►
From YouTube: 2014 Spring NuPIC Hackathon Kickoff Presentation
Description
Matt Taylor & Jeff Hawkins
A
Okay,
thanks
everybody
for
coming
to
our
spring
2014
hackathon
for
new
pic,
it's
great
to
see
some
of
you
that
I've
seen
before
and
other
hackathons
and
new
faces
are
always
good
too.
So,
thanks
for
being
here-
and
we
hope
you
have
a
good
time-
and
we
will
be
here
to
help
you
through
with
everything
first
of
all-
I-
am
recording
and
live
streaming.
This
feed
and
sometimes
other
things
that
might
be
going
on
so
just
to
let
you
know
you're
fairly,
warned.
A
If
you
don't
do
anything
you
wouldn't
want
the
internet
to
see
also
have
these
recordings
available
on
ustream
afterwards.
If
you
want
to
go
back
and
check
to
see,
if
you
did
anything,
you
didn't
want
the
internet
to
see
later,
so
you
might
be
caught
on
tape,
feel
free
to
take
as
many
photographs
or
videos
as
you
like.
This
is
an
open
source
project,
an
open
source
community.
We
are
very
open
about
all
of
this
stuff,
so
posted
on
Twitter
or
Facebook.
A
Whatever
you
can
use
this
hashtag
new
pic
hack,
if
you'd
like
to
we'd,
be
very
happy
if
you
did
that
to
help
promote
our
project.
Thank
you
to
the
people
who
have
helped
me
out
with
this.
All
of
the
staffers
have
got
these
hackathons
shirts
with
the
word
root
on
the
back,
because
we're
super
users,
and
so
I
just
put
a
list
of
people
that
are
working
or
helping
with
new
menta,
and
then
we
have
a
support
staff
of
great
support.
A
People
who've
helped
me
with
logistics
and
the
badging
and
all
of
that
sort
of
thing.
So,
thanks
to
everybody
for
helping
out
and
another
big,
thank
you
to
pinger.
This
is
their
office.
This
is
where
they
work
every
day,
so
they
have
volunteered
their
space
to
us
if
no
no
charge,
so
that
was
very
gracious
of
them
to
do
so.
So,
thank
you,
pinger.
You
want
to
know
what
finger
is
go
check
them
out
at
finger
calm,
so
I'm
going
to
let
Jeff
come
up
and
say
a
couple
words
to
kind
of
kick
off.
A
B
Does
that
make
a
difference
to
anyone
here?
Okay,
I
noticed,
first
of
all
that
my
shirt
does
not
say
rude
and
I
think
that's
it
because
he
you
know
he
listed
those
engineers
that
well
I'm
an
engineer,
but
I'm
the
only
person
from
the
method
here
actually
probably
doesn't
code
anymore.
So
if
you
come
and
ask
me
like,
where
is
this
and
a
github,
but
everyone
else
can
answer
any
technical
question
you
come
so
I
think
that's
why
I
don't
say
truth
is
that
right,
no.
B
Phil
you
later
okay
anyway,
so
anyway,
thank
you
for
all
coming
here
and
for
spending
time
with
us
and
working
on
new
pic
I
just
thought.
You
had
a
few
quick
comments.
I,
don't
know
how
much
time
I
should
spend
on
this
business.
A
few
quick
comments.
This
is
a
this.
Is
you
know
this
isn't
a?
We
think
this
is
really
important
work
we're
doing
here
and
that
you're
participating.
You
know,
as
you
all
know,
I'm
sure.
All
of
you
heard
me
speakers
new
welcome.
You
know.
B
Hopefully
somebody
you've
heard
about
it
before,
but
this
all
got
started
because
many
years
ago,
over
30
years
ago,
I
fell
in
love
with
Brian's.
I
said
my
gosh.
