– Thank you very much
for the introduction. Artificial Intelligence is increasingly taking over our everyday
jobs to make our life easier. Some of these examples
include autonomous cars, robots, wearable held,
and assistive devices. All these applications that you’re looking they require for lot of data processing and knowledge extraction in real time. On the other hand, there is a new trend of for new class of computing, which requires smaller,
higher energy efficiency with much more higher
computational capability for these devices. So back in a couple of decades ago, we had this gigantic mainframe computers that were taking the space of the room, and they were consuming
hundreds of thousands of watts, and nowadays, we have all
these variable smart devices, that by define do a lot more computations than that gigantic mainframe computers. But they require a lot lot
less power consumption. The new constraint is to have
much lower power consumption within a lot smaller devices, however, with much higher
computational capability. So I give you one example, so the headset that you are seeing there, it’s basically a headset
that has been developed in my lab. And can be used for
disabled people to control their environments, including
driving their wheelchairs. So the AI processor that
is behind the headset is responsible to perform
all the data processing knowledge extraction,
emotional learning, and AI, to basically make that
application working. However, in order to have that headset being able to be wearable and at most we want to
charge this one once daily, so we need to have that
AI processor in the back to be very low power consuming. So for example, this application we don’t want to have this
more than 100 milliwatts. On the other hand, we
don’t want to have any data sending out, because
again, to save the power, so the memory that is on the processor is required to be very limited. Much more smaller than you
see it on the computer, so it’s only 1.6 megabit. So together all these are requirements with the real time requirements of getting this wheelchair driving. It’s a lot of challenge
to put everything crunch to this headset processor. In my lab, me and my team are excited to develop the next generation of low power and programmable and scalable software and hardware processing engines that can be used for variety of the emotional learning applications. So some of these applications
are spanning from medical, autonomous
systems, assistive devices, stress and behavior monitoring and even environment, for example, for air pollution monitoring. I give one example here, so one of the projects
that we are leading is for brain signal processing, in order to measure the
human cognitive state and basically their focus and their stress and eventually use this brain signals to control the environment. So again, this is a wearable device and it’s basically require
a lot of processing and also low power consumption and for example for this example, we are looking at the user performed the commands for
this particular application every half second. However, there is another application that again, my other team
of students is looking at, is much more higher
computation completely, which is for driving autonomous drones. So these devices, they require to perform a lot of computations
within 30 milliseconds doing the image and frame per second, with very very limited power. So the AI processor that you’re developing is basically responsible
to address all these different applications requirements. And we recently built a new chip, an AI accelerator chip, which is basically tiny
enough that you can fit it in a quarter coin
and still addressing all this processing
application requirements. And we are looking at basically the device being fit
in one of the headsets that I showed to you, to basically perform one
of the brain applications of your looking. And at the end, I would like to thank my team all different generations, that they have basically
done so much great work on these exciting projects and making all these exciting projects moving forward and coming up with better solutions. Thank you. (audience applause)

GRIT-X 2019: Tinoosh Mohsenin
Tagged on:         

Leave a Reply

Your email address will not be published. Required fields are marked *