►
Description
2015 HTM Challenge Application Submission (ineligible for prizes)
B
B
What's
going
on
during
the
video,
we
made
a
game,
a
mind-control
game
where
you
wear
a
EG
headset,
and
by
getting
your
brain
wave
in
encoding
them,
we
can
train
a
model
and
classify
the
output
of
that
model,
to
control
the
little
minion
to
go
left
or
right,
and
what
you're
going
to
see
is
basically
a
demonstration
of
that
game.
If
you
want
to
go
in
the
details
of
how
we
implemented
it
after,
I
am
happy
to
answer
questions
with
making
teammates.
C
B
We
are
including
the
data
we
filter,
it
remove
some
noisy
frequencies.
Then
we
take
the
Fourier,
transform
and
extract
frequency.
That's
calling
you,
which
is
characteristic
of
little
context,
emulation
when
you
think
about
left
or
right.
You
can
detect
that
the
electrons,
so
we
stand
immune
frequency
included,
is
a
scalar
encoder
to
the
HTM
and
we
send
the
tags
as
labels,
as
we
were
saying
for
continuous,
continuous
learning,
and
so
each
team
produces
classification
results
and
that's
what
is
controlling
the
million
left.
B
C
B
A
B
Yes,
actually
cared
you'd
a
lot
of
work
on
so
removing
the
60
Hertz
frequency,
because
this
is
ambient
noise
of
electronic
devices.
Remove
that
we
remove
I
blinks
with
ica,
which
is
a
very
it,
doesn't
contain
the
information.
So
we
need
to
remove
this,
and
then
we
take
the
FFT
where
we
extract
that
one
frequency
that
is
interesting
and
representative
of
module
movement.
That
is
what
we
encode
and
feed
the
HTM.
How
many
channels
is.
E
F
We
had
eight
channels,
but
we
couldn't
really
feed
that
much
data
to
the
HTM,
so
we
ended
up
doing
like.
Unlike
the
BCI
literature,
they
call
like
a
Laplace
filter
but
basically
just
put
like
one
channel
in
the
middle
and
you
subtract
the
channels
around
it.
So
we
had
to
two
streams
of
data
that
we
fed
to
the
HTM
really.
B
Was
going
on
a
weather,
pre-trained,
yes,
so
indent
demonstration,
it
was
live
after
that,
a
better
way,
a
way
to
get
better
results
is
to
pre-trained
it
stop.
The
learning
and
just
average
camp
predict
when
you
do
that.
It's
nicer,
because
you
won't
tell
you
that
nice
pre-trained
it
I
said
that
you
have
we're
really.
You
had
the
user
carefully
label
carefully
produce
motor
movement
for
the
right
target
when
it's
live
and
continuously
learning.
B
G
B
G
E
Communion
with
you
guys
aren't
the
first
ones
I
assume
we've
tried
to
do.
This
may
be
the
first
one
to
do
with
HTM.
So
what's
what
is
the
expected
state-of-the-art
performance
of
you
know?
It's
not
even
clear
how
much
of
this
problem
can
be
solved
with
such
a
you
know
such
a
signal.
That's
you
know,
that's
kind
of
very
crude.
So
is
this
a
solved
problem
and
then
you're
trying
to
stop
it
again
using
this
technique
or
there's.
F
B
B
To
say
it's
hard
to
say,
because
if
it
was
because
of
the
nature
of
the
neurofeedback
application,
because
it
also
helped
the
person
using
the
game
or
if
it's
because
the
HTM
work
better,
it's
it's
helped
to
say
why
it
worked
better
than
the
other
application,
because
that
was
a
different
setup.
You
had
the
robotic
arm
moving
yeah.
F
D
Over
time
did
he
get
accurate
enough
so
that
the
user
could
think
about
something
else
and
the
minion
wouldn't
move?
But
then,
when
you
thought
about
left
and
right
again,
it
would
move
or
were
there
a
lot
of
sort
of
false
positives
where
you
just
like
you're
saying
someone
was
talking
to
the
user
and
that
made
the
menu
move
around
and
the
way
he
didn't
expect.
Or
was
it.
B
B
Thing
is
when
you,
when
you
train
the
HTM
we
trained
it
was
left
and
right.
So
really
you
couldn't
really
stop
with
that.
Pre
training.
Dataset
assume,
if
you
train
with
more
things
where
you
you
include
that
baseline,
which
we
should
do
then,
because
you
don't
produce
that
new
frequency,
then
it
should
start,
but
as
it
is,
it's
left
all
right.
So
oh
yeah.
D
B
Oh
yeah,
so
one
of
the
things
that
actually
yesterday
night
we
we
did
to
also
look
at
just
with
your
eyes
how
you
could
predict
this
as
well.
If
you
plot
the
output
of
these
two
electrodes
on
your
left
and
right
motor
cortex,
you
will
see
the
increase,
a
very
neat
increase
in
new
frequency
when
you
actually
move
your
left
or
right
hand.
So,
when
you
don't
do
anything
you
something
else,
you
don't
see
that
frequency
that's.