►
Description
Recap of Neuromatch presentations, non-neuronal cognition in cell colonies, various topics in developmental systems, and papers from the reading queue. Attendees: Susan Crawford-Young, Bradly Alicea.
A
A
C
B
B
A
A
B
A
A
Yeah
yeah,
it's
yeah,
so
I
mean
I
yeah
and
I
made
I
made
a
mistake
in
the
email
I
said:
2
pm
utc,
which
is
of
course
because
it's,
I
think
it
only
changes
for
certain
parts
of
the
world,
not
others,
but
you
know
it's
like.
Everyone
eventually
makes
the
change
to
the
same
time.
I
think
so
it's
it's
there's
a
couple
weeks
in
there,
it's
usually
pretty
chaotic,
where
no
one
knows.
A
B
B
B
C
B
A
Oh
yeah
yeah,
they
send
you
these
matches
and,
like
you
know,
you
can
contact
them
or
not
get
a
contact
point
yet
still,
but
it's
like
yeah,
it's
just
an
opportunity
for
you
to
like
talk
to
people.
You
know
matt
meet
people,
new
people,
you
might
not
meet
otherwise
and
let's
you
know
it's
pretty
fruitful.
You
can
have
a
discussion
about
different
things
and
that
sounds
like
an
interesting
paper.
B
A
Well,
you
can
put
it
in
the
chat,
I'm
just
doing
this
for
people
on
youtube,
so,
if
you're
viewing
this
recording
welcome,
we
have
a
couple
of
things
to
talk
about
today,
so
I
want
to
go
through
some
things
because
we
haven't
had
a
meeting.
You
know
we
had
a
meeting
last
week.
It
was
sort
of
informal,
but
I
wanted
to
go
through
some
things.
We
didn't
talk
about
last
week.
A
A
People
enjoyed
that
it
was
a
a
collection
of
different
thoughts
about
the
differences
between
artificial
neural
networks
and
biological
neural
networks,
specifically
like
the
kinds
of
things
that
artificial
neural
networks
don't
have
the
biological
neural
networks
do
so.
We
know
that,
like
biological
neural
networks,
something
you'd
find
in
an
organism,
a
network
of
a
dense
reticulating
network
of
neurons.
A
Of
course
can
do
a
lot
of
amazing
things
and
while
artificial
neural
networks
can
do
those
too
they're,
you
know,
there's
obviously
a
gap,
and
so
what
is
that
gap
consist
of?
And
so
we've
got
a
couple
of
candidates
in
the
talk?
Okay
and
one
of
them,
one
of
the
more
interesting
ones
I
think
was
where
we
had.
We
talked
about
the
sort
of
the
scale-free
nature
of
a
biological
neural
network.
A
That
is
where
you
have
processing
hubs
in
different
parts
of
the
brain
or
the
network,
and
those
are
both
for
information
processing
purposes
and
for
energetic
purposes
very
efficient,
and
so
your
average
artificial
neural
network
is
very
there's
sort
of
a
random
connectivity
and
then
it
you
know
it.
The
weights
are
shifted
so
that
you
can
find
patterns
and
learn
patterns.
A
You
know.
That's
out
in
the
literature,
then
that's
kind
of
obvious,
but
you
know
to
think
about
that
in
terms
of
a
artificial
neural
network
is
it'd,
be
something
people
don't
do
very
often,
and
so
so
susan
put
her
link
in
the
chat
here.
So
this
is
mechanical.
Surface
waves
accompany
action
potential
propagation,
and
this
is
a
nature
paper
from
2015,
and
this
is
open
source,
so
it's
available
for
free.
A
A
So
yeah
many
studies
have
shown
that
an
act,
a
mechanical
displacement
of
the
axonal
membrane,
accompanies
the
electrical
pulse,
defining
the
action
potential.
Despite
a
large
and
diverse
body
of
experimental
evidence,
there
is
no
theoretical
consensus
either
for
the
physical
basis
of
this
mechanical
wave
or
its
interdependence
of
the
electrical
signal
in
this
manuscript.
We
present
a
model
for
these
mechanical
displacements,
as
arising
from
the
driving
of
surface
wave
wave
modes
and
potential
energy
is
stored
in
elastic
properties
of
the
neuronal
membrane
and
cytoskeleton.
A
A
A
A
A
This
was
a
group
effort,
so
this
is
the
people
who
have
been
talking
about
this
topic
in
this
group,
so
dick
gordon
thomas,
harvick,
jesse,
parent,
esme,
singer
visual
sing
and
then
akshara
gopi
who's,
part
of
our
other
group
and
myself,
and
so
you
know
that
my
philosophy
on
authorship
is
pretty
broad.
It's
basically,
if
you
contribute
something
to
the
discussion
or
the
you
know,
putting
together
the
slideshow
or
whatever
or
if
there's
a
paper
contributing
to
the
paper
in
some
way,
then
you
get
an
authorship
credit.
A
A
Yeah,
so
that's
that's
so
this
is
the
I'm
going
to
go
through
the
presentation
here,
because
I
don't
think
anyone
here
has
seen
it
okay.
So
this
is
this
was
the
so.