We
can
figure
out
how
the
human
brain
works
in
my
lifetime
back
then
we
actually
had
it
no
idea
how
it
worked,
and
now
we
have
a
really
good
idea
how
it
works
and-
and
I
said
I
realized-
we
can
figure
that
out-
we
can
build
machines
at
work
on
the
same
principles
as
the
human
brain
and
in
the
goal.
B
B
It's
really
to
understand
how
the
brain
works
and
build
machines
that
work
on
the
same
principles
and
I'm
a
hun
percent,
confident
that
the
future
of
machine
intelligence
is
going
to
be
built
on
the
principles
that
are
in
new
pic
that
we're
working
on
and
you're
working
on
and
so
on.
You
know,
and
I
make
the
analogy:
I've
done
this
in
several
talks-
that
this
is
just
like
the
beginning
of
the
computing
here,
and
I
really
believe
that
we've
been
building
computers
now
for
80
years
or
so
on.
B
The
same
principles
goes
back
into
1930s
when
people
like
turning
into
a
Norman
wrote
papers
about
principles
and
the
concepts,
but
no
one
knew
how
to
build
a
computer
and
now
in
the
1940s
they
figured
out
how
to
build
them
and
they
entered
the
1950s.
We
had
a
business
and
it's
been
growing
and
growing
ever
since
we're
in
that
1940s
period.
B
Right
now
we
have
the
concepts
we're
starting
to
build
this
stuff,
but
we
are
literally
on
the
precipice
of
what
I
can't
imagine
ever
stopping
machine
intelligence
for
many
many
decades
to
come
and
we're
in
that
pioneer
state
and
one
of
the
things
about
being
a
pioneer.
Is
you
got
to
be
very
self-sufficient?
You
have
to
invent
the
stuff
as
you
go,
you
have
to
figure
out
what
works
and
what
doesn't
work.
It's,
not
the
infrastructure.
B
Isn't
there,
and
so
a
new
pic
is
perhaps
a
little
bit
unusual
in
that
regard
is
an
open
source
project.
Is
that
it's
not
like
easy
to
just
come
in
and
say:
ok,
I'll,
start
contributing
on
this
thing,
because
I
see
how
it
works.
New
pic
is
a
set
of
learning,
algorithms
and
concepts
that
are
difficult
to
understand.
B
Initially,
once
you
understand
the
really
elegant
and
simple,
but
initially
it's
a
steep
learning
curve
and
and
we're
figuring
out
we're
figuring
out
in
the
analogy
again
back
to
the
1940s,
we're
figuring
out
how
to
build
the
stuff
for
the
first
time
nowadays,
when
you
write
software
for
a
computer,
you
don't
really
know.
What's
going
on
side,
you
don't
know
about
all
the
registers
and
the
AL
use
and
how
they
do
a
floating
point,
operation
and
hardware.
B
You
don't
know
about
the
microcode
and
all
the
stuff,
but
back
in
the
early
days,
every
computer
scientist
have
to
know
everything
and
that's
kind
of
where
we
are
right.
Now
we're
trying
to
move
away
from
that
we're
trying
to
make
this
more
conceptualized
as
easily
build
applications.
That's
something
match
working
on
something
we're
working
on
or
building
frameworks,
they're
starting
to
go
frameworks
for
people
to
build
applications
without
having
to
know
all
the
details
inside,
but
it's
more
challenging
than
most
open
source
projects.
B
I
would
say-
and
I
think
that's
a
good
thing
and
it's
a
natural
thing,
and
so
that's
what
makes
us
all
pioneers.
So
that's
it!
That's
all
I
want
to
say
we
we're
here
to
help
you
and
I'm
going
to
be
around
all
day
today
and
tonight,
but
I
won't
be
spending
right
here
and
then
I'll
be
back
tomorrow,
and
we
have
a
lot
of
people
talk
about
so
I'm
just
going
to
talk
about
applications
and
other
things
we
can
do
so.
I
did
I,
probably
use
my
time
already
expired.
A
Thanks
Jeff,
now
back
to
what
I
know,
you've
been
all
waiting
for
hackathon
protocol.
