►
Description
Attendees: Bradly Alicea, Mayukh Deb, Jesse Parent, Ujjwal Singh, and Krishna Katyal. GSoC Updates and discussion, paper planning for Periodicity of the Embryo.
A
B
B
A
A
B
B
B
D
C
B
B
D
C
A
D
B
B
There
are
actually
videos
of
these
forms
like
these
empires,
okay.
So
when
I
plot
the
populations
of
these
values
at
AE
and
pcz
with
time,
then
these
values
are
actually
those
really
values
for
a
particular
video.
Is
this
like
a
grows
up
really
fast
in
the
blue
one?
You
can
see
the
deep
blue
one
which
grows
up
really
fast
right
and
then
there's
the
other
lineages,
which
are
like
really
really
small
populations,
but
they
also
grow
like
in
a
certain
rate.
B
The
Freddie
like
the
predictive
data,
okay,
so
then
I
was
kind
of
getting
a
very
I.
Don't
know
what
should
I
call
this
later,
let
us
finish
under
the
flat
top
set
on
the
initial
part
from
0
to
50.
On
the
left
side,
you
can
see
that
it's
really
biased
words
like
it's,
not
really
predicting
well
for
a
on
the
initial
things.
They
don't
initial
side
a
then
when
I
was
training
it
on
time,
values
which
were.
B
D
B
B
Yeah,
like
wait,
I'll
show
you
let
it
go
from
suppose
it
goes
from
zero
to
any
one
like
on
this
one
256,
yes,
okay
and
then
on
some
of
the
videos.
It
goes
from
two
fifties
like
equation
chain
like
the
beautiful
physics
and
then
it's
another
video,
so
the
data
suddenly
goes
back
to
zero
and
then
this
is
100.
So
you
get.
B
D
A
B
B
B
B
E
B
B
B
B
C
C
A
A
D
B
B
A
B
C
Said,
there's
UPC
on
the
image
with
an
F
inside
second
image:
yeah,
let's
fansite
second
original.
Second
go
first
image:
ya!
Miss
me
yeah
this
one.
Yes,
then,
you
can
see
this.
This
is
something
which
is
causing
like
Oh
problems
for
Mon
yeah.
As
you
can
see
like
this
is
all
noise
with
even
on
you
really
can't
end.
Like
always
it's
on
there,
you
can
see
legs.
C
You
can,
what
use
it
is
to
make.
You
can
be
definite
between
no
models
once
for
the
length
of
T.
This
might
work
other
than
that
you
can
just
use
a
demos.
A
technique
in
the
whole
model
used
did
such
CV
to
find
the
best
parameter,
which
is
simple.
These
at
least
these
one
reviews,
so
just
to
tell
me,
lies
the
model
so
that
specific.
A
A
D
A
B
C
B
So
you're
telling
me
like,
after
training,
the
inlet
after
training
in
all
of
the
images
we
just
run
it
on
all
of
the
training
again
and
this
thing
worked
on
the
losses
for
each
image
and
then,
if
the
loss
is
higher
than
a
certain
value,
then
that,
like
a
bad
image,
so
we'll
drop.
That
image
will
delete
that
image.
Yeah.
B
C
B
B
It's
like
it's
our
one
instance
and
given
a
few
frames,
I
mean
for
the
next
instance
for
the
next
instance.
There
is
only
one
thing
and
again
on.
The
next
instance
is
true,
then,
again
right,
it's
more
uniform
law
usually
uses
people
to
do
until
who
is
no
more
possible
only
within
the
level
anyone.
It
was
not
good
because
it
creates
offsets,
even
though
small
it
was
there
yeah.
There
are
the
things
that
yeah,
so
that
was
it.
A
A
A
A
D
F
D
A
D
A
A
Awesome
yeah,
thanks
again
for
the
update
on
the
google
Summer
of
Code
stuff
I,
think
we
had
a
good
discussion
on
that.
I,
don't
know
if
I
had
any
questions
right
now,
but
we
can
revisit
some
of
that
on
slack
I.
