►
From YouTube: Kickoff (2015 Spring NuPIC Hackathon)
Description
Protocol and kickoff by Matt Taylor and Jeff Hawkins.
A
B
A
Caught
on
tape,
just
know
that
if
you
would
like
to
take
any
photographs
or
videos
feel
free,
this
is
a
very
open,
transparent
thing
we're
doing
here.
So
if
you
want
to
use
a
hashtag
for
social
media,
it
is
hack
new
pic.
Let
me
say
thanks
to
everyone
who
is
helped
me
with
this
hackathon
real
quick,
Terry,
try
and
tailor
their
there
at
the
registration
desk,
downstairs,
they've
been
instrumental
and
for
logistics
and
planning,
and
thank
you
Jeff
and
the
rest
of
our
new
pic
engineers
on
the
right
here.
A
Who
will
all
be
here
willing
to
help
you
get
new
pic
installed
brainstorm
about
what
you
can
do
with
your
data?
We've
all
got
root
on
the
back
of
our
shirts.
So
if
you
see
somebody
with
root,
ask
some
questions
and
if
they
can't
answer
it,
they'll
find
someone
who
can
also
thanks
Cornell
Tech.
Are
there
any
Cornell,
Tech
students
here?
Okay?
A
C
Is
Evan
I
just
saw
this
logo
the
first
time
Jimmy
as
it
goes,
everyone
know
what
that
is.
It's
great.
My
wife
is:
what's
that
picture,
I
got
no
I
these
great
pose,
so
one
of
our
community
members
are
created
that
time
anyway,
thanks
Matt,
thank
you
for
coming
here
and
we're
trying
to
be
here.
This
is
our
first
hackathon
outside
of
Silicon
Valley,
and
so
it's
a
sort
of
a
new
experience
for
us.
Although
we've
done
some
before
Matt
always
runs
a
great
show
and
we're
gonna
try
to
make
this
really
valuable
for
you.
C
C
Ways
we
are
trying
to
invent
that
we're
trying
to
reverse
engineer
the
neocortex,
we're
making
good
progress
of
that,
and
we
want
to
build
machines
of
work
in
those
principles.
That's
what
new
pic
is
all
about.
It's
the
open
source
component
of
that
and
and
I
think
the
implications
of
this
work
are
going
to
be
very,
very
profound
in
the
future,
so
we
all
get
a
chance
to
participate
in
that.
It's
not
clear
talking
to
how
the
future
will
play
out
or
the
right,
what
time
frames
and
so
on.
C
It's
kind
of
a
messy
time
right
now,
but
it's
it's
it's
something.
It's
really
happening,
it's
starting
to
form
and
I.
Think
menta
is
sort
of
the
center
of
an
ecstasy
of
many
decades
of
machine
learning
based
on
cortical
principles
and
machine
intelligence
in
general.
So
it's
great
to
have
other
people
to
participate
in
that
and
I
know.
Some
of
you
come
from
a
long
way
to
be
here,
so
we're
gonna
try
to
make
it
all
very
successful
for
all
of
you,
it's
important
work
and
I
don't
really
want
to
delay
things
much
at
all.
C
I
just
want
to.
Thank
you
and
for
coming
and
we're
try
to
make
this
successful
and
not
just
this
event,
but
new
pic
as
a
whole
and
machine
intelligence
of
the
hall
and
brain
theory
and
all
that
stuff.
So
then,
thanks
for
coming
and
I'll
be
around
here
myself
for
a
part
of
the
day
and
then
I'll
be
back
tomorrow,
and
so
we
can
have
lots
of
conversations
that
people
want
to
chat.
But
I'm
going
to
give
this
back
with
Matt
who's
really
in
charge
and
running
everything,
and
it's
going
to
make
it
work
thanks.
C
C
C
A
Be
at
newman
org,
slash
hack,
if
you,
if
you
go
there,
it's
got
links
to
our
schedule,
to
or
code
of
conduct,
to
resources
to
everything
and
even
where
to
register
your
hack.