So,
first
of
all,
the
portal
for
the
hackathon
is
new
mentor,
org,
slash
hack.
This
has
our
code
of
conduct,
the
hack
registration
link,
our
schedule
for
the
entire
event,
and
it
also
has
links
to
an
announcements
page
if
we
want
to
inform
everyone
of
something
that's
going
on
and
a
link
to
our
live
stream.
So
speaking
of
schedule,
well,
first
a
word
about
the
code
of
conduct.
A
We
posted
it
downstairs
when
you
came
in
it
just
basically
says:
don't
do
anything
stupid
or
don't
harass
people
don't
I
mean
you
can
read
through
it.
It's
a
standard
anti-harassment
policy.
We
don't
expect
anyone
to
do
anything
dumb
or
offensive.
So
don't
do
it
or
we'll.
Just
kick
you
out
our
schedule
today
we're
running
a
little
late.
That's
fine!
We
are
on
the
kickoff
as
soon
as
this
is
over
with
we're
going
to
start.
A
We
have
a
few
of
our
engineers
that
will
help
you
get
new
pic
installed
and
running
if
you
haven't
already
so
we're
probably
going
to
do
that
in
this
conference
room
over
here,
but
I'll
take
a
show
of
hands
of
people
who
think
they
need
that
help
from
us
in
a
moment
at
one
o'clock,
Scott
Purdy
will
give
you
a
beginner's
guide
to
new
pic,
and
this
is
a
really
kind
of
new
pic
processes
and
functionality.
So
this
would
be
great
if
you're
new,
to
new
pic
to
sit
in
on
this
stock.
A
It'll
be
right
here
at
one
o'clock,
it's
great
for
everybody
to
come
and
watch
this
talk
if
nothing
else
just
to
be
in
the
glow
of
Scott's
presence
at
two-thirty,
zubat
I
Ahmed
will
talk
about
anomaly
detection
using
the
CLA.
This
is
something
that
new
minta
as
a
company
has
done
really
interesting
things
with.
So
we
we
know
a
lot
about
anomaly,
detection
and
the
community
in
general.
We
don't
talk
a
lot
about
anomaly
detection,
but
it's
a
really
interesting
capability
of
the
CLA
so
stupid.
I
will
talk
about
that
at
two-thirty
right
here.
A
At
four
Jeff
will
give
a
very
informal
whiteboard
talk
about
some
ideas
about
temporal
pooling,
so
come
enjoy
that
at
four
tomorrow
at
ten
o'clock
Chayton
sir
/
will
talk
about
the
cortical
learning
algorithm
and
how
it
is
implemented
in
new
pic.
This
will
be
a
very
detailed
technical
discussion,
so
sort
of
an
advanced
level,
but
if
you
want
to
get
into
the
nuts
and
bolts
of
the
CLA
and
how
it
works
in
new
pic
would
be
a
great
place
to
be
at
ten
o'clock
tomorrow
morning,
when
you're
all
really
well
rested.
A
I
will
talk
at
one
o'clock
about
the
state
of
the
new
pic
open
source
project
where
we've
been
over
the
past
years,
since
we've
open
sourced,
where
we
currently
are
at
and
they're
sort
of
our
vision
for
the
project
in
the
future
and
then
at
4pm
tomorrow
we
will
start
demos,
and
that
is
you
guys,
giving
demonstrations
of
what
you've
created
over
the
past
36
hours
onto
the
pinger
offices.
So
this
is
a
map
of
the
office.
A
You
are
right
here
facing
me
and
behind
you
are
some
work
stations
that
are
unavailable
and
some
that
are
available.
So
what
I'll
say
about
work
stations
is
use
your
discretion
if
they
look
like
they
are
occupied.
If
there
are
personal
effects
strewn
about
the
workspace,
don't
use
it
try
and
find
another
place
to
work.
