►
From YouTube: Inaugural Meeting: DevoWorm ML
Description
First (informational) meeting, September 4. Attendees: Bradly Alicea, Jesse Parent, Vinay Varma, Abraham Kohrman, and Richard Gordon.
Slides: https://docs.google.com/presentation/d/1K1Fmn0GTz1pNIhyyJ0b5Xe_RwPyslU4yRSYW02zlnR4/edit?usp=sharing
A
All
right,
why
don't
we
get
started?
They'd
hoped
they'd
have
a
few
more
people,
but
that's
okay,
I'm
gonna
record
these
sessions,
and
hopefully
people
can't
make
it
I
get
a
few
inquiries.
What
people
couldn't
make
it
and
hopefully
they
can
make
it
to
the
recording
so
welcome
to
this
inaugural
meeting
of
deva,
worm
ml.
So
I
know
that
you,
maybe
you
have
questions
about
what
we're
gonna
do
so
I'm
gonna
do
a
short
presentation.
A
A
So
why
don't
we
start
with
introductions?
I'm
gonna,
stop
sharing
for
a
minute
and
I
want
you
to
give
a
couple
pieces
of
information.
If
you
can
give
your
name
your
level
of
education
or
affiliation,
your
area
of
expertise,
whatever
you
think
that
is,
and
then
name
one
thing
you
would
want
to
learn
from
this
group.
A
C
C
Computer
vision
based
project-
that's
also
my
interest
area
interest
like
in
computer
vision.
I
want
to
pursue
my
personal
career,
so
that's
maybe
I
want
to
go
for
higher
studies
regarding
that
pursue
a
masters
regarding
involved
in
machine
learning
in
software.
So
that's
that
and
from
this
group
I
would
like,
like
I,
always
believe
when
we
share
our
resources
and
Michelle,
then
we
put
a
color
collaborative
effort,
then
I
think
it's
it's
good
for
everyone
like
elderman,
gets
to
learn
everything
and
I
think
so.
C
A
E
Computation
and
technical
techniques
to
studying
a
lot
of
things,
I
particularly
stand,
let's
say,
complexity
or
maybe
anachronistic
term
of
cybernetics,
but
I'm
I'm
happy
to
have
that
working
with
some
of
those
things
in
well.
They
came
up
in
the
agonal
labs
and
the
labor
group
over
the
US
over
the
summer,
so
I
understand
he's
continuing
working
with
those
things
and
exploring
that
over
just
in
complexity
and
particularly
within
the
field
of
cognition
and
things
that
contribute
to
cognition.
B
B
B
Well,
there's
a
nice
example
because
we
can
talk
about
individual
size
and
what
they're
doing
so
I
very
much
look
forward
to
this
in
terms
of
machine
learning.
I
am
only
a
ocean.
Expander
of
what's
been
going
on,
I
haven't
done
it
myself,
but
I've
done
enough
reading
to
be
skeptical
about
accepting
its
results.
Okay,
thanks:
okay,.
A
D
A
So
yeah
well
yeah,
it's
ready.
I
was
saying,
though
I
think
we
have
a
pretty
broad
set
of
interests
here
and
I.
Think
it's
interesting
that
we
have
a
healthy
skepticism
of
machine
learning.
At
least
maybe
you
know
this
is
one
of
things
we
can
explore
in
this
group
is
like
you
know,
you
see
a
lot
of
papers
and
they
purport
these
wonderful
results,
but
you
know
they're
not
what
what
do
they
really
mean
in
there?
A
F
A
A
Okay,
so
then
we
back
to
the
slides,
so
he
did
the
introductions
and
so
a
little
bit
about
machine
learning.
So
there's
like
I,
said:
there's
a
lot
of
hyper
on
machine
learning
and
so
I
just
took
this
definition
from
Wikipedia
machine
learning
is
defined
as
the
scientific
study
of
algorithms
and
statistical
models
that
computer
systems
use
to
perform
a
specific
task
without
using
explicit
instructions
relying
on
patterns
and
inference.
Instead,
it
is
seen
as
a
subset
of
artificial
intelligence.
