►
From YouTube: DevoWorm (2020, Meeting 12): Axolotl microscopy, Education Stack update, and EfficientNet
Description
DevoWorm meeting: April 6, 2020. Attendees: Richard Gordon, Ujjwal Singh, Bradly Alicea, Tom Portegys, Susan Crawford-Young, Vinay Varma, and Jesse Parent
B
A
B
C
D
B
E
E
A
F
B
G
B
B
F
B
B
B
B
B
D
B
B
B
E
F
B
B
D
C
B
A
B
B
G
A
A
A
A
B
A
C
E
E
F
C
F
C
F
F
F
F
E
E
F
F
F
E
C
F
C
F
F
F
F
F
F
F
C
F
F
F
E
F
F
F
C
F
C
C
F
F
F
C
C
F
F
F
C
F
F
C
F
C
C
F
C
F
F
F
C
A
Yeah,
thank
you.
Visual
was
a
good
presentation.
I
know
we
had
last
summer
I
think
we
had
discussed
depth
of
network
of
deep
learning,
networks
and
sort
of
the
advantages
and
disadvantages
of
that
depth.
So,
like
you
know,
going
deeper
versus
really
deep
versus
a
shallow
network,
and
so
I
think
we
talked
about
like
how
it's
generally
advantageous,
but
there's
like
an
increase
in
the
error
rate,
depending
on
the
the
way
it's
implemented.
So
the
scaling
I
mean
that's
a
problem
in
a
lot
of
areas
of
science.
A
A
A
So
Joey
about
like
how
to
best
use
your
computational
resources
just
yeah,
so
their
needs
can
get
better
performance.
Given
that
cuz
I
mean
you
know,
we
don't
we
don't
think
about
it
too
much
early
in
computer
Lee
years
of
computer
science
before
concerned,
even
obsessed
about
using
memory.
You
know
efficiently
because
they
didn't
have
a
lot
of
it.
But
of
course
they
still
for
deep
learning
models.
We
don't
have
infinite
memory,
and
so
it
would
be
good
to
be
able
to
use
things
efficiently.
A
Okay,
yeah
thanks
again
as
well
and
I
can,
if
you
can
send
me
the
slides
afterwards,
I
can
make
them
available.
Maybe
if
there's
a
paper
I,
don't
know
what
the
paper
was
associated
with
this,
but
I
think
it's
in
our
slack
channel.
So
if
you
want
to
check
out
the
deform
slack,
Channel
or
else
I
can
send
it
in
an
email
as
well.
A
A
You
know
what
this
is.
This
is
the
machine
learning
education,
so
it
combines
biological
modeling
machine
learning
and
then
some
good
old
fashioned,
Sidra
biological
data
analysis
topics
in
embryo
physics,
which
is
the
embryo
physics
seminar.
So
these
are
a
series
of
seminars
and
I'm
still
working
in
how
they
might
be
presented,
and
we
can
talk
more
in
depth
about
that.
A
A
So
the
whole
thing
is
like
I
said:
the
lessons
themselves.
The
materials
are
hosted
on
github
I.
Have
it
set
up
for
the
open-room?
Deform
curriculum,
for
example,
is
that
the
I
have
like
stubs
on
in
github
and
those
are
like
posted
as
markdown
files.
So
it's
like
you
go
there
and
it's
like
I
saw
of
us,
so
you
go
through
it
and
you
pick
a
lesson
and
the
lesson
might
be
on
YouTube.
A
It
might
be
hosted
natively
on
github
or
it
might
be
a
badge,
a
micro,
credential,
host
and
I'm
bad
juist
and
I.
Think
a
lot
of
you
know
what
badges
are
I'm,
not
gonna,
go
in
too
much
into
that.
But
the
idea
is
that
you
would
engage
with
each
of
these
things
and
then
you
could
submit
lessons
like
you
know.
You
would
review
the
material
and
then
there'd
be
some
assignment,
and
then
you
would
submit
the
assignment
on
get
out.
A
If
that's
what
we
call
for
in
the
assignment.
So
all
those
things
are
here
but
they're
all
kind
of
scattered
Google
classroom,
then,
is
what
I
use
to
replace
the
leotta
me
and
the
replacement
the
reason
I'm
using
Google
classroom
here
to
integrate.
All
of
these
is
because
it
allows
people
to
enroll
in
the
course
and
then
access
everything
that
you
could
access
to
github.
But
it's
like
manna
it's
a
course
management
system
where
you
can
actually
keep
track
of
people
and
you
can
keep
track
of
their
assignments.