If
you
can't
find
a
place
to
work,
come
and
talk
to
one
of
us
with
the
route
shirts
and
we'll
try
and
find
you
a
place
to
work
so
upstairs
all
these
offices
are
off
limits.
A
We
do
have
this
room
here,
which
is
sort
of
a
lounge
area
and
this
big
conference
room
you're
free
to
work
in
those
spaces
if
they're
not
occupied,
and
if
you
can
find
an
open
workspace,
you
may
work
there.
You
can
fit
two
to
three
people
in
any
of
these
little
workspace
areas
so
crunch
together,
get
comfortable
with
your
neighbor.
We
have
bathrooms
directly
behind
us.
They
have
a
code.
The
code
is
posted
on
the
door,
so
this
is
the
third
floor.
If
we
go
one
floor
down,
oh
yeah
there's
an
emergency
exit
plan.
A
However
I
don't
know
how
viable
this
is
because
I
checked
the
one
emergency
exit
and
the
door
to
the
stairwell
was
locked.
So
I
guess
it's
every
man
for
self
downstairs.
So
there's
this
big
landing
here
and
stairwell.
So
when
you
go
downstairs,
there's
a
big
kitchen
area,
it's
kind
of
directly
underneath
of
us
and
that's
where
we
will
be
serving
meals
and
snacks.
There
are
a
few
rooms
that
are
open
for
use,
there's
a
little
mini
scrum
room
in
the
front
lobby
area.
A
There's
there's
a
conference
room
as
well,
and
there's
a
couple
of
open
offices
along
this
back
wall
that
you
can
use
if
they're
unoccupied
generally.
If
the
door
is
closed,
it's
don't
go
into
it.
All
the
doors
seem
to
be
open
where
we
have
open
space
to
work.
There's
a
lot
more
open
workstations
downstairs
as
well
same
rule
applies,
use
your
discretion.
If
it
looks
open,
it
probably
is,
if
it
looks
occupied,
it
probably
is
and
leave
it
alone
same
disclaimer
about
the
emergency
exit
plan.
But
it's
there.
That's
the
proposed
emergency
exit,
okay!
A
So
a
couple
of
things
about
the
facility
on
the
weekends
we
have
no
control
over
this.
The
elevator
takes
a
key
to
go
up
to
go
up
from
the
bottom
floor.
You
can
always
go
down,
but
you
can't
get
back
up
unless
you
have
this
like
key
card,
so
we're
going
to
have
someone
posted
there
for
the
major
hours
of
the
day
today
and
tomorrow
to
help
people
get
up
and
down.
A
You
can
sleep
here.
We
actually
have
some
sleeping
pads,
so
it
could
be
a
bit
more
comfortable
than
the
floor.
So
if
you
want
to
stick
around
just
come,
ask
us
when
it
gets
late,
we'll
start
putting
some
sleeping
pads
out
just
feel
free
to
grab
one
and
find
a
dark,
cubby
cubby
hole
somewhere
and
go
to
sleep.
The
Wi-Fi
is
posted
on
the
back
of
your
badges.
A
A
Okay,
where
your
badges,
please
as
much
as
you
can
remember
to
food
and
drink,
is
downstairs
clean
up
after
yourselves.
The
the
paper
plates
and
the
plastic
utensils
and
stuff
are
all
compostable,
so
in
the
in
the
trash
bins
down
in
the
kitchen,
there's
to
compost,
bins,
so
try
and
try
and
throw
the
compostable
stuff
in
there.
A
Try
and
pick
up
after
yourself
as
well,
we'll
be
going
around
kind
of
picking
up,
and
this
generally
isn't
a
problem.
But
if
you
see
some
empty
bottles
and
cans
or
something
and
you
see,
a
trash
can
is
going
to
toss
it
in
and
I
already
talked
about
the
code
of
conduct.
We
I
don't
anticipate
any
problems
with
respectfulness.
This
is
pretty
tame
crew.
A
Okay,
communications
at
the
hackathon
I
highly
encourage
you
to
go
to
our
IRC
channel
on
freenode.