The
first
thing
is
we
had
a
couple
of
short
videos
that
I
clipped
from
some
of
the
data
that
we
have.
So
this
is
basil
area
in
the
wild-
and
I
know
maybe
some
of
you
haven't
seen
this
okay
gotta,
let's
restart
this.
D
A
So
this
is
the
isolated
colony,
and
so
this
is
where
it's
like
just
on
a
flat
focal
plane,
and
you
can
see
that
now
it's
doing
this
sort
of
accordion
like
movement,
you
have
a
leading
cell
here,
you
have
a
stationary
cell
here
and
you
can
see
it
sliding
and
you
can
see
different
modes
of
movement.
You
can
see
these
cells
are
kind
of
coming,
bunching
up
and
then
stretching
out
as
they
move
along
and
that'll
be
important
later
in
this
talk
to
understand.
A
A
Okay,
all
right,
I
guess
we're
okay,
now,
okay!
So
then
we
we're
thinking
about
this
in
terms
of
psychophysics,
so
non-neuronal,
cognition
and
I'll
get
what
that
I'll
get
into
what
that
is
in
a
minute.
But
psychophysics
is
this
branch
of
psychology's
branch
of
sort
of
physiology
that
deals
with
how
people
perceive
stimuli
and
you
know
encode
it,
and
then
you
can
measure
it
using
some
sort
of
mathematical
model,
so
there's
a
whole
literature
on
this
in
humans.
Of
course,
you
can
see
someone
in
a
virtual
environment.
A
People
have
also
observed
psychophysical
properties
in
mouse
visual
systems.
So
this
is
a
mouse
focusing
on
a
cat
and
then
focusing
on
another
mouse.
You
see
in
goldfish
there's
been
work
done
on.
I
think
color
vision,
so
they've
done
a
lot
of
they've
sort
of
extracted
a
lot
of
psychophysical
laws
and
goldfish,
and
then
the
question
is:
can
we
also
observe
things
like
this
in
diatoms?
A
So
this
is
the
slide
that
I
used
for
the
short
introduction,
so
they're
algae,
so
they're
single
cell
eukaryotes,
but
they're
not
really
animals.
They
have
plant-like
characteristics,
so
they
have
chlorophyll
elements
and
a
silica
cell
wall.
So
it's
very
much
unlike
animals,
but
they
do
have
a
lot
of
sensory
behaviors
much
like
plants
but
they're,
not
they're,
they're,
sort
of
on
the
border,
they're
sort
of
an
early
branch
of
evolution.
A
So
you
have
these
diatoms
they're,
basically
their
glass.
I
mean
they're
silica
cell
walls
and
they
come
in
a
bunch
of
different
shapes
and
so
in
this
talk
we'll
be
focusing
on
this
rapha
diatom.
So
this
rapid
class
of
diatom
is
you
have
a
bunch
of
you,
know,
properties
of
the
cell.
You
have
valves,
and
you
have
these
girdle
bands
and
you
have
all
these
properties.
A
A
So
it's
a
very
small
time
scale.
So,
needless
to
say,
they
don't
really
have
nervous
systems
because
there's
no
room
for
neurons
in
this
system.
You
just
have
a
bunch
of
these
rapha
diatom
cells,
at
least
for
the
the
genus
we're
going
to
be
talking
about,
and
so
this
is
an
analysis
again
of
bachelorion
culture.
So
this
is
the
again
a
repeat
of
the
video
that
I
showed
before
and
notice
a
couple
things.
One
is
that
you
have
this
accordion-like
movement
and
we
can
model
this
as
a
series
of
oscillators.
A
So
there
are
no
neurons
here,
but
it's
exhibiting
some
sort
of
complex
behavior,
collective
behavior
and
the
question
is:
what
is
this
behavior?
Is
it
generated
by
it's
not
generated
by
a
brain,
it's
generated
by
something
else
and
there's
information
processing
going
on,
because
we
know
that
it
does
respond
to
environmental
cues?
So
how
do
we
characterize
that?
A
A
A
So
if
you
know
about
the
way
people
kind
of
view
cognition,
it's
largely
information
processing,
and
so
this
is
consistent
with
this,
but
non-neuronal
cognition
doesn't
involve
neurons.
So
what
does
it
involve?
Well
in
a
baluska
and
levin
in
this
paper
they
talk
about
different
ways,
different
places.
A
This
refers
to
the
physical
traces
that
can
be
referenced
and
read
in
the
future,
so
in
ant
colonies,
for
example,
you
have
ants
moving
around
and
they're
navigating
a
landscape
and
they're
making.
You
know
they're
wandering
around
trying
to
find
food
and
they
can
form
these
trails,
and
the
trails
are,
of
course
not.
You
know
they
don't
necessarily
follow
any.
They
follow
cues,
but
they
do
more
than
that.
They
lay
down
pheromone
and
then
other
ants
will
follow
the
pheromone
of
their
con
specifics
and
eventually,
over
time.
D
A
Guide
them
to
these
resources,
so
one
ant
might
find
a
resource
somewhere.
They'll
lay
down
some
pheromone
along
a
trail
going
to
that
resource.
