akSJ3NucBsc

[Music]
before we begin I would like to say a
big thank you to today masa Mucha for
actually being a part of this video and
of course introducing me to the concept
of true north and now let's get into it
so hello there my name is Tanmay Bakshi
and this time we're going to be going
over a very interesting concept called
true north by IBM now just where
actually get into that though I would
also like to quickly say that I'm
actually currently in the IBM research
lab in Melbourne Australia and just over
weekend I'd like to introduce you to hit
day so today would you listen to no
thank you
I'm Hideo by didn't decide to Australia
I will show you the NAO robot which
TrueNorth sure definitely that was
actually very interesting and before I
tell you more about the backend
technology behind TrueNorth and what it
does I'd like to quick demo now this
over here as you've probably seen you
see now robot by Softbank
now we've been few modifications to the
robot and show you those in just a
moment there on the back of the robot
however I'm going to show you now if as
you can see over here on this computer
we've got a web interface and you can
actually see over here what's being
displayed from the robots eyes you could
say the robots camera which allows it to
see and you can also see there's a
neural network trained on the side part
and dataset and what this means is that
it will actually take the images capra
from camera run it through the five bar
ten-year-old network little output
predictions for what the robot is seeing
and if it predicts above a certain
threshold of confidence it'll say out
that I see this class out of the ten
classes and so now today we'll actually
go ahead and show them our robot views
of different images that we've got and
then from there we should be good to go
and we should see that seven now robot
shows us a few different tells us what's
in the images right now as you can see I
actually showed a huge of a horror
partition now an image of a ship you
cannot even wickedly webui what's
happening a fair together these auntie's
showing the images it's displaying the
result is live on this web interface um
now make cerumen or you probably do
realize this thing this image frame
hey there Nicole is actually updating
frequently and I'll talk about how it's
so fast in just a moment but earlier
take a look at the rest of the demo and
it's real line there y'all they asked
its classifying all these images and now
I trust from the camera of course it's
very different as you probably know so
now a little about itself has not nearly
enough computational power to actually
go ahead and classify images with such
high accuracy with such a deep neural
network at such great speed so how do we
achieve this and well that works by the
TrueNorth chip by IBM now let's talk a
little bit more about this over here now
this over here is essentially like a
little backpack that we've got for the
now Nova and what is this well
essentially this backpack in teams as
you can see over here something called a
true north chip by IBM and this chip is
essentially modeled off of how
biological neural networks or the human
brain works and so as you can see we've
got that little chip that's doing all of
the calculations for those neural
networks that's right these computations
were not done on a computer or on the
now robot
they're done on this limp chip on the
backpack that is now robot is wearing
now you can actually also see there are
only two wires going true and from the
TrueNorth backpack here and then our
robot one of that is a simple Ethernet
wire and this Ethernet wire is used for
data communication between the robot and
the backpack
however you can also see there's a
little black USB cable being used
between the robot and the backpack what
does that do well each
transmitted all of the power so the
chief needs to run that's right
within 5 volts and point 5x you can run
this entire chip which allows you to run
deep neural networks with a deep neural
networks with extremely high performance
in fact 15 frames frames per second
that's greatest and of course with such
high accuracy with this narrow robot so
now that you know a little bit more of
the actual technology that goes behind
this I love to today a few more quick
questions first of all how many synapses
faces how deep of a network can fit into
true north oh yeah - most have 256
million in up Skolnick that is
absolutely great we're gonna be able to
fit some really big neural network
architectures of this chip it's a lot of
questions what exactly were you doing
with tools and you in general yeah I'm
working on robotics own IP n IBM
Research Australia and listen free I'm a
frying through North for the Euro
votives image analysis and partition
parties yeah ok better yeah thank you I
think our focus behind exactly self
control for this video today thank you
very much everybody for joining in I
really do hope you enjoyed we're able to
learn something from this video and if
you did please make sure to leave a like
down below as well as if you have any
more questions or feedback since and
we've set down in the comment section
below if you email that to me at Tajima
name gmail.com or you can community at
Tai Chi Manny apart from that if you
really do with my content and you want
see more but please do consider
subscribing to my youtube channel as
well as it really does help out a lot
and of course that's going to be all for
this video today so thank you very much
everybody for joining in good luck