►
From YouTube: (2/5) Michael Berry. Plenary Lecture 3, LACONEU 2010
Description
Lectura Plenaria dictada por Michael Berry (Princeton U, USA) en el Instituto de Sistemas Complejos de Valparaíso (ISCV, www.iscv.cl), el día 22 de enero de 2010, en el marco de la VIII Escuela de Verano en Sistemas Complejos LACONEU 2010 (Latin American Summer School in Computational Neuroscience and Biomedical Applications). Más información: http://www.cnv.cl/laconeu2010/laconeu2010.htm
A
Some
variability
okay,
so
just
to
kind
of
characterize
the
total
redundancy
in
the
population.
One
of
the
things
you
can
ask
is
that
if,
if
you're,
if
you're
one.
A
A
Drops
off
at
you
know,
250
microns
away,
and
so,
if
you
just
add.
A
Of
course,
the
the
level
that
redundancy
is
is
not
terribly
high,
so
this
is
the
average
redundancy
for
cells
that
that
are
already
redundant.
It's
about
five
or
ten
percent,
okay
per
pair
okay.
So
so
it's
not
a
huge
effect
for
pairs
of
neurons,
but
but
basically
because.
A
It
ends
up
being
a
really
large
effect
in
the
population,
so
one
way
of
quantifying
that
is
to
calculate
what
we
call
the
over
representation
factor.
So
the
idea
is,
we
take
the
you
know:
here's
cell
a
we
take
the
raw
redundancy
with
all
its
neighbors
b.
Add
that
up
over
all
the
neighbors
and
divide
by
cell
a's,
visual
information.
So
when
cell.
A
You
know
how
what's
the
total
redundancy
with
neighbors,
and
so
that's
that's,
measuring,
given
cell
a's
information.
How
many
times
over
is
that
re-represented
by
other
neurons,
somewhere
on
on
the
retina
okay
average
over
all
the
cells
a
and
you
get
about
is
what
the
over
representation
factor
is,
and
I
just.
B
A
Point
out,
you
know
this:
isn't
a
measurement
of
redundancy
in
the
entire
population
that
calculation
is
is
very
hard,
but
in
in
kind
of
under
some
some
very
simple
assumptions:
kind
of
uniform
correlations.
You
can.
You
can
connect
this
to
the
population
information
and
if
these
assumptions
are
true.
B
A
This
over
representation
factor,
so,
roughly
speaking,
you
know
it's
reasonable
to.
A
Roughly
a
factor
of
10.
okay,
so
so
that's
really
pretty
high,
and
so
so
it
leads
to
kind
of
a.
A
Situation,
you
know
when
we
look
at
single.
A
A
lot
of
information
contained
in
in
each
spike
and
they're
sort
of
efficient
in
in
a
variety.
A
To
what
happened
to
this
single
neuron
efficiency-
and
you
know
just
to
just
to
quickly
point
out.
A
A
lot
of
you
are
familiar
with.
You
know,
ideas
about
redundancy
reduction
as
one
of
the
purposes
of
retinal
processing.
You
know
this
has
been
a
very
influential
idea
for
the
last
50
years
and
at
least
at
the
level
of
the
entire
population.
It's
just
completely
wrong.
That's
that's
not
what
the
retina
is
is
doing,
and
you
might
wonder
why
you
know
redundancy
reduction
seems
like
like
a
good
idea.
You
get
a
more
efficient
code,
so
so
why
is
this
code
so
so
inefficient?
Actually,
okay,
so.
A
Redundancy
is
is
really
a
beneficial
property
of
the
population
code
and
it
allows
the
population
code
to
be
fast,
unambiguous
and
and
low
error.
Okay.
So
let
me
let
me
kind
of
unpack
this
argument
a
little
bit
more.
A
About
about
vision
is
that
the
way
that
you
use
it,
you
know
kind
of
every
day
vision
is
extremely
deterministic
and
reliable.
You
know.
So,
if
I
do
something
like
you
know,
I
say
you
know
how
many
fingers
am
I
holding
up.
You
know:
what's
the
chance,
you
got
that
wrong.
A
Essentially
zero,
and
you
know
you
do
things
like
you
walk
out
into
the
street.
You
know
you,
look,
you
see,
there's
no
trucks
coming,
you
walk
you,
you
sort
of
you're,
trusting
that
that
visual
discrimination
well
enough
that
you
know
you
could
be
killed.
If
you
were
wrong
about
that,
but
you
know.
B
For
just
many
many
everyday
tasks.
A
Okay,
so
you
know
so
so
this
kind.
A
The
population
needs
to
be
highly
deterministic.
To
account
for
our
kind
of
normal
behavior
is
something
that
that
I
think
hasn't
really
been
explored
very
much.
Okay,.
B
B
A
So
we've
looked
at
this
in
guinea
pig
and
it
seems
to
be
about
the
same
number.
So
you
know
this
total
redundancy.
A
Easily
be
a
factor
two
different
in
in
different
species,
but
you
know
it's
definitely
the
case
in
mammalian
species
that
that
we've
recorded
from
that
there's
there's
a
similar
kind
of
issue
that
you
know
yeah
when,
when
you
define
cells
with
just
one
type,
they
don't
overlap
that
much
and
you
know
they
have
kind
of
minimal
redundancy
in
a
single
mosaic.
A
But
then
there
are
like
20
cell
types,
so
so
the
the
population
redundancy
is
is
is
going
to
be
quite
high
and
you
know
we
haven't
measured
it
as
as
kind
of
in
as
much
detail
in
those
other
species,
as
in
the
salamander.
A
Okay,
let's
see.
B
So
so
so
another.
A
Thing
that
I
think
is
is
is
really
quite
important.
Is
this.
A
That
they're
one
of
two
possible
stimuli
in
the
world.
You
know
a
lot.
B
Of
experiments
that.
A
You
know
that
have
that
have
been
done
to
study
how
you
discriminate
among
visual
stimuli.
Often
you
know
have
two
alternatives
and
you
want
to
take
the
population
and
figure
out
which
of
the
two
alternatives
had
occurred.
But
that's
that's,
that's
not
the
what
happens
in
real
life.
You
know
what
happens
is
it
could
be
anything
and
you
look.
A
B
A
Everything
else
another
important
characteristic
is
recognition
is,
is
often
very
fast
in
many
cases,
so
that
places
additional
constraints
on
the
population
code
and
another
thing
that
I
think
is
is
really
important,
but
but
isn't
considered.
A
lot
is
that
your
visual
system
can
detect
a
number
of
objects
and
they
can,
you
know,
be
anywhere
and
they
can
kind
of
be
present
or
pop
up
at
any
time,
and
so
what
that
means
actually
is
that
false
alarm
events
are
are
really
really
important.
You
know.
A
So,
if
you
think,
let's
say
you've
got
some,
you
know
detection
circuits
in
your
brain
that
recognize
you
know
spiders,
that's
kind
of
environmentally
important
stimulus.