Other
ants
will
follow
and
reinforce
the
trail
and
then
you'll
get
these
ant
trails
and
it's
collective
behavior
that
re,
you
know,
there's
a
brain
generating
it,
but
it
doesn't
require
our
brain
to
maintain
it
necessarily.
A
So
that's
what
they
mean
by
stigma
g.
So
this
is
an
example
of
the
extracellular
matrix
in
vessel
or
in
diatoms.
I
think
this
is
just
kind
of
like
a
generic
example.
These
are
secreted
polymers
that
completely
encase
the
cell
and
are
responsible
for
adhesion
and
interactions
with
the
external
environment.
These
are
called
misologe
trails
and
strands.
A
So
you
have
these
things
that
these,
I
guess
they're
polymers,
that
join
the
cells
together
or
they
link
the
cells
when
they're
decoupled
and
they
they
provide,
maybe
some
sort
of
memory
or
some
sort
of
way
of
mediating
those
collective
behaviors
by
the
individual
cells.
If
you
go
back
and
you
look
at
the
video
that
we
showed
here,
you'll
see
that
the
cells
will
stretch
out
and
then
they'll
they'll
be
able
to
stretch
against
one
another,
but
they
don't
break
that
bond
they're,
always
like
sort
of
touching
one
another
at
some
point.
A
So
this
immediate
this
ex
the
extent
of
the
stretch,
for
example,
you
see
a
little
bit
of
the
cell
in
contact,
and
so
the
question
is:
why
doesn't
it
fly
off
the
end
of
the
cell,
like
when?
Does
one
cell
not
fly
off
the
end
of
the
other
cell
and
break
the
colony
apart?
Well,
it's
because
of
these
mesolage
trails,
they're
kind
of
sticking
together,
and
so
it's
not
a
not
completely.
A
A
Sensory
input,
so
there's
the
sensory
world.
You
have
the
you
know
you
might
find
these
diatoms
in
a
marine
estuary
or
you
might
find
them
in
a
river
or
a
lake,
or
you
know
somewhere
in
a
pelagic
environment
where
you
know,
there's
a
lot
of
biodiversity
but
they're
they're
sort
of
in
the
water
there.
A
One
of
them
is
the
elevated
levels
of
co2
or
just
fluctuations
in
co2.
So
we
know
that
like,
for
example,
diatoms
send
co2
and
they
can
sense
concentrations
of
it
more
specifically,
and
that
affects
the
growth
and
so
the
growth
of
the
colony,
the
growth
of
the
cells,
relies
on
the
co2
concentration,
and
so
when
there's
high
co2,
for
example,
it'll
reduce
grazing
pressure
and
it
has
an
effect
on
photosynthesis.
A
We
also
deal
with
water,
fluid
shear,
fluid
motion,
and
so
there
have
been
a
couple
of
studies
looking
at
that,
especially
in
terms
of
perception,
so
there's
decision
making
physiology
and
perception,
so
we're
not
just
making
this
up
that
there's
like
something
going
on
there.
People
are
talking
about
the
single
virtue
and
diatoms,
have
sensory
systems
for
detecting
and
responding
to
fluid
motion,
osmotic
stress
and
iron
abundance.
So
you
know,
like
think
about,
like
that
water
column.
Again,
you
know
there's
sheer
going
on.
A
A
So
you
get
these
chemical
signals
and
they
get
modified
by
light
intensity.
You
can
imagine.
Seasonality,
of
course,
is
a
context
in
which
they
might
experience
these
kind
of
fluctuations
also
day
night
cycles.
So
there
are
a
lot
of
opportunities,
probably
for
to
observe
these
different
things
in
nature
and
of
course
we
can
implement
them
in
the
lab
artificially
as
well.
So
co2
and
light
availability
affect
diatom
physiology
and
they
in
this
paper
they
tested
three
light
intensities
and
they
found
that
low
light
intensity
affects
the
physiology
and
distinct
physiological
traits.
A
A
A
So
it
turns
out
that
there's
a
lot
of
stuff
that
can
potentially
be
varied
and
you
can
look
at
the
how
diatoms
respond
now.
The
question
is,
of
course,
is
how
do
we
model
this
behavior
in
an
ideal
fashion
and
then
what
happens
when
it's
exposed
to
these
different
fluctuating
environmental
factors?
So
the
first
thing
we
have
is
this:
what
we
call
collective
pattern
generator-
and
this
is
a
model
that
is
based
on
a
central
pattern-
generator
neuroscience.
D
A
You
know
these
are
kind
of
arbitrary
locations
of
these
actin
meiosis
complexes,
and
then
we
can
link
them
with
this
sort
of
connectionist
framework,
where
each
link
between
the
two
nodes
and
different
cells
are
labeled
with
sort
of
a
weight,
and
so
we
can
weight
these
connections
to
make
them
either
excitatory
or
inhibitory.
So
they
can.
There
can
be
a
gain
or
there
can
be
like
a
latency
between
the
two
things
and
that
will
govern
sort
of
the
movement.
A
So
we
can
use
a
series
of
coupled
sinusoids
to
model
the
ideal
copg.
