Applying AI to Real World Use Cases – MIT AI Conference 2019

Machine generated transcript…

Hi everybody, my name is Julie, choy and I am Im working in Tel Im. A VP in the AI products group very happy to be here today on this lovely Saturday. Thank you all for taking time out of your busy schedules to come and learn more about AI today, Im going to walk through some really exciting use cases, and just what were seeing as Intel in this AI journey. First, I just like to introduce myself as

A proud alum of MIT yeah, its been a while, since I was at MIT, its been around 20 years Im class of 99 management science course 15. For those of you. That might know MIT speak. Do we have any course, 15 people in the room all right represent those people are really important to know, because these are the people that will likely translate the benefits of the technology.

That the engineers are building to the rest of the world to buy right. One of the biggest things I learned from my MIT education was really about perseverance. You can solve anything. You just need to be able to stay up all night. Sometimes my first job out of the Institute was as a hacker, so I used to be hired to obtain root in to the fortune 500, which wasnt

That hard after that, I worked moved to the valley spent most of the past 20 years here in Silicon Valley and my day, job is Intel AI marketing, but my real job is mom to two amazing children, one of whom is in the room today. Thank you for coming and joining me son and today again Im here to talk about AI applications, and there are really three factors that are driving the explosive growth of AI today. The first is

algorithmic innovation the second is data and the third are advances in compute so lets walk through that quickly algorithmic innovation Im sure youve heard today from many people about this topic but when you think about AI and the explosive innovation that were seeing its really relatively a recent phenomenon right so seven years ago did we break through with Alec sous-chefs keys work with the Alex net model and since then there have.

Been a number of really amazing algorithm other algorithms developed for vision and speech and sparse data and its really just the beginning I think that? What were seeing is at least at Intel over the past three years that Ive. Been there, enterprises are finally starting to be very interested in how to apply AI its still early days and its really the algorithm that’s going to be at the heart of this transformation. The second factor contributing to AI application growth is, of course, data without data you’re not going to have

Algorithms right and so when we talk about data a lot of times you know we talk about the growth so, lets start there some project that by 2025 we will, have around 175 zettabytes of data worldwide well. What is 175 zettabytes? I asked my team to help me visualize what that would look like that is the equivalent of stacking, something like 25 trillion. Blu-Ray discs at maximum

capacity all the way from Stanford Hoover tower to the moon and taking that roundtrip around 23 times its an amount a massive amount of data and of course the amount of data is a big factor in feeding deep learning algorithms but AI algorithmic innovation is not just about amount of data its also about you know a variety of data as well as velocity of data and that’s something that I learned from Professor Stonebraker at MIT who maybe some of you took classes with the.

third factor contributing to AI growth is compute specifically the innovations in AI Hardware this is something that I do every day had Intel were working on innovations on the hardware and the full stack software that you need to enable that hardware and our friends at open AI which is one of the the leading think tanks in AI just released that the demand for AI compute specific.

for deep learning training is doubling around every three and a half months that’s a tremendous rate of growth and its a tremendous amount of demand for advances in compute so at the end of the day to create AI applications you need to think about this development cycle its about understanding your data being able to architect infrastructure and.

systems that can grapple that with that data and get the insights you need out of it and its about creating algorithms at scale and using the right type of tools and this is not a simple thing its a very complicated thing I think that over the past 20 years of working in tech most of the past ten Ive been working on developer tools this is like the most interesting and difficult problem area that Ive seen yet to solve and that’s whats driving a lot of the demand right that’s why so many enterprises are.

willing to invest because the payoff is is so huge and you know the the stakes are high for problem solving so lets talk about some of these applications so were going to walk through three application examples today of how AI is transforming the world I do have a few more videos and Im hoping the sound is resolved but my point here is that virtually every industry that you can think of is undergoing this transformation and its not just the number of industries its also the global nature of ai-ai-ai-ai application.

transformation and the potential for impact is truly a global phenomenon so lets walk through three examples from around the world will start in China with an example from healthcare specifically well take it well look at an example from the IRI health hospital system so iri hospitals are dedicated to vision and the exploration of technology to help with degenerative eye disease.

and intel partnered with ire health care on an AI solution that would assist the physicians in China with delivering insights to patients that were suffering from ophthalmic degeneration and one thing to note about China is that the number of physicians is actually its deficient compared to the number of population that requires help and so AI is incredibly useful in.