All
that
stuff,
we
will
register
on
a
website
called
challenge:
post
challenge:
post
calm
if
you're
going
to
have
a
hack
demo.
At
the
end
of
this,
you
need
to
register
their
more
infos
on
limited
on
org,
slash
hack.
All
of
these
slides
also,
so
these
slides
have
a
bunch
of
links
in
them
that
might
be
useful
to
you.
A
You
can
also
find
them
at
the
mental
net
org
slash
hack,
so
today's
schedule
I,
am
doing
the
kick
off
at
around
ten-thirty
or
as
soon
as
I'm
done.
With
this
we're
going
to
have
a
couple
of
new
pic
engineers
show
off
a
few
samples
of
things.
They've
done
with
new
pic
using
what
Chayton
is
has
a
demo
using
the
Unity
game.
A
Engine
and
feeding
data
from
Inuvik
and
Austin
has
something
with
the
project
tango
that
should
be
interesting,
1130
to
1230,
where
we
will
be
available,
probably
and
there's
a
big
conference
room
in
the
back
here.
So
we'll
have
nipping
engineers
to
help
you
get
who
big
installed
if
you're
having
problems
getting
it
compiled
and
running
on
your
laptop
at
12
30
to
1
30
in
the
room
back
behind
the
projection
screen
we're
going
to
have
sort
of
a
brainstorming
session
on
music
theory
with
new
pic.
A
A
Guest
with
us
that's
going
to
help
is
it
that
these
are
he's
our
in
house
music,
theorists,
Charlie
Gillingham
from
counting
crows,
is
the
keyboardist
for
counting
crows
and
he's
also
a
computer
scientist.
You
didn't
know
that
so
we'll
be
running
that
session
and
hopefully
pick
his
brain
on
some
music
theory
and
how
we
can
it
go
to
MIDI
data
into
sports
representations
I'm,
going
to
give
a
short
session
on
the
state
of
New
pick
two
o'clock
at
two-thirty
here
attend.
A
If
you
want
up
it's
going
to
go
around
like
what
the
communities
like
what's
going
on
in
the
community,
what
kind
of
projects
are
happening
where
we're
going?
What
we've
accomplished
over
the
past
year,
so
it's
more
focused
on
people
interested
in
contributing
or
in
the
development,
in
the
progress
of
what's
happening
with
the
new
pic
project
itself,
and
we
also
have
connor
Russa
mono
from
open
bc.
I
who
brought
a
bunch
of
hardware
stuff,
open
DCI,
does
a
kind
of
dyi
EEG
kits
so
that
they've
got
these
little
spider.
A
Contacts
and
you
can
get
the
EG
information
from
your
brain
and
we've
had
a
couple
hacks
in
the
past
of
getting
that
day
into
new
pic
and
using
it
to
control
things.
So
that
could
be
interesting
if
you're
interested
in
that
please
attend
that
session.
I
will
probably
be
back
in
this
room
behind
the
projection
screen
tomorrow.
I've
got
nothing
on
the
agenda
except
demos
at
four.
So
if
you're
going
to
demo
be
ready
at
four
o'clock
a
couple
things
about
the
facility
emergency
exits,
you
can
see
a
sign
immediately
upon
exiting
this
room.
A
It
says
exit,
that's
where
you
go
all
of
the
doors
in
here:
don't
open,
pulling
or
pushing
they're
all
sliding.
This
has
been
a
problem
in
the
past,
apparently
so
just
be
aware
of
that
it
doesn't
open
it's
because
you're
doing
it
wrong
slide
the
door
back
and
forth.
I
mentioned
bathroom
coats
earlier.
The
men
and
women
have
different
coats
there
on
the
door
on
the
way
out.
You
just
take
a
look.
Our
Wi-Fi
here
is
google
guest.