So
each
of
these
cells
is
one
of
these
trajectories
and
then
we
can
have
these
overlapping
sinusoids
depending
on
sort
of
the
frequency
of
movement
between
the
cells.
So
you
know
like
an
idealized
model,
would
be
where
you'd
have
something
at
like
half
phase
synchronization.
A
So
there's
a
lag
between
cell
one
and
cell
two,
but
these
can
be
you
know
these
can
be
tightened
up.
You
can
have
quarter
phase
or
you
know,
maybe
like
10
degrees
out
of
phase
depending
on
you
know
what
your
your,
depending
on
environmental
effects
or
what
the
model
system
should
look
like,
and
then
you
can
also
add
noise
into
the
signal,
so
you
can
simulate
discoordination.
A
So
if
you
have
overlapping
noisy
signals,
you
know
that
makes
it
much
harder.
So
you
can
just
think
of,
like
the
cell
may
be
like
moving
around
a
little
bit
like
stochastically
in
space
and
then
moving
against
one
another
stochastically,
which
would
be
what
this
is.
And
so
we
can
do
a
lot
of
this
like
numerically
and
then
plotting
out
like
functions,
and
we
can
think
of
it
that
way.
A
We
can
also
think
of
this
in
terms
of
like,
if
you
have
an
environmental
input
to
the
system,
maybe
on
one
end
of
it,
you
know
what
does
the
movement
look
like
if
you
change
the
weights,
for
example?
So
if
I'm
putting
like
light
sources
into
these
different
nodes,
I'm
trying
to
like
simulate
like
light
intensity?
A
A
Have
a
signal-to-noise
ratio,
so
you
can
have
you
know
you
have
these
forces,
but
again
you
don't
know
like
they're
forces
that
are
noisy
versus
forces
that
are
like
informative
to
the
organism
like
if
it's
some
shear
that
exists,
because
maybe
the
seas
are
rough.
Is
that
different
from
something
that,
where
it's
just
kind
of
like
some
other
type
of
sheer
force
that
I
might
experience,
maybe
you
know
there's
a
a
large
predator
coming
nearby
and
they
don't
really
bother
vessel
area,
but
that
might
be
a
signal
for
something.
A
A
D
A
Of
things
and
look
at
what
the
outcome
is
so
then
that
was
the
end
of
the
talk
and
I
point
out
thomas
harvick's
observation
of
dietarium
site,
which
is
nice.
He
has
a
lot
of
stuff
on
diatoms
there
on
his
observations
of
the
of
the
colonies,
so
he's
helping
us
with
some
of
the
observations,
empirical
observations
and
then
I
put
in
a
plug
for
the
divorm
groups.
A
I
think
that'll
be
something
that
we
can
explore
later
in
terms
of
how
to
actually
write
up
a
paper
on
this,
so
I've
proposed
a
couple
of,
or
we've
proposed,
a
couple
of
models,
potential
models
here
that
might
be
you
know,
informative,
maybe
not
just
for
diatoms
but
for
the
whole
enterprise
of
you
know,
collect
you,
know,
cell
collectives
or
cell
colonies,
and
how
they
respond
to
different
things
in
nature.
So
might.
B
There
I
turned
myself
back
on
yeah,
I'm
I'm
curious
about
how
big
they
are
and
whether
I
can
put
them
in
my
new
ball
microscope
and
see
them
because
then
you
could
observe
them.
3D!
Oh
wow
yeah.
If
you
know
the
microscopes
I've
got,
are
I
don't
know
they?
You
can
get
one
millimeter
field
of
view.
So
I
wonder
if
you
could
see
them
in
that?
That's
as
narrow
focuses.
I've
got
currently.
A
A
A
Well,
I
mean
you,
you
go
and
you
get
a
sample
of
them
and
I
think
you
can
get
them
in
like
a
river
or
like
a
maybe
a
brackish
water
source
or
somewhere
like
that,
and
you
just
collect
them
like
collect
a
water
sample
and
you
and
you
look
for
these
colonies
and
then
you
can
isolate
them
into
a
culture
dish
and
the
culture,
I
think,
is
there
some
nutrient
mix
that
they
use
these.
You
know.
A
B
A
I
don't
know
I
mean
like,
I
think
they
probably
do,
maybe
in
the
summertime,
not
in
the
wintertime,
but
I
I
know
like
germany,
they
have
them
in
florida,
so
they're
they're
kind
of
like
c
elegans,
where
you
find
them
all
over
and
they're
in
the
soil
and,
like
you
know
you
can
I've
like.
I
did
c
elegans.
I
I've
isolated
some
c
elegans
here
where
I
live,
and
I
know
you
can
do
it
all
over
the
world
and
you
can
find
different
like
whale
type
strains
all
over
the
world.
A
Like
you
know
a
lot
of
stuff,
that's
up
and
characterized.
So
that's
that's
an
issue
like
you
know.
You
don't
really
know
what
the
exact
with
bacillaria
it's
just
basically
with
it's
like
basically
a
genus
they
don't.
They
have
species
divisions
but
they're
a
little
tricky
to
to
define,
but
I
mean
they
shouldn't
be
too
hard
to
find.
The
question
is,
like
you
know,
I
think,
getting
finding
identifying
the
right
diatoms
that
maybe
you
want
to
look
at
and
getting
them
maintaining
them
in
culture.