A
So
it's
still
kind
of
scattered
I
mean
you're
gonna
have
maybe
some
people
doing
things
to
get
out,
maybe
some
through
Google
classroom,
maybe
some
through
bad
juist.
But
at
least
you
know
in
this
way
you
can
track
who's
actually
pursuing
the
course
and
you
know
doing
assignments,
and
then
you
know
you
can
further
verify
that
their
badge
list
or
github-
and
you
can
you
know
you-
can
evaluate
their
work
and
maybe
give
them
some
sort
of
credit
in
the
organization.
A
These
are
an
accredited
courses.
There
just
may
be
casual
learning,
that's
that's
where
what
we're
aiming
for,
but
I
think
you
know
the
idea.
Also
is
you
want
to
keep
track
of
what
people
are
doing,
so
you
know
you
can
interact
with
them
or
they
can
follow
up
with
your
organization
if
they
want
to
contribute
their
different
ways
to
you
know
reward
people
for
certain
courses
and
so
forth.
A
So
we
won't
get
into
that
contribution
mode,
but
that's
that's
basically
the
model
for
it
for
learning,
but
then
there
are
also
a
couple
other
tools
that
are
sort
of
on
top
of
Google
classroom
that
augment
this
so
eliademy
had
a
chat
function
which
you
could
communicate
with
course,
mates
and
that's,
of
course,
an
essential
part
of
education
and
so
to
make
us
integrated.
I've,
started
some
channels
on
something
called
git,
er
and
together.
I
think
some
of
you've
used
it
before
for
an
evil
worm.
A
But
it's
a
chat
platform
where
you
have
channels,
and
then
you
post
a
message
in
the
channel
where
people
respond
to
you.
The
advantage
of
using
git
er
here
is
that
it's
integrated
with
github
here,
so
that
there's
an
integration
of
materials
and
of
you
know
messages
and
things
like
that.
So
it's
integrated
in
one
way
that
it's
a
Google
classroom
directly
but
with
github
and
then
the
other
tool.
Is
this
interactive
Milt
making
tool
called
hypothesis
and
I?
Don't
know
how
many
of
you
have
used
this?
A
It's
it's
a
so
you
plug
it
into
a
Chrome
browser.
If
you
use
Chrome,
if
you
don't
use,
Chrome
I
think
you
can
still
use
it
through
other
means,
but
you
go
to
their
website,
which
is
hypothesis
da,
is
so
it's.
This
word
just
type
it
into
your
browser
and
you
could
go
to
the
site
and
they
you
know
it's
like
a
downloadable.
It
basically
allows
you
to
annotate
web
pages,
so
I
can
annotate
any
web
page
I
want
a
visit.
A
A
web
page
I
can
put
an
annotation
on
the
page,
and
then
people
visit
that
page.
If
they
have
this
tool,
they
can
view
the
notes.
So
this
is
actually
something
I
pioneered
in
the
open
when
beaver
worm
curriculum
where
they
would
go
to
a
stub
on
github,
and
they
would
look
at
the
lesson
and
then
there
would
be
notes
put
on
the
page
so
that
you
would
see
okay,
I
love.
A
The
page:
this
is
a
lesson
there
are
eight
notes
here
and
the
it
notes
might
range
from
like
this
is
a
nice
lesson
to
some
problem.
They
had
with
it
to
some
question
they
had
about
the
curriculum
and
they're.
All
just
kind
of
you
know
arranged
there
for
different
users,
and
so
it's
a
way
to
sort
of
annotate
your
experience,
something
you
know
you
couldn't
do
before
with
it.
A
I
use
this
before
with
eliademy,
but
it's
something
that's
not
integrated
into
a
lot
of
MOOC
platforms,
because
the
MOOC
platforms
really
I
tried
other
MOOC
platforms
and
they
weren't
very
good
at
this
sort
of
scale.
I
wanted
to
get
out
with
these
courses,
and
so
that's
that's,
where
I'm
going
with
this
sort
of
thing
and
I'm
gonna
incorporate
the
embryo
physics
materials
into
this.
A
A
So
hopefully
this
works
out.
Then
this
is
an
example
of
it
here.
It's
it's
optimized
for
Firefox!
It's
not
it's
not
great!
In
at
least
this
is
the
platform.
This
is
the
Google
classroom
platform.
I
should
say
it's
not
great
in
chrome,
but
it's
better
in
Firefox,
but
you
can
use
it
in
Chrome,
and
so
it's
set
up
like
this.