Try
to
encourage
you
to
do
that,
because
I
think
it's
a
good
place
to
kind
of
clear
up
some
of
the
issues
to
do
that,
and
so,
let's
see
I
had
a
couple
of
things
myself
that
I
wanted
to
present.
A
So
the
first
thing
is
that
Oshawott
asked
me
a
couple
days
ago
on
slack
about
the
periodicity
in
the
embryo
paper,
and
so
this
is
the
paper.
That
is
it's
it's
a
way.
Well,
we
had
prepared
a
proposal
for
it
like
about
a
year
ago,
and
the
special
issue
never
happened.
So
we
have
this
repo
that's
up,
but
we're
not
like
we
haven't
done
it.
So
we're
kind
of
pre
revisiting
this
paper
for
another
journal.
A
A
A
What
we
do
have
is
that
same
type
of
data
that
my
yolk
was
showing,
where
you
have
the
and
it's
cleaned
up
considerably
from
the
raw
data
that
he
was
exploring.
So
you,
basically
what
you
have
is
you
have
like
a
cell
number,
the
time
that
it
happens
in
the
spatial
position,
and
so
you
can
create
three-dimensional
graph,
so
they
don't
have
the
graphs
in
repository.
I
may
not
have
the
demurrer.
I
may
just
have
the
graphs
that
I
have
in
the
paper.
There.
A
Sure
should
have
brought
this
with
me
here,
but
I
have
these
sort
of
three
dimensional
graphs,
where
and
you'll
have
to
trust
me
on
this,
like
I,
actually
put
it
on
slag,
where
you
have
this
three-dimensional
space,
and
we
have
these
points
that
represent
the
cells
that
are
tracked
and
they're
averaged
out
to
a
common
position
and
then
they're
located
in
this.
Basically,
what
ends
up
looking
like
this
big
sphere
of
dots,
you
know.
A
So
one
of
the
graphs
I
have
here
is
differentiation
by
divisions.
So
this
means
this
is
a
c
elegans
embryo
and
this
access
is
in
minutes
all
right.
So
this
actually
goes
way
out
to
like
after
the
egg
hatches
and
it's
and
they
call
this
post
embryonic
cell
division,
but
this
is
the
pre
embryonic
period
from
about
zero
to
about
eight
hundred
minutes
or
so
and
at
about
750
or
so
you
can
see
this
blue
line.
This
is
actually
beyond
hatch.
So
it's
a
bird
about
here
but
anyways
these.
A
Well,
actually,
that's
not
what
that
means,
but
it's
it's
about
here,
so
these
orange
pulses.
You
can
see
kind
of
the
lines
coming
up.
These
are
like
subdivisions
and
embryogenesis
and,
as
you
can
see,
it's
not
uniform.
You
know
and
I'll
divide
it.
You
know
in
our
consistent
intervals,
or
you
know
it
goes
from
two
to
four
to
sixty
or
from
two
to
four
to
eight
to
sixteen
cells
sort
of
uniformly.
There
are
these
pulses,
so
you
get
these
pulses.
A
It's
not
immediately
clear
what
the
timescale
is
here:
they're
actually
multiple
time
scales,
but
if
there's
definitely
a
pulsing
that
happens-
and
you
see
this
in
the
post,
I'm
working
on
a
case
to
where
you
have
different.
They
happen
in
clusters,
they
don't
happen
in
uniformly.
Then
you
have
this:
this
craft
divisions
during
embryo.
Just
again
this
is
a
histogram
of
C
elegans,
embryo
Genesis
and
again
you
have
cell
divisions,
and
this
is
set
up
sort
of
a
close-up
of
this
period
of
embryogenesis
from
zero
to
four
hundred
and
forty
minutes.
A
And
you
get
these
like
these
bursts
of
activity.
It's
not
always
uniform,
so
200
minutes
there
aren't
any
cell
divisions
recorded
the
same
thing
at
350,
360
minutes,
and
so,
but
you
do
get
these
periods
where
there
and
this
frequency
is
I,
think
that
the
proportion
of
cells
born
at
a
certain
time.
