
with
hormones
to
also
make
modifications
to
the
embryo
to
see.
If
what
happens,
when
you
do,
you
know
x
or
y
and
manipulation
as
much
remains
unknown.
A
But
anyways
yeah-
I
guess
this
is
the
the
orange
here
or
the
yellow
is
the
yolk.
The
embryo
is
the
gray,
and
so
oh
the
cirrhosis
on
the
outside.
I
see
that's
the
blue
and
then
the
amnion
is
the
orange,
which
is
this
sort
of
like
this
outline
here
on
the
inside,
and
so
this
is.
These
are
different,
anatomical
views.
So
this
is
the
mid
sagittal
section
view
in
a
and
then
b
is
the
transverse
section
view
at
the
position
indicated
by
these
at
this
slice
right
here.
A
So
that's
a
that's
the
anatomy
of
it
and
then
this
is
the
phylogeny.
So
we've
talked
about
phytogenesis
as
being
these
evolutionary
relationships,
so
you
can
see
where
they
fit
into
the.
These
are
the
holo
metabolis
insects
here,
and
these
are
the
different
insects
as
well
that
that
expressed
this
trait.
A
So
here
are
a
nice
set
of
images
here
where
they
show
the
serosa
and
the
amnion
in
this
flipping
process
of
the
cirrhosis
out
here
the
amnions
here
and
then
you
have
this
sort
of
process
where
there's
this
folding
and
you
saw
in
the
video
where
it
looks
like
these
little
fingers
are
coming
out
and
that's
kind
of
what's
happening
here.
You
kind
of
see
this
is
the
cartoon
on
the
left.
A
So
you
can
see
that
process
of
folding,
where
there's
a
where
it
kind
of
comes
out
and
then
goes
underneath
and
then
there's
this
turning
of
the
egg
and
the
cirrhosis
reduced
in
this
and
the
amnion
is
expanded
and
the
embryo
is
out
here.
So
that's
basically
what's
happening.
This
is
the
whole
process
of
blastokinesis.
A
Okay,
so
that's
just
an
image
that
we
just
saw
and
then
my
favorite
hand-drawn
illustration
of
catatrepsis
was
done
by
kozo
miya
kawa
in
1987,
who
surveyed
a
bunch
of
different
odin
odonates,
finding
intriguing
diversity
in
the
dynamics
of
reorienting
embryos.
So
this
is
a
paper
on
the
topic
here,
and
so
these
are
the
drawings.
You
can
see
the
hand.
Drawings
are
very
there's
a
very
high
art
of
hand,
drawing
in
developmental
biology,
as
in
some
you
know.
Also,
you
know
different
parts
of
biology.
A
They've
been
doing
this
sort
of
thing,
for
you
know
over
100
years,
where
they
draw
things
out
by
hand,
and
it's
very
useful
in
some
cases,
to
really
understand
what's
going
on.
But
this
is
the
process
here,
this
reorienting
of
embryos-
and
so
this
is
something
that
this
is
the
picture
of
it
actually
close
up.
A
And
then
this
is
the
paper
that's
based
off
of
so
this
is
positioned
in
germ
rudiment
and
rotation
of
embryo
and
eggs
and
of
some
dragonflies.
So
odonata
is
a
dragonfly
genus.
I
believe-
and
this
is
the
german
rudiment
and
rotation
of
embryos
in
the
egg
and
so
see
if
there
are
any
pictures
in
here
that
we
might
look
at
so
this
is
the
same
thing
we
saw.
This
is
a
revolution
with
a
180
degree
rotation
of
an
embryo.
A
A
So
this
these
arrows
are
pointing
in
different
directions.
They
look
like
a
clock,
but
it's
just
telling
you
where
the
ventral
side
of
the
embryo
is
so
you
can
see
the
ventral
side
of
the
embryo
is
shifting
around
in
these
different
specimens.
So
you
have
this
or
it's
down
and
then
it's
up
and
then
it's
sideways
here
and
then
it's
kind
of
ticking
downward
and
then
it
goes
back
to
being
sideways
and
this
this
specimen
here
at
the
top.