You
you
come
into
the
you,
have
a
series
of
lessons
or
a
series
of
courses
that
are
listed.
A
Then
you
go
into
a
course,
and
this
is
one
of
the
courses,
so
this
is
the
open
line,
beaver
room
curriculum
and
you
have
your
little
banner
here.
You
have
the
stream,
which
is
like
where
you
have
messages
and
things
that
have
been
contributed.
You
have
the
coursework
tab,
which
is
and
I
know
it's
hard
to
see,
but
where
you
have
the
actual
assignments
or
the
actual
modules
listed
so
for
open
one
diva,
one
curriculum,
it's
all
of
these
stubs
on
github
sort
of
organized.
A
In
this
way-
and
in
this
way
we
can,
you
know,
give
people
they
can
go
to
a
lesson.
They
can
go
to
all
the
resources.
So
this
is
not
just
from
get
out.
This
is
also
the
badges.
So
these
are
the
micro
credentials
here
for
development
from
get
from
badges
and
they're
integrated
with
the
lessons
from
github.
So
these
different
modules
here
and
then
you
can
add
YouTube
videos
into
this
mix
as
well.
So
you
could
say:
here's
a
YouTube
video
and
then
give
them
instructions
on
what
to
do
with
it.
A
Then
there's
a
people
tab
which
tells
you
the
people
in
the
course
and
then
the
grades,
which
that's
something
that
I
think
if
you're,
using
like
this
for
elementary
education
or
collegiate
education
and
actually
is
an
important
tab,
but
we're
not
worried
about
that
so
much
right
now.
So
that's
the
thing
I
want
to
talk
about
the
education
stack
I
just
wanted
to
give
an
update
on
that.
If
you
have
any
thoughts
about
that,
let
me
know.
Let
me
see
what
the
comments
are
here.
C
A
G
G
G
G
G
A
C
A
C
C
Would
say-
and
you
say
we
see
a
kind
of
thing
in
my
frontier
master,
but
I
was
looking
at
classrooms
for
more
things
and
one
of
my
friends
is
I've
been
looking.
They
making
an
online
course
wait
to
be
one
of
my
trainers,
so
I
will
talk
about
some
of
those
things
later,
but
it's
interesting
to
see
all
together
the
ecosystem.
So
thanks
for
that,
oh
yeah.
A
Yeah
I
mean,
if
there's
any
feedback
again,
we
can
talk
about
it,
I
and
then
I,
just
kind
of
came
up
with
this
I
mean
that's
what
I
came
up
with
during
the
Mozilla
open
Leaders
program,
so
it's
like
that
was
the
sort
of
a
template.
I
was
using
there
probably
more
efficient
ways
to
do
this
with
that
again
without
you
don't
want
to
I
didn't
want
to
go
for
something,
that's
really
for
scaled
up
classrooms.
A
A
Okay,
just
I
just
wanted.
Okay,
so
now
I
think
we'll
go
to
the
exa
lotto
data
and
we
can
talk
about
the
soft
line
too.
If
you
want,
let's
see
the
X
allotted
that
is
seasons
here,
I
just
want
I
haven't
had
a
chance
to
really
dig
really
deeply
into
this,
but
I
do
have
some
some
animated
gifts
to
show
you
so
Susan
sent
me
these
data,
and
these
are
of
her
flipping
microscope
where
she
takes
an
X
a
while
embryo
and
the
microscope
flips
it
like.
A
You
know
it
flips
it
on
its
end,
so
that
the
whole
thing
flips.
You
know
a
liquid
medium
and
you
get
this
full
view
of
the
embryo,
and
so
she
sent
me
some
images
and
there
are
a
lot
of
different
shots
of
these
embryos
and
there
you
can
come
down
to
a
single
view
of
the
embryo,
that's
like
about
maybe
9
or
10
images.
Deep.
So,
like
you
get
this
process
of
flipping,
what
does
vinay
have
to
say
here?
Last
year,
I
worked
on
some
assignments
for
badges.
A
It
was
a
good
and
nice
way
for
someone
new
to
get
both
theoretical
knowledge
and
practical,
no
yeah,
that's
how
we
were
aiming
for
with
the
badges
so
well.
You
know
this
is
a
little
bit
more
in
depth
than
the
badges,
but
badges
are
good,
so
anyways
that
let
me
go
back
to
the
axolotl
data,
so
you
get
these
these
sequences
of
about
nine
or
ten
images
that
sort
of
characterize
us
hoping
process.
So
this
is
the
one
of
the
gifts
that
I
created
here.