A
A
Oh
well,
you
can
find
c
elegans,
you
can
harvest
them
from
like
a
compost,
pile
if
you
have
that
or
like
the
soil
somewhere,
where
it's
like
cool
and
wet.
Like
you
know,
in
an
area,
that's
wooded,
so
you
go
to
the
soil
and
you
take
a
sample
of
the
soil
and
you
you
put
it
in
a
culture
dish
like
an
egg
or
plate
just
little
samples
of
soil
and
you
pack
them
down.
A
And
so
then
you
have
to
take
like
a
pit,
a
worm,
what
they
call
worm
pick,
which
is
a
little
metal
thing,
and
you
have
to
pick
them
out
of
the
agar
plate
and
put
them
in
a
new
plate
with
food
which
is
like
a
medium
of
bacteria,
and
then
you
can.
Let
them
sort
of
you
know,
isolate
them
on
those
plates.
And
then
you
end
up
with
a
bunch
of
nematodes,
but
you
don't
know
which
one
is
c
elegans,
the
ones
that
are
c
elegans
will
self-reproduce.
B
A
Yeah
and
I
have
a
protocol
for
this
on
the
website,
so
it's
like
a
it's
a
nice
thing.
A
nice
thing
would
be
a
nice
thing
to
do
for,
like
someone
is
trying
to
learn
biology
and
how
to
isolate
like
different
things
and
put
them
under
the
microscope.
But
it's
yeah.
It's
definitely
not
too
hard.
I
mean
it's
hard,
but
it's
not
like
so
difficult.
It's
not
like
a
molecular
protocol,
but.
A
I
did
that
I
did
that
once
and
I
wrote
it
up
and
it's
pretty
good
yeah,
so
you
can
and
you
can
observe
them
and
that's
why
they
use
c
elegans
as
a
model
organism,
because
you
can
find
it
everywhere.
So
where
do
you
find
a
worm
pick?
Oh
people
make
them
usually
like
they'll,
take
like
a
one
of
these
glass
pipettes
and
they
break
the
tip
off,
and
then
that
leaves
you
a
little
hole
at
the
end
of
the
like
the
fat
part
of
the
glass
pipette.
A
And
then
you
take
a
piece
of
metal
or
little
piece
of
metal
wire
and
you
stick
it
in
there
and
you
kind
of
close
it
up
with
under
using
a
bunsen
burner.
You
close
it
up.
You
melt
the
glass
around
it
and
then
you
like
tap
the
end
of
that
the
other
end
of
that
glass,
wire
or
the
metal
wire,
so
that
the
end
is
flat
with
like
a
hammer.
You
tap
the
end
of
that
with
a
hammer
and
it
becomes
flat
and
then
you,
then
you
have
something
that
you
can
scoop
up.
A
A
B
A
Yeah
well
so
that's
yeah,
that's
psychophysics
of
non-neural
cognition
next
thing
I
guess
I
was
going
to
talk.
I
was
hoping
krishna
would
make
it
to
the
meeting.
So
we
could
talk
about
this
paper
that
he's
writing.
So
he's
writing.
His
paper
reinforcement
learning
algorithms
sarsa
unexpected
sarsa,
so
this
is
something
he
wrote
he's
in
the
process
of
writing
it.
A
A
It'll
choose
from
different
policies
which
are
ways
to
like
you
know,
be
rewarded
for
certain
behaviors
and
it'll
end
up
learning
things
in
this
way,
and
so
reinforcement
learning
focuses
on
the
sequence
of
decisions
to
maximize
a
reward
which
encodes
the
success
of
the
outcome
of
an
act
and
achievable
in
an
uncertain,
potentially
complex
environment.
So
we
talked
about
reinforcement,
learning
and
one
of
the
divorm
ml
lectures,
there's
a
lecture
on
it
and
so
he's
writing
this
paper
on
temporal
difference.
A
Learning
approaches
which
is
something
that
is,
we
didn't
really
talk
about
in
the
vassal
area
or
the
diatom
talk,
but
we,
this
is
something
that
you
find
in
a
lot
of.
Maybe
biological
nervous
systems
and
they've
tried
to
model
it
in
artificial
neural
networks.
So
it's
basic!
It's
it's
what
it
it's,
how
it's
described
so
temporal
difference.
Learning
is
where
you
look
at
like
a
sequence
of
quantities,
and
you
try
to
sort
of
extract
a
trend
out
of
that
sequence
by
looking
at
the
differences
between
the
the
selection
of
things
that
you
presented
with.
A
So
if
you
know
you
had
like,
for
example,
a
series
of
days
where
it
either
rained
or
didn't
rain,
and
you
wanted
you
observed
those
days
and
you
said
it
all
right
it
rained
today,
but
it's
not
going
to
rain
tomorrow
and
then
it
rained
the
next
day
it
won't
rain
the
next
day.
So
what
what
should
I
expect
for
the
future?
Can
you
assemble
like
a
forecast
of
when
it's
a
when
it
should
rain
or
when
it
shouldn't
rain?