It's
very
fast.
It's
free
and
no
password.
If
you're
sleeping.
F
A
Feel
free,
you
might,
hopefully,
you
brought
a
pillow
or
sleeping
bag
or
something
because
there's
no
couches
in
here,
but
if
you
want
to
talk
about
accommodations,
you
come
talk
to
me
later
and
maybe
I
can
give
you
some
tips
on
where
to
crash.
Okay,
a
little
more
protocol,
you
do
need
your
badges
to
get
in
the
door
tomorrow.
Keep
this
thing
and
bring
it
back
tomorrow
on
the
back
of
it
is
the
real
important
thing.
This
is
what
the
security
guards
are
going
to
want
to
see.
So.
A
Also
has
a
gift
card
for
food.
We
usually
provide
food.
These
events
because
of
logistics
problems
of
this
facility.
We
were
not
able
to
provide
meals,
so
we
compensated
by
giving
you
a
gift
card.
If
you
can't
find
food
out
on
the
street,
there's
something
wrong
with
you.
There
is
times
and
food
around
here.
It's
lots
of
places
to
eat,
so
please
go
use
that
gift
cards,
so
you
can
go
eat,
there's
trash
cans
all
over
the
place,
please
use
them.
I
will
be
making
sweeps
every
few
hours
or
so
to
pick
up
trash.
A
It's
a
lot
easier
if
you
just
clean
up
up
to
yourself
be
respectful
of
the
facility.
If
you
don't
know,
if
you
should
be
touching
something
come
and
ask
me,
I'll
tell
you
if
you
can
touch
it
or
not,
and
also
please
take
a
look
at
our
code
of
conduct.
It's
a
typical
code
of
conduct.
Don't
do
anything,
that's
that's
harassing
or
offensive
to
people.
If
you
think
it
might
be
offensive,
it
probably
is
offensive,
so
don't
come
up
with
any
hacks
that
that
might
offend
anyone.
If
you
want.
A
A
There
is
a
general
discussion
list,
which
is
probably
your
best
bet
and
there's
one
for
development
and
there's
one
for
theory,
so
feel
free
to
sign
up
for
that,
and
and
remember
for
those
of
you
taking
pictures
of
the
slides
that
the
slides
are
available
in
limited
a
dork,
/,
hack
and
I
have
all
these
links
in
them.
We
also
have
a
chat
room
on
IRC
which
I
sometimes
on,
and
we
also
have
a
git
er
chat
room.
B
A
Stuff,
so
ok,
a
few
questions
before
I
go
on.
Please
raise
your
hand
if
you've
ever
been
to
a
hackathon
before
a
lot
of
people.
That's
good!
Ok
is
it
who
has
we've
compiled
and
built
new
pic
on
their
laptop.
This
is
a
better
indication
of
how
busy
we're
going
to
be
helping
you,
ok,
so
maybe
half
that's
not
too
bad
oops,
who
has
run
any
of
the
example
code
tutorial
stuff
like
that.
A
F
A
The
wiki
is
laid
out
in
a
way
to
kind
of
drive
you
in
a
path
that
will
help
you
build
and
install,
give
you
tutorials
on
how
to
use
new
pic
and
then
how
to
learn
like
what
new
pic
is
doing
and
the
theory
behind
it
and
that
sort
of
thing
so
thats,
a
good
resource
to
start
at
here
are
some
suggested
links,
especially
for
you,
hackers,
if
you're
new
to
new
pic,
and
you
want
to
figure
out
what
it's
all
about.
We've
got
a
couple
of
tutorials.
A
We
call
Jim
it's
about
energy
consumption
over
time.
This
is
a
really
nice
use
case
or
new
pic
because
it
recognizes
these
patterns
and
energy
consumption
in
buildings
really
well.
So
both
of
these
have
screencasts
and
sample
code,
and
me
talking
on
and
on
and
on
about
how
to
build
it
out.