A
A
Yeah
so
I
mean
I
just
wanted
to
get
it
like
this,
so
I
could
visualize
what
was
going
on
with
the
process.
So
I
mean
this
is
that
these
are
the
images
if
we
I
can't
stop.
The
gif
I
should
have
made
a
movie
of
it,
but
you
can
see
that
there
are
these
cells
here
and
you
can
see
the
boundaries
and
you
can
see
that
there's
one
side
and
then
this
other
side,
and
one
of
the
advantages
here
is
that
you
can
see
you
know
kind
of
the
in
you
know
a
little
bit
in.
A
You
can
actually
get
some
depth
information
out
of
these
images.
So
this
is
a
series
of
images
here,
maybe
about
eight
seven
or
eight
images
in
sequence
and
they're,
all
sort
of
a
line
so
they're
all
and,
like
a
the
you
know,
identical
reference
frame.
So
it's
not
like
you
have
to
align
them
too
much.
You
can
see
that
they're
sort
of
in
sequence.
A
So
now
we
have
this
embryo
that
we
can
define
it's
moving
a
little
bit,
but
you
can
get
those
cells
and
you
can
remind
sort
of
the
boundaries,
those
cells
over
time
as
it's
moving.
You
can
maybe
correct
for
that,
but
you
have
a
pretty
good
view
of
the
whole
thing.
So
you
have
this
depth
information.
You
have
these
sides
of
the
embryo
and
so
forth.
So
now,
so
that's
the
the
images
that
we
have
and
we
have
a
sequence
now.
The
next
step
is
to
try
to
segment
the
images
so
how
it
may
be.
A
You
create
a
visualization
paradigm
for
it
where
we
we
talked
about
using
sort
of
a
Google,
Maps
type
approach
where
you
have
like
the
surface
of
the
embryo
spread
out,
and
you
can
search
the
surface
different
frames
on
the
surface.
I
can't
remember
exactly
the
details
on
that,
but
that
was
one
of
the
things
we
talked
about,
but
there
are
other
options
as
well
and
that
we've
been
talking
about
using
machine
learning.
For
some
of
these
you
know
some
segmentation,
stuff
and,
and
some
of
some
other
analyses.
A
So
you
know
this
might
be
a
good
opportunity
to
apply
some
of
this
computation
to
this.
This
data
set-
and
there
are
a
lot
of
images
here.
I
just
showed
you
a
couple
examples,
but
there's
a
lot
of
data
that
I
have
I,
don't
have
a
number
on
it.
I
mean
some
kind
of
vector
model.
The
vector
model
might
be
useful
here.
A
A
They
take
it
off
of
that
sphere
and
they
flatten
it
out
and
the
flat
flattening,
introduces
a
mathematical
transformation
or
you
have
to
use
a
mathematical
transformation
to
flatten
it
out,
and
so
you
take
the
curvature
out
and
you
have
as
a
flat
representation.
That
means
that
somewhere
in
the
map,
you
have
to
make
distortions.
So
there
are
different
projections
of
the
earth
of
the
surface
of
the
earth
that
are
based
on
just
projecting
that
spherical
information
to
flat
information.
There
actually
are
not
models
that
minimize
that
distortion.
A
A
Well,
yeah
I
guess
we
could
do
that.
We
could
have
like
you
could
have
like
a
you
know,
could
zoom
in
on
different
parts
of
it,
but
there
yeah.
There
are
a
lot
of
options
here.
I
don't
know
like
you
know,
we
couldn't
build
a
3d
model
like
you
know,
with
a
wireframe
model
and
then
put
information
on
it.
That
way
too.
It's
not
you
know
you
could
mean
something
like
blender
to
sort
of
model.
A
Yes,
but
you
can
see
the
whole
sphere
in
Earth,
yeah
and
Google
Earth.
You
can
see
the
sphere,
but
the
tiles
of
it
I
think
are
in
treaty
the
way
it's
rendered
I'm
not
really
sure
how
exactly
how
they
do.
It
I
mean
they've
done
it.
That's
what
the
tradition
we
did
it,
but
you
can
zoom
out
to
see
the
globe
but
I'm
not
sure
exactly
how
they're
building
that
projection.
A
A
Is
that
you
have
you
know
if
you
have
depth
information,
you
can
do
a
lot
of
tricks,
yeah,
there's
Google,
Mars,
I,
think
Google,
Venus,
Google,
moon,
they've
done
a
lot
of
different
planets
and
it's
just
the
information
from
like
space
probes,
but
they've
mapped
the
different
parts
of
the
planet,
and
actually,
that
might
be
a
good
presentation
for
another
week,
is
to
show
how
those
kind
of
models
actually
work.