A
As
you
can
imagine,
that's
a
little
bit
difficult
to
do,
but
you
can
do
you
know
you
can
do
some
sort
of
difference.
Learning
on
that
temporal
sequence
to
sort
of
get.
You
know,
information
about
the
intervals
and
other
things
that
you
might
need
information.
You
might
need
to
know
to
make
a
better
prediction,
and
so
you
get
you
know
you
do
this
over
time.
You
revise
your
sort
of
model
of
what
you
know
what
you
should
expect
and
it
should
bring
you
nearer
to
the
prediction:
a
successful
prediction:
that's
the
idea.
A
So
this
is
what
he's
going
to
talk
about
this
method,
and
this
is,
of
course
your
reinforcement
learning
model.
You
have
an
agent
and
it
interacts
with
its
environment
through
an
action,
so
the
agent
engages
the
environment
by
observing
it,
maybe
and
then
there's
a
state
that
it
observes
and
a
reward
for
doing
something
in
that
environment,
so
it
observes
the
world
and
it
either
makes
some
decision.
A
A
So
you
know
there
might
be
a
tone
when
there's
a
food
pellet,
so
you
know
every
time
there's
a
tone
with
a
food
pellet,
the
rat:
either
you
know,
selects
the
pellet
or
it
doesn't
and
then
there's
the
state
of
you
know
whether
it's
been
rewarded
or
not
and
there's
a
reward.
It
may
not
be
rewarded
or
may
be
rewarded
for
doing
some
sort
of
behavior
and
that's
how
it
works,
and
now
I'm
not
really
making
that
wasn't
the
most
elegant
description
of
that.
But
you
know
it
in
reinforcement
learning.
A
Yeah
yeah,
it's
they've,
yeah,
they've
developed
quite
a
terminology
for
it
so
and
that's
that's
a
whole
another
thing,
but
this
so
yeah.
This
is
gonna,
be
a
nice
paper
when
it's
finished,
I'm
not
sure
when
we're
working
on
it.
So
I
don't
know
like
how
it
applies
necessarily
to
what
we're
doing
directly
in
the
group.
But
I
you
know
it's
something
that
is
sort
of
part
of
the
machine
learning
mission
here,
so
that
would
be
interesting
to
see
what
he
does
with
it.
A
As
for
reinforcement
learning,
we've
done,
we've
talked
about
this
individual
ml
a
bit.
So
if
you're
interested
in
sort
of
what
people
are
doing
in
reinforcement,
learning,
there's
some
lectures
there
to
look
at
there's,
also
things
online
that
we've
talked
about,
that
they
give
you
tutorials
in
terms
of
like
how
to
implement
it
in
a
machine
or
a
robot,
especially
so
there's
that
I
just
wanted
to
also
remind
us
that
this
then
mainly.
A
This
is
for
myself
that
this
journal
of
open
source
science,
article
is
still
outstanding
and
I
have
not
had
a
chance
to
really
kind
of
work
on
it
because
I've
been
busy
with
neuromatch
and
some
other
things.
So
this
is
the
next
sort
of
one
of
the
next
things
I
need
to
do,
and
this
is
just
again
going
through.
This
is
from
last
summer's
from
the
devo
learn
implementation.
So
this
is
the
actual
program
and
we're
going
to
be
working
on.
I
just
want
to
add
some
more
a
couple,
more
things
to
it.
A
So
it's
probably
even
I'm
just
kind
of
holding
things
up
here.
I
need
to
get
on
it
and
finish
it
off.
I
think
the
thing
I
I
was
thinking
of
doing
was
to
kind
of
give
maybe
a
little
bit
more
information
about
like
how
it
works
and
how
it's
integrated
into
this
larger
evil
learn
systems
or
this
divalern
effort.
So
I'm
going
to
do
that.
A
A
A
A
So
in
this
in
this
topic,
which
I
actually
like
this
topic,
we
haven't
talked
about
it
that
much
in
the
group
they
talk
about
reaction
norms,
which
are
these
reaction
norms,
are
means
of
conceptually
graphically
mathematically,
describing
this
total
variance
of
genetic
variability
environmental
variability
in
the
interaction
of
them
as
a
powerful
tool
for
decomposing
it
into
its
constituent
parts.
Nature
nurture
critically
their
interaction,
so
reaction.
A
A
But
then,
after
you
take
away,
the
starvation
it'll
continue
to
grow
and
depending
on
the
genotype,
the
growth
is
either.
You
know
it
either
starts
back
at
a
linear
rate
of
growth,
or
it
starts
back
at
a
suppressed
rate
of
growth,
where
sometimes
it
will
try
to
make
up
for
lost
time
and
have
an
enhanced
growth.
And
so
you
can
measure
this
by,
like
looking
at.
A
A
I
talked
about
this
in
the
I
gave
a
talk
to
the
virtual
worm
group
this
last
spring
and
I
talked
about
this
work
in
doing
this
in
c
elegans.
So,
if
you're
interested,
I
can
send
you
that
link,
but
in
reaction
norms
are
a
very
powerful
way.
I
think
of
dealing
with
this
issue
of
sort
of
what
is
you
know,
what
is
environmental
plasticity
versus?