So
hopefully,
that's
understandable,
there's
one
for
prediction
and
there's
one
for
anomaly:
detection,
which
follows
the
prediction
one.
A
We
also
have
a
beginner's
guide
to
new
pic,
which
is
Scott
pretty
gave
at
a
previous
hackathon,
which
is
a
really
nice
dive
into
like
the
details
of
encoders.
How
how
we
encode
scalar
data,
for
example,
into
sparse
representations.
So
that's
a
good
resource
and
there's
a
link
here
to
a
bunch
of
different
code,
examples
on
our
wiki
cpu
usage,
spatial
and
temporal
bullying
classification,
all
that
stuff.
So
we
all
have.
We
have
code
examples
of
all
this
stuff.
A
So
if
you
see
something
that
sparks
an
interest
you
take
hey,
my
hack
might
fit
in
that
start,
start
with
that
and
see.
If
you
can
run
those
examples,
we
also
have
a
bunch
of
old
demos
from
old
hackathons
yeah
from
old,
hackathons,
I
think
14
demos
from
last
hackathon,
something
like
that
from
the
one
before
so.
If
you
want
some
ideas
of
what
other
people
have
been
tried
and
all
other
hackathons
that
go
there,
if
you
do
not
want
to
use
new
bakery.
D
A
You
don't
want
to
you
want
one
right:
icon
code.
There
are
other
HTML
plement
ations
you
could
potentially
use.
We
have
HD
mjv,
which
is
a
port
of
the
new
pic
python
HTM
algorithms,
by
david,
ray
david
ray
he's
right
back
there
yeah
there
you
go
embarrass
him,
okay
and
we
have
comport
decks
by
Felix
Andrews.
This
is
right
here,
which
is
a
closure
implementation
of
HTM.
It
doesn't
closely
follow
the
our
Python
implementation
of
the
HTML
gorillas
like
HTM
Java
does,
but
it's
still
an
HTM
okay,
a
little
guidance
about
your
demos.
A
The
most
important
thing
I
think
is:
if
you
have
an
idea
about
something
think
about
your
data,
think
about.
Do
you
have
a
lot
of
data?
Is
it
Tim
poro?
The
biggest
question
is:
is
the
data
you're
thinking
about
using
for
hack
temporal?
Does
it
have
a
time
component?
Because
if
it
doesn't,
if
it's
just
spatial,
you
might
not
have
much
success
key
thing
about
html5
emery
is
a
temporal
component
of
that.
So
make
sure
you
have
a
temporal
component,
your
data,
or
at
least
you
can
somehow
give
it
a
temporal
component.
A
A
C
C
A
You're
gonna
need
at
least
a
thousand
data
points,
probably
no
matter
what
okay,
a
great
way
to
learn.
Htm
is
to
build
an
HTM
in
your
favorite
programming
language.
That's
great
I
love
it
when
people
do
this,
but
it's
not
a
good
hack,
because
you're
just
standing
here
telling
people
about
what
they
already
know
and
here's
here's
my
life
HTM
in
whatever
so.
A
At
the
hackathon,
but
don't
make
it
a
hack
demo,
what
would
be
happy
to
help
you
do
that?
But
it's
not
a
good
at
home.
Okay.
So
let's
talk
quickly
about
anyone
having
any
ideas.
L
word
right
on
schedule,
so
good
I
talked
a
little
about
music,
so
I'm
really
interested
in
encoding
MIDI
data,
somehow
so
that
new
pic
can
understand
patterns
and
music,
and
this
is
an
example
that
Jeff
uses
a
lot
when
he's
talking
about
brain
about
HTM
and
involved,
and
we've
actually
had
demos
in
the
past,
our
first
hackathon.
A
A
Impressive
to
me,
but
it
wasn't
like
a
true
encoding
because
they
hacked
it
into
scalar
data.