If
you're,
not
an
IRC
person,
that's
that's!
Okay!
I
wasn't
really
an
IRC
person
for
the
longest
time,
but
I
get
into
a
lot
of
good
discussions
on
IRC
with
people
in
the
Newport
community
asking
questions
trying
to
figure
out
how
to
do
things.
So,
if
you
want
to
be
engaged
in
some
of
those
discussions,
please
join
in
and
also
use
the
mailing
list
as
well.
A
Questions
for
the
audience
want
a
little
bit
of
participation
by
show
of
hands
here,
so
we
can
get
an
impression
of
how
we
can
help
you
guys
out.
First
of
all,
who
has
ever
been
to
a
hackathon
before
good
good
number
of
you.
Okay,
that's
great!
So
if
you
find
yourself
confused
about
what's
going
on
or
what
the
protocol
is
just
ask
them,
one
around
you
apparently
and
chances
are
they
will
have
been
to
one
before
who
has
actually
built
new
pic.
A
Okay,
that's
better
than
I
thought
great.
Okay,
every
hackathon
it
gets
a
little
higher
the
percentage,
so
very
good.
So
I'd
say
that
was
most
of
you
that's
great,
so
who
is
actually
run
any
of
the
examples
like
the
hot
gym
example,
the
sine
wave
example?
Any
of
the
examples.
The
CPU
example
there's
an
audio
example
great
most
of
you
excellent
who
has
taken
your
own
data
and
passes
into
new
pic
or
at
least
attempted
to
pass
it
in
to
do
pick
good,
okay,
these
you're
encouraging
numbers.
Thank
you.
A
So,
on
that
note,
who
needs
help
installing
new
pic
12?
Don't
be
shy,
please!
This
is
going
to
help
me
decide
where
I'm
going,
to
put
you
guys
for
the
install
athan
great
we're
going
to
put
you
guys
in
this
conference
room
right
here
as
soon
as
this
meeting
is
over,
you
know,
after
maybe
a
10-minute
break
congregate
in
that
conference,
room
right
there
and
we'll
have
a
couple
of
our
engineers
trying
get
you
up
and
running
with
new
pic
on
your
machines.
A
Thanks
for
everybody,
we
do
have
if
you,
if
you
want
more
information,
I'm
trying
to
guide
everyone
kind
of
down
the
same
path
for
some
starter
resources
for
learning
about
how
new
pic
works.
Learning
about
how
you
can
use
new
pic
yourself
I
would
encourage
you
to
go
to
the
wiki
which
I
have
circled
here.
So
this
is
new
pic
org,
there's
a
link
to
the
wiki
right
on
the
menu
and
if
you
go
to
the
wiki,
there's
kind
of
a
flow.
A
If
you
want
to
learn
about
the
theory,
there's
a
link
to
that
there's
videos
with
some
of
our
engineers
talking
about
how
the
CLA
works,
subitizing
IDEO,
about
a
deep
dive
into
the
sea.
La
there's
Jeff
talking
about
sensory
motor
integration.
That
gives
a
lot
of
background
into
the
core
theory
of
CLA
at
HTM
there's
also
a
page
on
building
and
installing
new
pic,
which
will
help
you
with,
and
the
using
new
pic
link
gives
videos
tutorials
on
how
to
build
applications
using
a
new
pic
as
much
as
we
have
available
at
the
moment.
A
I'm
constantly
trying
to
build
this
out,
put
more
tutorials
up.
We
need
more
stuff
about
swarming
and
about
anomaly,
detection
and
stuff,
like
that.
So
I'll
be
populating
this
page
during
this
year
as
well,
but
there's
already
some
good
resources
there.
If
you
want
to
watch
them
some
videos
or
screencast
on
how
to
use
me
pic
be
a
good
place.