So
these
bins
are,
you
know
there
isn't
any
activity
until
the
first
division.
So
this
is
a
single
division.
A
Then
these
are
two
divisions
and
oh,
this
is
like
I
think
these
are
kind
of
in
the
forest
celled
eight
cell
stage,
and
then
it
goes
out
like
that,
but
you'll
notice
that
there
are
these
rounds
of
divisions
that
are
that
occur
over
time,
and
so
this
I
mean
this,
and
you
can
also
see
this
in
zebrafish
where
you
have
early.
This
is
the
first
stage
of
zebrafish
development
and
zebrafish
development.
Is
a
bit
different
from
C
elegans,
but
you
see
actually
it's
very
distinct
here,
where
you
have
this
early
spot
I.
A
Don't
really
have
the
graph
sleeper
labeled
very
well.
This
is
like
a
fertilization
and
then
this
is
like
a
hundred
and
fifty
minutes
or
something,
but
you
see
this
sort
of
periodicity
and
then
it
breaks
down
after
a
while,
but
I
mean
the
whole
idea
of
the
papers
is,
is
similar
to
this,
where
you
know
in
I,
don't
want
to
make
it
like
kind
of
like
this,
this
sort
of
undisciplined
tour
of
things,
so
it's
gonna
have
to
be
tightened
up
a
bit.
A
There
are
maybe
some
specific
questions,
but
the
idea
is
that
you
have
these
frequencies
of
self-worth
and
things
like
that,
and
you
know
it's
an
attempt
to
sort
of
talk
about
that.
It
may
be
in
a
quantitative
manner,
because
a
lot
of
the
explorations
of
this
have
been
with
data.
You
know
you'll
have
like
a
paper
where
they
describe
like
cell
divisions
and
they
give
a
graph
which
I
gave
a
graph
just
to
characterize
it,
but
they
don't
really
give
very
much
interpretation
in
terms
of
maybe
the
system's
level.
C
A
Yeah
and
I
mean,
like
you,
know,
they're
they're
different
methods
we
can
apply
to
like.
We
could
use
a
machine
learning
approach
to
some
of
it
and
then
you
know
present
like
a
sort
of
just
from
the
data.
You
know,
like
you,
have
your
your
your
divisions.
You
know
you
can
and
can
characterize
divisions,
but
you
can
also
you
know
kind
of
find
deeper
patterns
in
the
data.
So
you
know
that's
kind
of
what
in
some
ways,
people
will
use
machine
learning,
find
like
a
deeper
pattern
in
the
data.
A
C
A
C
A
A
A
A
It's
not
image
data,
though
it's
tabular
data,
so
in
that,
in
that
repository
they
have
a
lot
of
the
tabular
data
and
that's
what
we
worked
from
in
the
past.
A
lot
of
that
where
it
gives
you
like
information
about
it,
sort
of
distilled
down
so
like
for
the
EPIK
data
set
that
my
oak
was
showing
a
lot
of
that
is
like
you
know,
they
have
an
algorithm
that
they
used
to
align
the
images,
but
it
isn't
like
anything.
A
That's
like
machine
learning
is
just
kind
of
an
alignment
algorithm
that
they
use
to
produce
like
process
data.
But
the
problem
is:
that's
not
images
anymore.
They
just
kind
of
like
extracted
directly
from
the
images
and
produce
numbers,
so
they're
all
useful
for
machine
learning
directly.
Like
you
know,
if
machine
learning,
we
want
to
find
things
and
images
but
they're,
you
know
they're
just
kind
of
aligning
the
data
and
it's
you
know
like
you've,
seen
the
data
either
these
like
bring
dots
so
they're,
just
taking
the
green
dots
and
then
localizing
them
and
they're.