So
the
the
real
challenge
here
is
how
to
get
midi
data
or
any
type
of
some
type
of
musical
representation
into
a
sparse
format
that
that
encodes
more
information
than
just
the
pitch
of
the
note
right,
because
there's
more.
D
A
D
A
E
I
I
Basically,
you
you
shoot
light
at
something
and
then
whatever
light,
doesn't
bounce
back?
Tells
you
what's
in
the
object
that
you're
the
light
at
or
the
light
is
passing
from
depending
on
how
do
you
want
to
do
it,
but
I
can't
think
of
a
good
way
to
make
that
temporal,
because
it
doesn't
seem
like
something
that
would
depend
on
how
long
you
shoot
the
light
at
it
most
great
sense
of
goofy.
Maybe.
A
A
K
I
I
A
C
It
depends
on
what
you
kind
of
data
get
out
of
you
know,
quarter
clio
did
these
word
SDRs
and
they
actually
get
really
interesting
results
without
even
the
temporal
component,
just
using
the
sort
of
semantics
and
coatings
of
st
origin.
You
might
be
able
to
something
like
that,
at
least
with
the
words
that
we
were
able
to
then
say:
okay
now
become
a
sequence
of
the
sentences,
you
might
not
be
able
to
be
there,
but
it's
still,
even
just
the
SDR
component
of
the
theory
might
be
useful
for
you
in
some
way.
C
F
A
L
My
name
is
Raja.
It
came
to
that
last
hackathon
aight.
Then
I
get
a
chance
to
contribute
a
lot,
so
I've
been
waiting
about
new
pic
in
the
hack
that
I'm
thinking
about
much
even
sure
this
was
a
good
hack,
but
I
recently
write
a
paper,
a
stanford
paper
which
was
trying
to
predict
outcome
of
baseball
games
and
they
had
compared
about
six
or
seven
AI,
so-called
AI
algorithms,
and
predicting
what
the
outcome
of
baseball
games
is.
M
M
O
A
A
M
J
My
name
is
ashish,
I'm
working
for
a
doing
a
project
or
New
York,
Presbyterian
Hospital
this
summer
on
patient
onboarding
and
also
at
home
care
poster
discharge.
Some
kind
incorporates
some
kind
of
decision
analysis
before
they
get
to
the
hospital.
So
look
up,
you
know
what
makes
people
think
which
hospitals
to
choose
when
what
is
the
decision
making
I
am
not
sure
how
that
translates
into
this
project,
but
hopefully
there
are
some
application.
Second,
somebody
can
help
me
with.
P
My
name
is
Dave.
Petrillo
I
had
an
idea
for
using
HTM
at
automotive
data.
So
can
we
add
just
a
few
extra
sensors
to
a
car
and
get
a
better
idea
of
what's
wrong
with
the
car
for
consumer
applications
or
get
a
better
idea
of
how
the
drivers
doing
for
like
performance
driving
applications?
It's
like
it
seems
like
there
should
be
more
information
in
the
data
than
just
okay.
This
sensor
is
too
hot,
and
so
then
the
iTunes
do
hot,
but
instead
the
rumble
came
from
the
front
right
corner
of
the
car.
P
Q
Right,
my
name
is
Patrick
Ryan
I
am
interested
in
trying
to
model
student
data.
I
have
some
generated
student
data
based
on
students
that
are
good
and
bad
and
I'm
thinking
about
making
it
ten
oro
by
encoding.
The
student
and
the
question,
and
maybe
taxonomy,
is
about
that
question
and
then
followed
by
their
answer
and
whether
it
was
correct.
It's
just
a
simple.
You
know
212
sequence,
but
it
might
work.
R
I'm
Steve
deviled
egg
and
for
my
dissertation
I'm,
trying
to
model
interaction
between
the
hippocampus
and
the
cortex
and
there's
a
particular
rap
experiment
that
that
I'd
love
to
be
able
to.
I
started
to
try
to
put
into
new
paper.