A
A
So
so
that's
the
arrows
show
the
ventral
side
of
the
embryo.
The
circle's
enclosing
the
arrow
top
showing
eggshells
and
posterior
view
with
ventral
side
downward
developmental
time
is
indicated
by
days
after
oval
position
and
then
the
let's
see
so
they're
in
these
different
stages.
So
r
is
the
revolution
stage.
A
E
is
the
stage
where
the
eyes
are.
Pigmented
m
is
the
stage
where
muscles
become
functional
and
h
is
hatching.
So
you
can
see
at
the
end
there's
this
h.
So
the
hatching
is
at
different
time
lengths
at
different
time
periods,
and
then
you
also
have
these
different
stages
here,
where
different
things
happen.
So
the
eyes
form
here
the
muscle
forms
here
and
then,
after
between
the
muscle
and
hatching,
you
actually
do
see
some
rotation
even
then,
and
so
that
kind
of
you
know
that's
just
a
nice
little
diagram
of
that.
A
So
that's
that's
all
I
wanted
to
say
on
that.
So
that's
catatrepsis
and
I
thought
that
might
be
a
good
thing
to
look
at.
If
people
are
interested
in
this
topic,
maybe
we
can
follow
up
on
it.
I
don't
know
too
much
about
how
it'd
be
useful
in
c
elegans,
but
it's
interesting
to
know
that
that
process
happens.
A
A
A
A
So
this
is
a
model
organism
that
has
so
the
one
sentence
summary
this
paper,
the
genetic
and
behavioral
diversity
of
zebra
finch,
both
in
the
wild
and
in
captivity,
make
it
well
suited
for
neuroethological
studies
of
vocal
learning,
culture
and
social
bonding.
So
this
is
not
embryogenesis
anymore.
This
is
organismal
development
in
a
bird-
and
this
is
a
couple
people
here
work
in.
A
The
zebra
finch
will
continue
to
inform
research
into
culture,
learning
and
social
bonding,
as
well
as
adaptability
to
a
changing
climate.
So
we
have
this
review
of
this
organism
again
I
get
into
the
natural
history
of
it.
So
this
is
you
know
this
is
a
model
organism
that
people
use
for
different
reasons
and
they're
kind
of
going
over
the
natural
history
of
it,
which
is
something
that
isn't
really
considered.
Often
in
times
we
talk
about
model
organisms
we
usually
just
take.
A
For
example,
if
we're
interested
in
c
elegans,
we
usually
use
either
some
wild
type
or
back
crossed
mutants.
We
don't
really
worry
about
the
natural
variation
or
natural
diversity,
and
so,
in
this
case
they're
talking
about
you
know
they
structure
it
according
to
sort
of
the
evolutionary
history
they
talk
about
where
the
zebra
finch
came
from,
and
then
they
talk
about
analysis
of
sex
differences
and
vocal
learning
and
production,
so
they
consider
how
the
model
organism
is
used
and
they
consider-
maybe
you
know,
is
this
something
that
you
might
find
in
nature.
A
A
And
then
genes
and
brains
for
vocal
learning,
so
there's
a
lot
that
we
can
learn
from
sort
of
the
genetics
and
the
brains
of
zebra
zebra
finches,
not
zebrafish,
zebra
finches.
I
just
wanted
to
make
that
clear
since
we
talked
about
zebra
fish
here
a
lot
these.
This
is
a
lot
of
vocal
learning
studies
that
they've
done
with
this.
A
D
A
A
Maybe
the
second
paper
I'm
going
to
talk
about
is
a
new
kind
of
science,
a
15-year
view,
and
this
is
something
I
found.
I
think
this
weekend
and
I
was
looking
through
some
stuff,
and
this
is
a
nice.
This
book,
a
new
kind
of
science,
came
out
back
in
2002,
I
believe,
and
so
we're
kind
of
at
well,
I
think
2001
technically.
A
What
can
they
tell
us
about
the
world?