Cuz,
I'm,
not
up
to
date
on
the
state
of
the
art.
A
A
A
Ya
know,
and
it's
something
I
think
we
could
put
like
turn
into
a
tool
put
on,
because
we
have
our
Devo
zoo
and
we
have
our
machine
learning
tool
that
we
have
now,
which
is
going
to
be
updated.
This
summer
sure
you
know
just
improved
upon
this
summer,
and
so
maybe
we
can
integrate
it
that
into
the
set
of
tools
and.
B
A
A
So
that
I
mean
there
are
a
lot
of
ideas
here.
Why
don't
we
yeah?
Why
don't
come
up?
Maybe
next
meeting
we
can
revisit
this
and
come
up
with
some
ideas
of
how
to
do
this
and
it's
gonna
be
like
you
know:
they're
gonna
be
two
categories:
there's
gonna
be
what
I'd
love
to
see
and
then
maybe
what's
technologically
feasible,
you
know
in
not
like
three
year
period
but
like
something
we
can
do,
and
then
you
know
if
we
can
integrate
things
like
machine
learning
into
it.
That
would
be
great
too.
A
A
He's
got
it
in
the
chat,
so
I
was
gonna
present
a
paper,
but
I'm
not
gonna,
be
able
time
today,
I
wanted
to
say
one
more
thing
before
we
go,
and
that
is
that
I'm
headed
off
to
this
organizational
meeting.
This
is
brain
web.
This
is
a
new
collaboration
that
this
group
is
putting
on
their
European
group
and
I've
been
talking
with
them
about
different
things.
I
met
them
at
a
comp,
two
conferences,
the
spring
virtual
conferences,
and
they
have
this
nice
site
in
it's
a
collaboration
site.
A
So
what
you
can
do
is
you
can
attend
their
events.
They
have
these
packet
ons
happen
in
every
three
weeks.
They
have
this
community.
So
if
you
go
to
the
community
tab,
you
can
join
this
network
of
research
and
there
are
some
people
from
open
room
in
this
network.
So
this
isn't
just
a
small
group
of
people.
A
So
this
is
the
brain
web.
These
are
neuroscience
people,
but
there
are
other
people
as
well
a
lot
of
people
here
if
you
join,
if
you
join
and
you
register
and
you
put
some
tabs
as
to
your
interests,
you
can
join
this
network
and
it's
you
know
classified
by
these
tags.
So
there's
a
lot
of
experience
in
neural
imaging,
but
also
in
other
areas
of
neuroscience
and
there's
overlap
with
development,
I
guess,
people
doing
microscopy
and
imaging
of
different
types
so,
and
and
also
open
programming,
open
science.
A
So
there's
a
lot
of
stuff
there
and
then,
of
course
they
have
projects
that
are
ongoing,
so
they're
collaborations
that
they
have
calls
for
they
don't
have
any.
They
have
a
couple
listed
now
and
these
are
like
github
repositories
that
they
can
link
to.
So
so
that's
the
that's
the
thing
I'm
headed
off
to
next
to
invite
you
to
join
in.
If
you
want-
or
you
know
if
this
is
something
you're
interested,
we
can
talk
about
more
as
well
and
so
I
think.
A
That's
it
for
this
week,
we're
at
the
top
of
the
hour
next
week,
I'll
present
these
papers.
That
I
was
going
to
present
this
week
and
maybe
we'll
talk
more
about
the
axolotl
embryo.
How
to
you
know,
proceed
on
the
analysis
and
talk
you
know
and
then,
if
you
have
anything
else,
you
want
to
bring
up,
please
bring
it
up.
I
know,
Tom
and
Jesse
are
interested
in
education.
Maybe
you
know
you
can
we
can
discuss
that
a
little
bit
more
or
anything
else
that
you'd
like
to.
B
A
B
C
B
E
A
Yeah,
that
would
be
great
if
we
could
do
that.
I
mean
we'd.
Add
like
wooden
boards,
would
you
know,
have
a
representation
but
I
think
that
would
be
excellent
information.
So,
okay,
this
is
the
Google
embryo
paper
that
dick
published
about
11
years
ago.
So
this
is
the
paper
Google
embryo
for
building
quantitative
understanding
of
an
embryo
as
it
builds
itself.
So
this
is
the
thing
that
they
had
it's
in
the
chat
here.
I
can
download
it.
I
can
send
it
out
to
the
group
in
an
email.