But
only
gotten
a
little
way
in,
and
that
has
to
do
with
a
rat
will
project
out
of
their
hippocampus.
A
trajectory
about
where
they're
going
to
move.
And
if
you,
if
you
were
to
feed.
R
A
C
N
I'm,
my
name
is
Scott,
so
I
consider
using
anomaly
detection
for
fraud
detection
in
the
mobile
space,
so
right
now,
mobile,
advertisers
publisher,
specifically,
are
left
to
sort
of
set
up
their
own,
their
own
metrics,
using
mostly
thresholds.
If
you
can
use
simple
device
identifiers
like
the
idea
of
a
the
country,
maybe
their
locale
in
the
operating
system,
you
could
potentially
do
some
useful
things
in
detecting
the
types
of
traffic
and
the
determinant
ama
listing,
so
preserve
that
to
team
security.
N
A
O
Want
to
be
smooching
so
I,
so
a
couple
days
ago
there
was
I
quite
interesting
analysis
of
the
lstm
you
need.
There
was
some
guy
posted
this
from
my
from
Stanford
to
guess
and
I
figure
out.
Well,
let's
try
to
create
a
little
water.
Don't
worry
too
sweet
the
Centers.
You
know
and
say
hi
and
you
know
try
to
network
and
you
know
see
what
the
network
is
going
to
come
up
with.
That's
a
previous
person
had
you
know
like
a
bunch
of
Shakespeare's
as
well
as
we
could
be.
The
eyes
was
caramel
sauce
interview.
O
S
As
part
of
trying
to
understand
the
new
pic
algorithm,
I
was
interested
in
doing
us
like
a
sensitivity
assay
of
the
algorithm
itself,
using
generated
generated
data
train
on
a
classification
problem
and
then
perturb
that
data
and
see
okay.
Where
does
the
classification
start
breaking
down,
if
you
add
randomness
or
drift
or
something
to
to
the
input
patents?
So.
A
N
E
D
D
E
B
K
How's
it
going
everyone
I'm
Connor,
from
open
bc.
I
I'm
interested
in
taking
an
open,
pci
device
and
plugging
or
connecting
16
channels
to
the
kind
of
lower
forearm
and
seeing
if
we
can
use
new
pic
to
classify
individual
finger
movement
because
right
now,
I've
got
a
system
rigged
up
where
I've
got
a
single
EMG
input
that
can
make
a
hand
flex
in
like
an
analog
motion,
but
I'm
really
interested
to
see
if
we
can
identify
individual
fingers.
So.
T
E
T
T
O
F
A
U
A
D
Hi,
my
name
is
John
yay,
I'm
working
on
two
projects.
One
is
a
acoustic
project
that,
by
listening
to
the
noise
of
the
faucet,
sound
won't
be
able
to
detect,
which
faucet
and
hot
or
cold
where
it
is,
and
the
other
one.
I
want
to
you,
I'm
starting
to
kick
around
this
idea.
It's
actually
using
the
switch
matrix
on
the
h-town,
the
fpga
device
instead
of
hard-coding,
the
actual
fpga
want
to
be
able
to
do
the
weighted
learning
as
part
of
the
switch
matrix
on
the
fpga
p.
A
H
F
H
F
V
O
C
T
C
T
C
You
know
some
especially
very
nude
in
the
pic.
If
you've
got
an
idea,
you
want
to
run
City
thinking,
Craig's
and
data
running
through
all
you
have
some
data
try
to
find
one
of
the
men
to
employees
and
just
discussing
a
little
bit,
it's
very
easy
to
make
incorrect
assumptions
about
how
the
data
should
be
the
sampling
rates
and
what
is
temporal
data,
and
so,
if
you
have
even
any
any
thoughts
at
all,
do
this
it's
really
useful
more
people
trip
up
on
the
data
I
think
than
anything
else.
So
it's
it's
worth.