So
you
know
there
in
that
definition,
you
know,
people
might
start
to
have.
A
You
know,
might
have
a
different
definition,
but
I
think
that's
I
mean
that's
kind
of
you
know
some
sense.
What
we
want
to
look
at
with
respect
to
biology,
you
know,
is
you
know,
can
we
apply
these
types
of
methods
to
biology?
And
you
know,
is
it
useful?
What's
the?
What
are
we
getting
out
of
it?
So,
and
we
have
you
know
some
computer
science
type
people,
some
biologists
and
people
who
know
something
about
the
methods
I'm
going
to
talk
about
here,
which
is
a
graphical
history
of
machine
learning.
A
That's
that's
nice
that
we're
able
to
learn
about
those
things
but
they're
also
things
not
on
this
list
that
probably
have
contributed
in
machine
learning
as
well,
and
maybe
those
are
things
that
we
can
discuss
in
the
group.
So
quick,
summary
of
machine
learning.
Basically,
one
of
the
sort
of
the
models
for
machine
learning
is
this
idea
of
taking
data
labeling
it
in
some
way
and
then
training
things
and
then
testing
actually
how
these
labels
mixed
up
here
on
this
diagram?
A
Okay,
you
start
with
a
data
set
here,
and
you
know
it
has.
It
can
be
fairly
large,
although
it
doesn't
have
to
be,
and
you
want
to
find
patterns
in
the
data
like
you
usually
do
with
the
statistical
analysis,
and
then
you
can
either
use
that
unlabeled
data
there's
what
we
call
unsupervised
techniques
or
you
can
label
the
data
and
something
we
call
supervised
techniques,
and
so
supervised
techniques
are
maybe
easier
to
do.
But
people
have
done
a
lot
of
things
with
unsupervised
techniques
and
so
labeling.
A
The
data
means
putting
names
on,
like
maybe
segments
of
an
image
or
attaching
functional
information
to
some
data.
Point
things
like
that
and
then,
when
you
get
to
your
actual
model,
that
you're
going
to
use
for
machine
learning,
there
are
two
aspects
to
this
analysis.
First
is
the
training
period,
which
is
where
you
take
some
data
and
you
train
the
machine
or
you
train
the
kernel
or
whatever
it
is
to
look
for
patterns.
So
you
give
it
like
a
fair
data.
A
You've
give
it
as
an
input
and
then
the
Machine
kind
of
learns
the
associations,
and
then
it
has
a
base.
It
has
a
model
to
work
with
and
then
at
the
end
of
it,
you
have
this
testing
phase,
which
is
where
you're
testing,
for
you
know,
consistent
results
of
that
training
model.
So
this
is
the
result
that
you
get,
and
so
this
is
the
sort
of
thing
when
we
talk
about
machine
learning.
This
sort
of
you
know
three
level
approach.
Where
have
the
data
set
or
label
data?
A
So,
like
you
know,
if
you
had
hands
you
know,
would
you
want
to
train
the
Machine
and
only
hands
of
five
fingers?
Or
would
you
want
to
train
it?
Maybe
on
some
images
of
four
fingers
or
six
fingers?
There
would
be,
you
know,
adversarial
examples
in
the
training
set
and
then
that
would
improve
your
accuracy
on
the
testing
set.
So
this
is
the
kind
of
thing
I'm
talking
about
when
I
talk
about
machine
learning
as
well
and
I.
A
Think
I
think
a
lot
of
you
when
you,
if
you
like,
read
papers
on
it,
you'll
see
what
I
mean.
This
is
the
basic
approach,
but
this
is
only
one
type
of
approach.
We're
interested
in
machine
learning
is
kind
of
a
you
know
that
name
casts
a
wide
net,
and
so
it
may
or
may
not
include
a
lot
of
the
things
I'm
gonna
tell
you
about
here,
but
we're
also
interested
in
things
that
go
beyond
maybe
a
traditional
definition
of
machine
learning.