So
we
have
all
these
things
that
he's
sort
of
you
know
he
reviews
this
idea
of
using
cellular
automata
to
model
nature,
and
he
operates
from
this
principle
of
what
we
call
computational
equivalence
and
it's
actually
mentioned
in
the
poster
here,
and
so,
if
I
go
back
to
the
poster,
if
we
go
back
to
computational
equivalence
and
the
definition
of
it,
okay,
the
definition
is
systems
found
in
the
natural
world
can
perform,
perform
computations
up
to
a
maximal
universal
level
of
computational
power.
A
What
that
means
is
that,
basically,
if
I
have
a
system,
I
can
find
a
cellular
automata
or
a
set
of
rules
that
are
equivalent
to
that
thing
in
nature.
So
that's
a
pretty
bold
stance,
because
you
know
you
would
think
well.
Models
are
generally
useful,
but
they're,
not
necessarily
something
that
we
can.
You
know,
extract
the
entire
natural
phenomena.
A
Stephen
wolfram
is
kind
of
saying:
well,
no,
there's
an
equivalence
there
and
you
can
use
computational
fines
for
these
universal
rules,
and
so
it's
just
a
matter
of
you
know
getting
the
right
combination
of
the
right,
computational
rules-
and
you
can
do
this
with
something
like
a
cellular.
Automata
so
since
it's
been
15
years
since
I
published
my
book
a
new
kind
of
science,
more
than
25
since
I
started
writing
it,
so
he
spent
like
10
to
20
years
working
on
this,
but
with
every
passing
year.
A
I
feel
I
understand
more
about
what
the
book
is
really
about
and
why
it's
important.
I
wrote
the
book
to
contribute
to
the
progress
of
science,
but,
as
the
years
have
gone
by,
I've
realized
that
the
core
of
what's
in
the
book,
actually
goes
far
beyond
science
into
many
areas.
That
will
be
increasingly
important
in
defining
our
whole
future.
A
But
for
me,
one
of
the
achievements
of
the
book
is
the
realization
that
one
can
explore
such
fundamental
things
concretely
by
doing
actual
experiments
in
the
computational
universe
of
possible
programs.
So
you
may
you
can
build
these
programs
and
they're
not
only
like
sort
of
approximations
of
nature.
They're.
Actually,
you
know
sort
of
they
have
this
equivalence
to
the
natural
system,
and
so
so.
A
In
retrospect,
I
considered
incredibly
lucky
that
all
those
years
ago
I
happen
to
have
the
right
interest,
set
up
the
right
skills
to
actually
try
what
is,
in
a
sense,
the
most
basic
experiment,
computational
universe,
to
systematically
take
a
sequence
of
the
simplest
possible
programs
and
run
them,
and
then
so
then
you
got
all
these
interesting
results
where
you
get
these
patterns
and
then
you
have
to
interpret
the
different
rules.
So
this
is
rule
30
up
here.
This
is
his
all-time
favorite
discovery.
A
It's
one
of
the
simplest
programs.
You
can
imagine
it
operates
on
a
row
of
black
and
white
cells,
starting
from
a
single
black
cell
and
then
repeatedly
applying
the
rules
recursively
and
then
down
to
this
bottom
section
here,
and
the
crucial
point
is
that,
even
though
the
rules
are
by
any
measure,
extremely
simple,
the
pattern
that
emerges
is
not,
and
so
it's
this
generative
system
that's
generating
these
really
complex
things.
A
Much
like
you
would
have
like
in
a
genome,
it's
generating
complex
movements
in
in
the
embryo,
or
it's
generating
these
complex
bird
songs,
and
so
we
can
use
that
to
sort
of
understand.
What's
going
on
there,
it's
it's
a
representation
in
the
sense
that
you
know
maybe
doesn't
include
like
genes
encoded
with
information,
but
if
you
take
the
rules
as
being
these
transformational
elements,
you
can
generate
complexity
from
simplicity
and
that's
kind
of
what
we're
trying
to
do
with
understanding
the
genome.
A
I
can't
ascertain
it,
but
I'm
pretty
sure
that's
the
case,
and
so
this
has
been
a
shift
of
looking
at
the
world.