What
is
you
know,
encoded
sort
of
in
the
genome
and
then,
when
you
have
a
defined
genotype,
you
can
actually.
A
A
It
could
be
that
the
organism
doesn't
respond
to
it
at
all,
but
then
there's
a
response
coming
away
from
that,
which
is,
it
could
be
like
making
up
for
a
lost
time,
which
would
be
some
sort
of
exponential
growth.
It
could
be,
you
know,
linear,
which
means
it
just
continues
its
growth,
but
it
doesn't,
but
it
is
affected
by
that.
So
it's
basically
you
know
it
suppresses
growth.
But
after
you
take
away
the
stress,
growth
continues
linearly
or
just
totally
destroys
the
growth
potential
of
the
organism
altogether
and
c
elegans
are
different.
A
Depending
on
your
mutant,
I
can
report
that
there
is
that
difference,
and
so
you
know
basically
the
mutate,
the
the
defined
mutation
is.
You
know
some
functional
aspect
of
the
of
the
organism.
It
could
be
metabolism,
it
could
be.
You
know
some
sort
of
other
developmental
thing
that
you're
taking
away,
and
so
then
that
has
an
effect
on
how
the
genotype
responds
to
environment.
A
So
it's
a
really
interesting
topic
in
and
of
itself,
and
so
this
actually
another
word
on
this
is
that
I
know
some
of
the
people
in
the
group
are
interested
in
plasticity,
and
so
this
is
actually
very
tied
very
tightly
to
plasticity
and
understanding
how
that
works
in
a
in
a
model
system.
So
you
know,
if
you
want
to
talk
about
you
know,
developmental
plasticity,
for
example,
reaction
norms
are
a
part
of
that
understanding,
sort
of
the
plastic.
You
know
plasticity
of
an
organism.
A
D
A
So
this
is
about
like
early
organisms
before
the
cambrian
explosion,
or
you
know
maybe
right
around
that
same
time,
why
they
sort
of
evolved,
kept
evolving
into
crabs,
and
so
you
had
these
like
different
shapes
of
things
that
were
existing
in
the
ocean
at
the
time,
and
why
did
they
keep
evolving
into
crabs?
And
the
basic
argument
here
is
that
there
is
this
shape
economy
of
shape.
A
So
crab
is
like
what
they
refer
to
as
crabs
is
like
a
circular
or
a
spherical
type
phenotype,
and
so
you
know
why,
instead
of
like
evolving
this
long
phenotype,
they
keep
evolving
into
crabs.
It's
kind
of
an
interesting
video-
and
I
have
it
in
this
folder-
I'm
not
going
to
play
the
whole
thing,
because
it's
quite
long,
but
it's
an
interesting
question.
I
I
don't
know.
B
D
A
A
That
is,
you
know.
What
do
the
organisms
look
like
and
the
question
is,
I
don't
really
know
I
don't
know
if
people
know,
but
we
can
kind
of
look
at
the
phylogeny
and
take
a
gas,
and
then
you
know
again
related
to
that
is.
Why
would
you
get
like
these
different
types
of
development,
they're,
just
kind
of
like
emerging?
Why
would
they
emerge,
and
you
know,
because
they're
obviously
emerged
very
far
back
in
phylogeny,
because
you
know
they're
they're
sort
of
some
of
the
basic
modes
of
development.
A
A
A
So
yeah,
that's
an
interesting.
I
mean
it's
just
kind
of
one
of
those
things
that
I
ran
across.
I
thought
it
was
interesting
for
the
group
there's
a
paper
that
you
might
want
to
check
out.
This
is
evolving
l
systems
in
a
competitive
environment,
so
l
systems
are
something
that
lindemeyer
systems
are
interesting,
like
they're
computational
models,
they're.
Basically,
where
you
have
this,
it's
a
symbolic
system
where
you
can
put
in
different.
A
A
The
diffusion,
limited
aggregation
models,
except
it's
using
a
little
bit
different
methodology.
It's
not
using
a
random,
like
you
know,
like
things
attracted
to
other
things
and
sticking
together.
This
is
where
you
have
an
initial
condition,
and
then
you
apply
production
rules,
sort
of
recursively
and
then
you
watch
the
thing
grow
and
because
it's
a
symbolic
system,
they're
not
right.
Random
rules,
they're
they're,
almost
like
grammar
rules,
and
so
you
can
build
visualizations
out
of
this
and
they
use
linda
meyer
systems.
A
A
lot
to
model
plant
growth
so,
like
you
can
watch
like
you,
can
build
root
systems
and
branching
systems
using
l,
l
systems.
So
this
this
paper
kind
of
talks
about
using
l
systems
in
a
competitive
environment.
So
this
is
an
example
of
a
simulation
where
you
know
they
look
at
the
importance
of
competition.
A
So
again,
they're
modeling
these
little
plant
systems
and
they
actually
can
replicate
some
stages
of
the
evolutionary
process.
So
they
can
look
at
leafless
agents.
They
can
look
at
different
types
of
agents
producing
multiple
seeds,
so
you
can
start
off
at
a
very
low
level
of
complexity
and
build
up
to
agents
that
are
more
complex,
and
so
this
is
a
an
environment.