A
So
there's
something
called
reinforcement,
learning
and
that's
behave
based
on
what
they
call
behaviorism,
which
was
something
these
dinh
psychology
years
ago,
but
has
lived
a
very
robust
life
in
computer
science
as
a
stimulus-response
sort
of
machine
learning
technique,
and
so
sometimes
you'll
run
into
papers.
Like
that,
oh,
we
could
talk
about
those
generative
models,
many
outputs
not
directly
related
to
input.
So
you
know,
if
you
think
about
like
something
like
a
genetic
algorithm
or
even
a
neural
net.
Those
are
things.
A
Are
you
put
an
input
in
and
it
generates
a
lot
of
potential
outputs
that
you
have
to
judge
whether
they're,
good
or
not,
and
but
the
model
that
generates
those
is
called
a
generative
model.
Now
we're
interested
most
things
as
well,
and
finally-
and
this
is
Jesse
mentioned
this
earlier-
we're
interested
in
systems
and
complexity
models,
so
things
like
dynamics,
dynamical
systems,
networks
and
game
theory
and
maybe
more
traditional
cybernetics.
A
So
what
kinds
of
applications
do
we
have?
The
developmental
biology
that
we
can
think
of?
You
know
what
kinds
of?
Why
is
it
useful
to
developmental
biology?
So
one
one
is
image
processing
which
includes
like
segmentation
of
images,
classification
of
objects
in
those
images
and
pattern,
recognition
and
microscopy
data?
So
you,
you
know
you
can
think
of
a
host
of
different
applications
in
developmental
biology,
and
maybe
we
can
bring
some
of
those
to
bear
on
this
group.
Discuss
them
see
what
we
think
about
like
different
papers,
different
applications
and
so
forth.
A
Then
you
know
standard
data
analysis,
so
we
have
problems
in
developmental
biology
that
require
you
know
some
sort
of
analysis
more
than
a
t-test
or,
more
than
like.
You
know
some
sort
of
like
correlation
thing.
You
know
where
you
would
calculate
basic,
descriptive
statistics.
In
this
case
you
know
you
might
have
a
statistical
model
for
looking
at
movement
of
things
or
some
regulatory
prize
and
the
genome.
A
So
there
are
a
lot
of
papers
where
people
in
systems
biology,
especially
where
people
have
attempted
to
analyze.
The
very
large
amounts
of
you
know:
sequence,
data
or
gene
expression,
data
and
they're,
using
like
say
like
Network
models
or
other
things,
and
machine
learning
is
definitely
one
of
those
areas
that
you
know.
People
have
applied
to
these
problems
and
they
haven't
done
it
at
at
such
a
rate
that
we
really
have
good
methods
for
them.
A
So
they're
kind
of
just
people,
just
public,
or
at
least
they
used
to
do
this
in
the
near
past,
they
publish
a
paper
and
it
was
just
like
we
applied
this
technique
to
the
genome.
You
know
some
gene
expression,
data
and
here
the
results,
and
we
don't
really
know
if
it's
a
good
result
or
not
or
a
good
model,
they've
just
they're,
the
only
one
of
who's
ever
done
it.
A
Finally,
there
are
things
in
the
area
of
modeling
and
simulation,
so
people
have
done.
You
know
things
that
basic
largely
are
independent
of
data.
Algorithmic
approaches
to
growth
and
form
come
to
mind.
So
you
know:
you'll
have
like
a
model
of
like
morphogenesis,
actually
I'm
working
with
someone
right
now.
He
he
didn't
make
it
today,
but
I
know
he's
interested
in
the
group
who's
interested
in
modeling
morphogenesis,
and
so
you
know,
one
of
those
things
you
can
do
to
understand.
A
Morphogenesis
better
is
to
create
a
model
that
just
has
a
series
of
rules
attached
to
it,
and
then
the
system
evolves
by
those
rules,
and
then
you
can
analyze
those
outcomes
and
there's
no
data,
that's
put
into
it.
It's
just
a
series
of
rules
and
an
agent-based
model
that
responds
to
those
rules,
and
so
you
know
those
are
maybe
he'll
talk
about
one
of
those
types
of
models
in
the
group,
but
that's
something
we're
interested
in
as
well.