Some
of
this
is
because
we've
had
better
computers,
but
some
of
it
is
also
just
the
this
has
been
kind
of
demonstrated
a
bit
more
in
you
know.
The
view
of
nature
has
gone
from
this
sort
of
hardcore.
A
A
What
about
this
one
simulating
a
print?
Oh!
This
is
very
short.
This
is
just
an
abstract,
so
we
might
do
two.
So
this
is
something
I
found
it's
a
nurip's
abstract
from
2020.
It's
simulating
a
primary
visual
cortex
on
at
the
front
of
convolutional
neural
networks.
That's
what
cnn's
means
improves
robustness
to
image
perturbations.
A
So
this
means
that
they
model
this
network
after
the
structure
of
e1
and
its
operation
and
which
is
an
area
of
the
human
brain
or
primate
brains
that
analyzes
blobs
and
stripes
of
images.
So
it's
like
a
very
low
level
of
image
processing,
but
that's
basically
what
they're
trying
they
you
know
they've
had
this
idea
that,
basically,
if
you
simulate
that
it
should
improve
the
robustness
of
the
model
and
they
found
that
using
an
adversarial
attack
model,
they
can
do
this.
They
can
overcome
these.
They
can
make
it
more
robust.
A
A
The
v1
block
is
based
on
a
classic
neuroscientific
model
of
v1,
the
linear
non-linear
poisson
model
consisting
of
a
biologically
constrained,
gabor
filter
bank
and
simple
and
complex
non-linearities
in
a
v1
neuronal
stochasticity
generator
after
training.
V1
nets
retain
high
image
net
performance,
but
each
is
substantial
and
more
robust,
performing
the
base
cnn
and
state-of-the-art
methods
by
18
and
3
respectively,
on
these
benchmarks
that
they
use
they
use
adversarial
attacks
and
these
common
image
corruptions.
So
they
have
two
strategies.
A
One
is
to
show
like
adversarial
sort
of
attacks
on
the
image
corrupting
the
image,
and
then
the
other
is
sort
of
a
common
image.
Corruption,
like
different
errors,
that
you
might
find
in
an
image
like
noise,
like
a
noise
mask
or
maybe
something
else
that
you
know
like
some
missing
part
of
the
image
might
be
blurry.
A
Finally,
we
show
that
all
components
of
the
v1
block
work
in
synergy
to
improve
robustness,
and
so,
while
current
cnn
architectures
are
argument,
arguably
brain
inspired,
the
results
presented
here
demonstrate
that
more
precisely
mimicking,
just
one
stage
of
the
primate
visual
system
leads
to
new
gains
in
image
net
level.
Computer
vision
applications.
A
So
this
is
something
that
you
know
we've
talked
about
before
about
you
know:
if
you're
simulating
the
brain,
what
are
you
really
gaining?
Can
you
build
better
artificial
neural
networks,
or
it's
just
just
not
an
exercise
and
futility
or
uselessness
and
they're
suggesting
in
this
abstract
that
it's
actually
improving
the
model's
performance
actually
in
in
quite
a
simple
and
straightforward
way?
And
so
that's
that's
something
that
we
should
probably
mention
in
our
poster
something
similar
to
that
and
then
I
think
I
don't
know
if
I'm
going
to
go
through
this
paper.
A
A
A
So
the
abstract
is
recovering.
3D
phase
features
of
complex
multiple
scattering
biological
samples,
traditionally
sacrifices,
computational
efficiency
and
processing
time
for
physical
model
accuracy
and
reconstruction
quality.
This
trade-off
hinders
the
rapid
analysis
of
living
dynamical
biological
samples
that
are
often
of
great
greatest
interest
to
biological
research.
A
This
sampling
method,
using
a
novel
physics
model,
simulator
based
learning
strategy
trained
entirely
on
natural
image
data
sets.
We
show
our
network
and
robustly
reconstruct
complex
3d
biological
samples,
arbitrary
size
and
structure.
This
approach
highlights
that
large-scale
multi-scattering
models
can
be
leveraged
in
place
of
acquiring
experimental
data
sets
or
achieving
highly
generalizable
deep
learning
models.