A
If
you
find
a
bug-
and
you
think,
hey
I
might
be
able
to
fix
that
that'll
show
you
what
our
process
is
for
getting
your
changes,
your
code
contributions
into
our
software
okay,
so
we're
almost
done
last
thing:
I
want
to
talk
about
his
ideas
for
hacks,
for
the
hackathon,
so
I'll
start
out,
because
I
have
an
idea
that
I've
been
wanting
to
do
for
a
long
time.
I've
never
had
the
chance,
but
I
heard
about
this
hackathon
this
weekend.
That
I
might
be
able
to
do
this.
A
So
I've
been
wanting
to
do
some
audio
processing.
There
was.
There
was
a
guy
named
Tim
that
did
an
example
that
we
ended
up
merging
in
our
code
base.
Is
he
here,
Tim,
McMurray,
macrina
or
Karina
know?
Okay
anyway,
he
did
this
audio
processing
example
where
he
took
audio
from
his
own
mac
and
streamed
it
in
into
new
pic
and
I.
Don't
know
all
the
extent
of
it,
but
it
worked
somewhat
so
I
want
to
do
something
like
that.
This
is
my
hack.
A
If
I
can
at
least
get
some
anomaly
indications,
maybe
it
could
be
like
a
really
repetitive,
techno
song
or
something
with
patterns
that
are
easily
distinguishable,
so
that
I
can
note
when
the
song
changes,
something
like
that
so
I'm
interested
in
doing
some
type
of
hack
like
that
and
I
know.
Other
people
have
other
interesting
ideas,
and
this
is
a
chance
where
I
want
you
guys
to
actually
stand
up
and
say:
I
have
an
idea
I'm
looking
to
do
this.
A
B
But
there's
something
we
want
to
do
it
no
Manta
and
I
think
we
agreed
that
we
just
throw
it
out
and
see
if
anyone
wants
to
work
on
it.
We've
done
a
lot
of
work
with
a
nominally
detection
of
numerical
streams
of
data
and
we'd
like
to
try
to
apply
to
two-dimensional
geospatial
data.
So
the
idea
is
you
attract
someone's,
be
movement
micron
and,
as
I
don't
know,.
A
C
B
There
we
go
I
suppose
supposed
to
stand
over
here,
so
people
can
stream
it
anyway.
The
idea
would
be
to
get
to
get
data
from
like
your
phone
as
you're
moving
around
walking
around
downtown
or
something
like
that,
and
you
have
to
build
an
encoder
to
turn
that
into
a
sparsely
stupid
representation.
We
have
some
ideas:
how
to
do
that,
which
I
could
share
with
you
and
then
that
would
be
the
and
then
you'd
also
want
a
visual
Zeiss
component.
You
want
to
show
their
tracking
with
anomalies
as
you're
moving
through
space.
So
how?
B
How
unusual
is
that
you've
gone
down
this
street
at
this
time
of
day
and
things
like
that,
so
there's
a
bunch
of
components
to
that.
If
anyone
sits
in
that,
we
can
tell
a
bit
more
about
up
again
I'm
working
on
it,
but
something
we
actually
want
to
see
done
then.
Maybe,
if
someone
interested
you
could
work
on
that.
A
Aaron
I'm
gonna
volunteer,
you
where's,
Aaron
yeah.
You
had
some
ideas.
Would
you
come
up
and
say
a
couple
words
about
them.
Just
go
tight-lipped
tight-lipped,
okay,.
A
E
I'm
interested
in
modeling
language,
syntax
phrases
in
particular,
sorry,
I'm,
sorry,
yeah,
I'm,
interested
in
modeling,
language
phrases,
syntax
and
I
have
basically
have
an
algorithm
for
doing
that
which
I'd
like
to
implement
in
a
new
pic.
So
anybody
is
interested
in
that
kind
of
project.
I'm,
not
such
a
coda,
but
I
have
some
ideas
on
the
subject
and
basically
anybody
who's
interested
in
language
love.
To
talk
to
you.
A
Thanks
Rob,
no
he's
back
there.