C
A
So
one
of
the
things
that
I
like
to
do
is
I
like
to
have
you
know
a
number
of
channels
where
people
can
ask
question
meet
in
some
way,
so
one
of
those
ways
is
through
github.
We
have
a
github
repository.
I
can
send
out
the
links
to
those
you
know
those
resources
where
that
repository
is-
and
this
is
a
if
those
you
were
aware
of
what
github
is
it's
a
room.
You
know
it's
it's
a
place
where
you
can
submit
things
for
groups,
it's
it's
based
on
computer
code,
but
we
can
exchange.
A
You
know
we
can
do
editing
of
documents
in
it
as
well.
So
it's
a
version,
control
repository,
so
you
have
things
that
are
in
get.
You
know
committed
what
they
call
committed
to
github
and
those
things
then
can
be
edited
by
anyone
in
the
group,
and
you
know
you
can
contribute
new
things.
We
can
also
use
Google
Docs
for
that
as
well.
If
we're
working
on
a
document,
but
that's
one
point
of
collaboration,
another
is
we
have
a
slack
channel
and
I
think
it
was.
The
slack
channel
was
mentioned
in
the
email.
A
Then
of
course,
we'll
have
YouTube
for
like
meetings
like
this
or
maybe
tutorials
or
hackathons,
where
we
get
together
and
we
solve
some
problem.
So
then
you
know
if
we
can
post
that
to
YouTube
other
people
can
see
it
who
weren't
able
to
make
the
event
and
then
finally
I'm
using
the
node.
Then
so
abraham
mentioned
the
node,
it's
a
blog
for
notifications
so
or
essays.
It's
basically
a
publishing
platform
and
you
can
post
things
and
it
goes
up
to
the
developmental
biology
community.
A
So
that's
a
good
way
to
engage
people
in
developmental
biology
on
these
methods.
You
know
and
and
of
course,
if
you're
interested
in
sort
of
you
know
getting
people
and
than
these
methods,
one
way
is
to
communicate
with
them.
You
know
people
who
might
understand
it
a
little
bit,
but
don't
really
understand
it
a
lot,
and
you
know,
by
engaging
people
in
in
sort
of
the
place
where
they
go
to
look
for
things.
You
know
you
can
get
people
interested,
so
I
think
that's.
A
There
are
probably
more
places
we
can
work,
but
that's
that's
my
philosophy
about
working
together,
so
we
can
work
together,
multiple
ways
and
so
the
structure
of
this
group
over
time-
and
we
can-
you
know,
move
the
meeting
time
if
this
is
an
optimal
for
people
and
a
couple
of
people
say:
Wednesday
didn't
work
but
where's.
It
maybe
still
working
on
that.
But
anyways.
A
The
structure
of
this
group
will
be
we'll,
have
a
meeting
and
then
people
might
want
to
lead
a
discussion
of
some
type.
So
people
would
choose
a
date
to
lead
a
discussion,
so
you
know
you
might
pick
a
paper
to
review
or
you
might
want
to
show
a
demo
of
something
you're
working
on
it
could
be.
Anything
could
even
be
just
like
a
document
or
a
PowerPoint
presentation.
A
If
you
think
that
our
feedback
would
be
useful
to
you
or
we
can
brainstorm
ideas
if
you're
not
really
sure
what
you
want
to
do
or
what,
how
you
want
to
contribute
things
I'm
interesting
to
you,
but
you
don't
really
know
how
to
perceive.
Then
we
can
brainstorm
in
our
group
and
give
you
feedback
also.
We
we
could
do
this
weekly,
but
we
might
end
up
doing
it
by
weekly.
A
We
might
do
like
one
week
or
we
check
in,
and
you
know,
report
on
what
we
you
know
our
progress
on
things
or
you
know
if
we
have
something
interesting
that
we
want
to
share
with
the
group,
so
the
meeting
frequency
is
you
know
could
be
weekly
for
a
while
and
then
bi-weekly
it's
up
to
people
in
the
group,
and
then
we
want
to
at
the
meetings.
We
also
want
to
plan
for
outputs.