This
is
an
environment
resulting
from
a
simulation
with
a
high
growth
density
factor.
So
again
you
can
play
around
with
the
parameters
and
you
can
get
you
can
get.
You
know
variation.
A
A
You
know
different
types
of
agents,
so
some
of
these
agents
are
brown
spheres.
Some
of
them
are
these
long
tall,
green
things,
and
so
they
they
compete
for
tallness.
Basically,
they
become
taller
as
the
competitive
process
moves
on,
and
that's
because
they're
competing
for
sunlight
because
they're
more
of
them
and
it's
it's
a
more
complex
ecosystem.
A
A
So
this
agent
is
evolved
in
an
environment
where
the
light
shines
only
directly
down.
So
in
this
case
here
you're
getting
sunlight
from
one
direction
only,
and
so
that
affects
the
way
it
grows,
and
that's
interesting
in
light
of
the
stuff
that
we
were
talking
about
with
diatoms
is
that
you
can
do
this
with
light.
You
know
light
sources,
you
can
simulate
a
colony
or
something
and
how
it
responds
to
light.
A
So
again,
this
is
something
that
you
can
look
at
later,
if
you're
interested
and
then
there
are
a
couple
other
papers,
I'm
not
going
to
talk
about
there's
this
one.
That
was
this
paper
that
dick
sent
me
on
from
molecular
systems
biology.
This
is
deep
cycle,
so
this
is
something
that
reconstructs
a
cyclic
cell
cycle
trajectory
from
unsegmented
cells,
cell
images
using
convolutional
neural
networks.
D
A
A
Deep
cycle
was
evaluated
on
2.6
million
single
cell
microscopy
images,
and
so
they
what
they
did
is
they
visualized
different
cell
and
molecular
components
that
are
involved
in
cell
cycle,
using
fluorescent
markers
and
so
those
fluorescent
markers
and
retract,
and
they
were
able
to
sort
of
characterize
cell
division
in
the
cell
cycle.
So
the
cell
cycle
was
this
very
specific
thing
in
biology
where
a
cell
will
go
from
a
single
cell
to
two
cells.
A
A
Yeah
yeah,
so
they
validated
the
deep
cycle
trajectories
by
showing
its
nearly
perfect
correlation
with
real-time,
measured
images.
So
they
were
looking
at
real-time
images
and
they
used.
They
used
those
real-time
measurements
and
they
correlated
with
they
did
with
deep
cycle,
and
then
this
is
the
first
model
that
resolved
the
closed
cell
cycle
trajectory,
including
cell
division
solely
based
on
unsegmented
microscopy
images.
So
they
didn't
have
to
segment
the
cells.
They
just
had
to
follow
the
fluorescent
markers
and
they
were
able
to
do
this.
A
A
They
use
a
umap
method,
to
sort
of
characterize
of
you
know
the
different
phases
of
cell
cycle
and
where
they
lie
in
the
space
and
so
they're
able
to
separate
out
g0
and
g1
from
g2
and
m,
and
that's
basically
how
they
did
it
without
getting
into
too
much
technical
detail.
But
if
you
again,
if
you
want
to
read
this
paper,
it's
in
the
folder
here
so
yeah
any
any
thoughts
about
that
susan
other
than
the
axolotl
example.
B
If
you
had
those,
then
you
could
do
fluorescent
imaging
of
the
whole
surface
of
that,
the
egg
dividing
and
initially
at
least
up
until
wow
six
cell
divisions,
you're
that's
what
you're
looking
at
is
simply
the
cell
divisions
that
are
happening
like
there's
nothing
happening
underneath
until
a
little
later,
so
very
early
on
in
development
you're.
Just
seeing
this,
they
all
of
the
cells
that
are
dividing.
A
A
You
know
that
that's
going
to
break
the
algorithm
because
it
doesn't
have
the
ability
to
I
mean
you
can
train
it
on
cells
that
are
dividing,
but
you
know
it's
kind
of
hard
to
know
it
can't
break
the
algorithm
in
terms
of
like.
Is
this
a
cell
or
not
or
is
this?
Are
these
two
separate
cells,
or
so
you
know
I
like
the
way
they
do
this
with
markers
of
cell
division
and
being
able
to
predict.
A
That
being
said,
it's
not
exactly
what
we
want
to
do
with
embryos,
but
yeah.
It's
a.
I
think
it's
a
step
towards
understanding
that
it's
really
good,
so
yeah.
So
again,
that's
all
in
that
folder.
I
can
make
that
folder
available
to
people
if
they
want
to
email
me
or
if
susan
has
the
link
there
and
it
should
be
shareable.
A
Otherwise,
I
think
that's
it
for
today
thanks
susan
for
attending
and
this,
so
we
just
changed
time
our
time.
So
we
chant
and
change
the
utc
time
of
the
meeting,
and
hopefully
people
can
make
it
in
subsequent
weeks
know.
The
last
couple
weeks
have
been
a
little
rough
because
we've
had
things
going
on
and
but
I
think
next
week
we'll
have
an
update
on
maybe
some
projects.
I
think
those
wall
might
be
able
to
attend
next
week
and
give
an
update
on
they.