So
if
you're
interested
in
language
Rob,
you
should
talk
to
Francisco
Weber
he's
in
the
back
he's
the
founder
of
cept.
If
you're
familiar
with
his
word,
SDR
technology
they're
doing
some
really
interesting
stuff
that
in
that
space,
so
talk
to
him.
I
saw
another
handout
yeah.
Will
you
come
up?
F
F
The
first
one
is
simply
building
a
hierarchy.
That's
kind
of
in
a
column
shape
adding
a
few
regions
on
top
to
a
numerical
data
prediction
to
see
if
the
predictive
capabilities
of
the
algorithm
improve
as
you
do
that
the
second
one
is
building
a
hierarchy
to
do.
Predictions
in
economic
data
start
on
state
levels
in
the
united
states,
yet
a
few
economic
variables
and
try
to
do
a
prediction
model
off
of
that.
E
B
G
So,
for
example,
take
like
a
knight
Destiny's,
let's
say,
like
Syria,
for
example,
may
be
using
that
keyword,
aggregate
data
from
Twitter
like
around
Syria
and
then
maybe
detect
anomalies
as
surrounding
keywords
around
Syria
change
in
the
tweets.
So,
for
example,
you
know
the
narrative
in
Syria
is,
like
you
know,
kind
of
going
a
certain
way,
so
there
are
like
common
keywords
surrounding
Syria,
but
maybe
over
time
something
new
happen,
Syria
and
known
the
war,
and
maybe
some
new
keywords
and
rich
and
maybe
detect
anomalies
around
that
and
other
than
that
I
was
you
know.
G
D
I'm
Steve
levis,
I'm
interested
in
motor
control,
I'm
thinking
about
making
a
sort
of
a
primitive
1d
organism
that
might
play
like
pong
or
shuffle
puck
and
have
vision.
But
it
only
starts
out
where
it
reacts
just
to
what
it
sees
in
front
of
it
and
then
using
new
pic
to
do
predictions
to
help
it
improve
its
selection
of
which
direction
to
move
in
and
see
how.
Well
it
works.
H
High
at
the
previous
hackathon
I
had
done
some
work
with
adding
an
imagination
function
to
the
CLA,
and
this
was
basically
the
idea
that
you
know.
Could
you
could
you
make
a
decision
by
predicting
what
the
possible
outcomes
were
of
a
range
of
choices?
So,
if,
let's
say
you're
an
amazing,
you
wanted
to
turn
left
or
right.
You
know
which
ways
better.
Well,
you
need.
You
would
like
to
be
able
to
ask
the
CLA.
H
Some
work
then
and
I'm
thinking
of
continuing
that
work,
but
in
this
case
not
simply
predicting
where
cheese
might
be
in
a
maze,
but
instead
on
evaluating
on
on
on
taking
hints
overtime,
let's
think
of
it
as
though
a
mouse
sees
a
signpost
or
another
hint
or
a
scent
trail
and
based
on
those
kinds
of
inputs,
be
able
to
be
more
robustly,
find
find
the
cheese.
Shall
we
say
so
I'm
thinking
of
continuing
the
work
that
I
had
done
time
with
imagination
and
and
making
good
decisions
thanks
thanks.
I
All
right,
this
is
more
for,
like
cross-pollination
than
really
because
I'm
kind
of
still
learning
it
but
I'm
looking
at
taking
just
event
streams
from
just
like
traditional
client
software.
Look,
for
instance,
I'm
using
windows,
Forks
I
happen
to
know
a
bunch
of
events
they
have
set
up
and
just
trying
to
find
where
new
pic
can
make
sense
of
it
and
find
out
what
I
need
to
send
in
between.
For
instance,
if
you
have
two
instances
of
internet
internet
explorer
open,
do
you
need
to
send
them
extra
data
before
it
can
realize?
I
I
Till
I
can
just
see
if
I
can
reach
a
point
where
an
anomaly
is
suddenly
useful
to
me.