A
We
devise
a
new
model
based
data,
normalization,
pre-processing
procedure
for
homogenizing,
the
sample
contrast
and
achieving
uniform
prediction
quality,
regardless
of
scattering
strength
to
achieve
highly
efficient
training
and
prediction.
We
implement
a
lightweight
2d
network
structure
that
utilizes
a
multi-channel
input
or
encoding
axial
information,
so
they
demonstrate
this
framework
on
measuring
epithelial,
glucal
cells
in
in
c
elegans
and
so
they're.
Using
this
approach
to
sort
of
examine
a
certain
type
class
of
cell
and
c
elegans
and
adult
c
elegans,
we
highlight
the
robustness
of
this
approach
by
evaluating
dynamic
samples
of
a
living
worm,
video.
A
We
emphasize
our
approaches:
generalizability
by
recovering
algae
samples,
evaluated
with
different
experimental
setups,
to
assess
the
production
quality.
We
develop
novel
quantitative
evaluation
metrics
and
show
that
our
predictions
are
consistent
with
experimental
measurements
and
multi
multiple
scattering
physics.
A
So
this
is
a
imaging
method
that
they're
using
to
look
at
so
here
you
can
see
in
the
image
you
have
this
microscopy
technique.
You
know
your
microscope,
you're
looking
at
the
intensity
and
the
intensity
spectra
and
then
you're
generating
objects,
you're
simulating
intensities
and
you're,
making
a
linear
estimate,
and
so
then
you're
using
this.
These
data
to
train
the
network
and
then
recover
these
slices,
and
so
this
is
the
linear
approximation
here,
which
is
this
section
b
here.
This
is
the
output
for
the
c
elegans
and
then
there's.
D
A
A
A
It's
using
the
microscopy
data,
it's
sampling
at
different
intensities
and
then
it's
putting
using
a
neural
network
to
put
this
together
and
then
you're,
putting
these
different
output
images
together
to
get
this,
and
so
it's
a
very
nice
image,
so
yeah,
I
don't
want
to
get
in
too
much
into
the
technical
details,
so
I
will
leave
that
to
people
to
look
at.
I
will
put
that
in
the
chat
and
then
I
want
to
go
through
the
chat
because
we
have
like
12
things.
A
If
you
need
permissions,
let
me
know
so,
okay,
so
actually
I
said
I
have
my
final
exam
tomorrow,
so
gotta
go
now
well,
thank
you
for
attending
akshay
dick
says-
and
this
goes
back
to
the
the
first
thing
I
talked
about.
Do
these
acrobatics
correlate
with
whether
early
developmental
development
of
an
insect
is
centital
or
cellular?
A
I
don't
think
so.
I
think
that
it's
just
they're
just
observing
it
in
the
model
organism
that
we
this
catatrepsis,
is
what
he's
talking
about.
I
think
it's
just
in
this.
I
don't
know
if
it
has
anything
to
do
with
sort
of
the
centricial
nature
of
it.
I
think
it's
just
the
stage
of
embryogenesis,
so
I
don't
really
know
so
again.
This
is
the
this
is
the
process
here
where
this
the
embryo
kind
of
unfolds
and
then
come
pops
out
of
this
earlier
structure
and
then
it
kind
of
comes
underneath.
A
So
I'm
not
really
sure
I
think
the
early
part
is
centititial,
maybe
and
then
the
embryo
is
popping
out
of
cellular,
but
I'm
not
really
sure
yeah.
So
that's
I
don't
know
the
answer
to
that.
Necessarily
then.
My
next
is
having
to
leave
now
see
you
all
next
week.
Thank
you,
my
name
for
attending
and
look
forward
to
your
next
update
next
week.
A
A
A
A
A
Yeah
so,
okay!
Well,
if
you
have
anything
you
put
it
in
the
slack
channel
or
email
or
whatever,
we
could
talk
about
it
next
week,
yeah
next
week,
we'll
have
minot
giving
another
update
on
his
activities
and
then
if
people
want
to
present
something-
or
we
just
want
even
an
informal
talk
about
discussion
about
something
we
can
do
that.