A
So
it
ideal
we
like
to
do
something
with
this
group
I
mean
we
like
to
talk
and
interact,
but
we'd
also
like
to
do
things
like.
Maybe
you
know,
review
or
maybe
work
on
papers
or
blog
posts
or
presentations,
and
then
maybe
even
educational
content,
where
we
can
educate
people
say
in
developmental
biology
about
different
types
of
you
know:
machine
learning
techniques
or,
conversely,
we
could,
you
know,
try
to
teach
computer
scientists
a
little
bit
of
developmental
biology
in
ways
that
they
might
not
be
able
to
encounter
in
a
classroom.
A
I've
had
a
lot
of
people
in
our
evil
worm
group,
a
lot
of
computer
science,
people
ask
about
like
things
and
development,
and
sometimes
they're
very
good
resources
and
sometimes
they're
no
good
resources
for
these
things.
So
to
make
like
the
biologically
oriented
people
aware
that,
like
there's
an
opportunity
to
educate
you
know,
computer
science,
people
and
something
in
developmental
biology
that
you
know
really
doesn't
necessarily
exist
in
digital
form
or
you
know
some
easily
accessible
form.
So
this
is
my
contact
email.
A
B
A
We
can
look
into
that
yeah.
Well,
some
of
that
might
be
interesting
to
people
Thank,
You,
Jesse,
so
Jesse
said
no
questions,
but
I'm
excited
about
all
this,
so
yeah
I
mean
again,
like
Jesse
I
think
we
did
this
over
the
summer,
where
yeah
I
check
around
the
node.
That's
an
interesting
place.
They
have
a
lot
of
good
resources
there.
So
yeah.
C
A
If
you
have
an,
if
you
have
just
a
brainstorming
idea
that
you'd
like
to
explore,
you
know
maybe
like
do
a
short
presentation
on
the
idea
like
what
am
I
what's
coming,
you
know,
what's
in
my
head
and
you
know,
maybe
pose
some
questions,
you
know
doing
a
little
bit
structured
way,
but
definitely
something
where
people
can.
You
know
like
well
of
a
machine
learning
person
who
might
say:
oh,
that's
a
good
idea
or
not
and
be
a
biology
persons
as
well.
A
A
A
You
know
maybe
moving
the
time
every
every
no
and
again
two
different
days
or
you
know
we'll
see
what
happens
with
that.
But
why
don't
we
plan
on
about
next
week
at
the
same
time?
Is
that
good
for
everyone
here.
D
A
A
So
it
would
be
like
a
working
document
where
you
know
you
have
a
lot
of
resources
in
the
same
place
and
then
eventually
we
could
hammer
that
out
into
something
like
a
paper
or
you
know
some
sort
of
tutorial.
So
you
know
there
ways
to
use
github
for
that.
The
other
thing
that
even
a
posted
is
this
second
link
and
that's
an
example
actually
of
someone
working
in
deep
learning
for
if
it
gets
microscopy
data
or
it's
a
C
elegans
related
things.
So
this
is
a
an
example
of
deep
learning
in
I.
A
Think
it's
a
network
science
model
so
I
mean
this
is
like
just
an
example
of
what
people
are
doing
in
the
area
and
it's
it's
pretty
like
you
know,
I,
don't
know!
If
you
can
you
really?
If
you
look
at
it,
you
know
you
may
or
may
not
be
able
to
understand
what's
going
on
in
the
repository,
but
you
know
this
is
something
that
maybe
if
we
had
a
paper
to
review,
you
know
the
basic
concepts
it
would
be.
A
A
We'd
like
to
make
it
so
that
people
are
more
comfortable
with
these
things,
so
I
mean
that
that's
basically
one
of
our
goals,
so
yeah.
If
you
have
quick
references,
we
can
post
them
in
the
github
repository
as
well,
so
so,
okay,
well
thanks
for
attending
I,
think
we're
gonna
conclude
things
for
this
week
and
I
just
wanted
to
get
a
I
just
wanted
to
have
an
organizational
meeting,
and
hopefully
we
have
more
people
joining
our
group
and
contributing
and
hope
that
everyone
has
a
good
week.