D
A
But
but
that's
that's
so
those
are
you
know
they
deal
with
the
irregularities
in
the
data
that
way,
so
we
do
have
like
tabular
data
for
a
lot
of
these,
and
but
we
can
easily
you
know
we
could
use
image
image
and
get
it
more
different
types
of
information
out
of
it.
So
that's
not
too
much
of
a
problem,
but
I
can
like
produce
like
a
sort
of
an
outline
to,
or
you
know
we
can
have
a
discussion
about
this
in
slack.
A
I,
don't
know
what
if
this
would
be
like
good
for,
like
you
know,
individual
discussions
or
if
we
can
just
talk
about
on
the
channel
I
guess
we
can
talk
about
in
the
evil
worm
channel.
It
would
be
fine,
but
I'll
try
to
come
up
with
some.
You
know
and
I
kind
of
an
outline
of
what
we
might
want
to
expect
what
we
might
want
to
get
in
terms
of
data
and
organizing
that
a
little
bit
horror.
F
A
Well,
we
could
what's
of
them
working
kind
of
on
developing
like
pseudo
data
sets
for
embryos,
so
you
know
kind
of
basing
it
on
the
C
elegans
example,
where
you
try
to
figure
out
like
what
the
you
know,
try
to
come
up
with
different
functions
or
distributions
for
characterizing
cell
divisions
and
so
I.
Don't
know
if
that's
I,
don't
think.
That's
in
that
repository,
but
like
basically
creating
datasets
that
are
like
not
real
embryos
but
they're,
based
on
some
of
the
principles
that
were
extracting
from
my
C
elegans.
C
A
A
F
C
F
F
Yeah
but
gas
you
know,
provide
more
offers
versatility
to
the
data
right.
If,
even
if
we,
you
know
augmented
data
in
case
of
you
know,
for
example,
if
we
are
flipping,
you
know
sales
notification,
it
might
you
know,
change
the
whole
semantics
of
the
situation
right
if
they
are
like
dividing
from
left
to
right
and
if
we
pit
now
they
are
kind
of
dividing
from
top
to
bottom.
It
might
you
know,
change.
A
F
F
So
there
it
may
or
may
not
do
you
know
flipping
the
dot
or
augmenting
the
data
you
know,
may
change
the
essence
of
the
dot,
unlike
if
there
is
something
that
the
cell
is
dividing
from
left
to
right
will
flip
it
or
you
know
we
kind
of
transpose
it.
It
fits
now
kind
of
from
right
to
left
or
from
top
to
bottom.
F
It
might
change
the
essence
so,
but
if
you
try
to
generate
some
of
the
artificial
data
which
may
or
may
not
be
beneficial,
but
you
know
we
can
always,
you
know,
try
to
do
the
thing
with
it
works
or
not,
yeah
sure.
And
then,
if
we
like
the
of
the
tabular
data-
and
we
can
you
know
somehow
I-
don't
know
that
if
this
thing
will
go
up
or
not,
but
we
can
always
try
everything
you
know
convert
this
into
time
see
this
thing
makes
sense
we
have.
A
A
yeah,
just
to
kind
of
decide
on
that.
But
I
would
out
prepare
like
a
document
and
we
can
work
from
there
and
then
we'll
go.
You
know
and
then
kind
of
create
issues
from
there.
So
get
up.
Issues
I
think
we're
at
the
end
of
our
meeting
time.
I
wanted
to
well
I,
guess:
I,
don't
have
to
really
go
over
issues
today.
I
think
we're
on
the
G
side
board.
I
think
we're.
Okay,
I
was
going
through
that
this
last
week
and
I.
A
D
F
D
Bits
would
whatever
people
want
to
go
into
me.
You
can
make
us
stuff
about
that
at
the
Semak
directory
system,
illogical
directory
and
clear
your
own
resources,
or
it's
very,
it's
very
freeform
and
you've
been
asked
and
Bradley
we're.
Making
an
issue
like
hey
I'd,
like
to
look
more
into
this
topic,
so
feel
free
to
go
ahead
and
anybody
can
go
ahead
and
use
it
observer.
Otherwise,
yeah
yeah.