I'm
like
whoa
this
season
anomaly,
because
there's
something
wrong
happening
so
I
want
to
see
if
there's
anything
there,
just
kind
of
handed
him
learning
new
pic
and
seeing,
if
there's
any
treasure
to
unburied
in
this
space.
So
that's
it.
B
Very
interested
in
looking
at
we've
had
people
have
been
interested
in
looking
at
groc
our
product
for
anomaly
detection,
but
applying
it
to
unusual
behavior
on
insider
intrusion,
detection.
This
like
they
calls
as
the
Snowden
problem,
detecting
someone
who
is
like
going
rogue
inside
of
an
organization
and
the
idea
of
detecting
unusual
behavior
and
there's
a
big
big,
big
business
opportunity
there.
So
one
of
the
things
is
can
what
just
kicking
simple
metrics
off
of
someone's
pc
like
keyboard
statistics
or
application
statistics.
B
You
can
measure
at
a
regular
time
and
then
see
if
they
start
behaving
differently
at
some
time
in
the
future,
thats
related
to
what
you
were
just
talking
about.
I
think
that
is
a
big
application.
A
lot
of
people
are
interested
in,
so
that's
something
we're
interested
in
as
well,
so
I
can
just
throw
the
thoughts
about
if
you
want
ok,
yeah,
that's.
A
C
Okay,
so
not
necessarily
an
idea
but
a
resource
that
hopefully
might
spark
some
ideas.
I
worked
on
a
project
called
fluent
for
this
preparation
for
this
hackathon,
so
people
can
it's
along
the
side
of
language
prediction:
it's
a
API
and
a
Python
library.
It's
a
web-based
API
in
a
Python
library
that
you
can
feed
it.
Words
and
it'll
predict
the
next
word
using
the
set
technology.
Those
guys
are
there
in
the
back
and
the
CLA
together.
C
So
it'll
actually
try
to
understand
what
you're
you
know,
what
the
words
mean
and
what
words
should
follow,
and
so
you
can
feed
it
a
word.
It's
pretty
simple.
You
feed
it
a
word
as
a
sequence
of
words
and
will
predict
the
next
one,
and
so
maybe
that's
that's
a
resource
available
for
you
guys
to
use
for
those
of
you
interest
in
language
in
your
projects,
maybe
in
making
some
kind
of
language
based
games
or
apps
of
that
kind.
C
C
A
A
J
Cricket,
a
and
I've
posted
this
actually
on
the
new
meta
discussion,
mailing
list,
angry
bots
in
AI
world,
and
if
you
saw
this
a
3d
game,
I'm
sort
of
taking
the
data
from
that
I
want
to
train
new
pic
on
it
to
play
the
game
and
I've
taken
the
images
from
within
the
game
and
actually
transform
them
through
an
open
source.
Pre-Trained
neural
net
called
cafe
and
at
one
like
some
imagenet
competition.
J
So
I
have
basically
a
vector
thousand
real
numbers
floating
point
numbers
from
the
image
that
I
want
to
feed
into
a
new
pic,
along
with
the
like,
was
shot
and
shot
landed
like
boolean's,
because
it's
a
first
person
shooter.
So
it's
just
a
bully
Affleck
did
I
get
shot
that
I
hit
a
guy
and
what
were
the
moves?
J
G
A
Okay,
alright,
so
that
wraps
it
up
for
this
kickoff
presentation.
Let's
let
me
just
give
a
quick
shout
out
to
to
Jeff
full
who
designed
our
t-shirts,
hey
Jeff,
keys
right
here
up
in
front
other
than
that
with
let's
take
a
ten-minute
break
ten-minute
break
and
for
those
of
you
who
need
help
installing
a
new
pic
meeting.
This
conference
room
right
here
in
10
minutes
so
that'd
be
about
11,
ish,
ok
and
then
chayton's
going
to
do
a
little
fluent
demo
in
about
10
or
15
minutes,
whatever
you're
comfortable
with
and
we'll
just
okay.