augmenting the physician there will be no replacement of physicians no robot doctors in China will be in augmentation and what we see here in this particular example is again that application life cycle here the data was taken from 5,000 patients and it was actually an image data of the inner lining of the eye from these 5,000 patients and this image data was used to train a model a convolutional neural network 26 layer squeeze net to be exact and using that model that was trained it was also inference in a private cloud so the deployment of this AI was through.

a private cloud and the ultimate value was delivering a solution to the physician again to help the physician diagnose these problems to 93% accuracy and this is much higher because the average accuracy for a physician in the field in China is roughly 70 to 80 percent accuracy so this increase from the AI is very helpful to the physician in diagnosing and treating the problems this is happening for 30,000 clinics.

across China reaching roughly 30 million patients its incredibly valuable and we see these types of medical imaging AI applications in many different hospital systems around the world this is just one example the second example is very different it comes from the UK from an organization called resonate and I do have a video and if it doesnt work Ill talk us through it but it of resonates vision technology director Darren wood and hes going to talk about.

luminate which is the platform that resonate has developed its an AI solution to transform railway transportation and logistics lets take a like millions of journeys had taken by rail every day a lot freight moves by rail its still very important to the economy and getting people to jobs getting people to friends and family the challenge for our operators is to deliver increasing capacity without.

being able to build new infrastructure so increasingly is about how you deliver more for less and increasingly how you deliver a great customer experience by minimizing delays were working with railway operators to see how we can use artificial intelligence to deliver a more proactive better managed railway our solution is called the luminate digital platform and a core part of that you see artificial intelligence algorithms its trained upon 10 years of performance data on the networks and is looking to spot patterns but when things.

are about to improve or deteriorate so that operators can intervene and take positive action Loon Lake doesnt just look at trains being late right now the key part looks at is where theyre going to be in the future in the next hour or so and looking at things around platforms where two trains might be occupying the same platform if illuminate traffic managers are able to make much better informed decisions to have a simulation environment in which they can test out their changes they can test them against performance redo the forecast say no with confidence.

of those decisions are much better weve been working with a European rail operator to try illuminate and weve been seeing some very encouraging results one example is when they had a large power outage at one of their main stations and the following day train during the wrong place and they made extensive use of the delayed to reschedule and get back to an acceptable level of performance okay thank you Darren so again Darren is the director of vision technologies at resonate and resonate has created an AI platform its a software product that’s.

delivering insights to railway systems across Europe ultimately railway and transportation systems across Europe care most about delays nobody likes the ways in the airport or in the train station or wherever and so theyre using the luminate platform to move trains around to optimize how many trains can be on existing railways and ultimately reduce delays and disruptions and increase customer satisfaction and the AI equation if you think about it in this case the data is coming from ten.

years of historical records on the movement of these of these rail railway machines were also getting data from real time gateways that are installed at the track level and detecting the motions and the movements of the trains were also looking at you know maintenance and issues with trains and ultimately creating algorithms that are predicting the delays and getting ahead of that so.

its a very interesting solution that is using on the compute side CPUs as well as Intel distribution of Python and other tools like math kernal library and other software kits to help derive the algorithm at the heart of the solution the third and the last example today comes from North America its actually a hybrid this is a company called who box which is a robotics company headquartered in Houston but.

also has operations in Brazil and what who box is doing is its leveraging AI to deliver independent mobility I have a very short video that kind of tees that up so lets watch this make sure the volume told me okay this is AI powering a worlds first face controlled wheelchair is AI on Intel all the chips all right so who box is an incredible example of transforming lives and giving people a whole new level of mobility right so they have three applications and one of.

them that we collaborated with the line is called the wheelie which is a wheelchair that is AI enabled whats happening here is that people that have lost the ability to move from the neck down and sometimes they may even have impairments in their facial expressions have traditionally been you know its been challenging to have independent mobility but what who box has done is created a system whereby we can add an AI kit that is computer vision enabled as well as using deep learning inference on device so its mini PC so the combination of a real sense camera and a.

mini PC on the wheelchair its capturing facial expression data so the data set here our facial expressions captured from the operators of the wheelchair and what we can produce is an algorithm that has 99 9 percent accuracy on eleven different facial expression types that can now be used to direct the motion of the wheelchair its a phenomenal application and truly transforming the lives of really 1% of the globe 1% of.

the globe uses wheelchair for mobility and this is the type of transformation that could not happen without deep learning and the data and the algorithm all working together so I just have one last question for everyone in this room you know at the end of the day there have never been as many tools as much data or as many problems as we have in.

the world and so I would just to ask what are you going to transform and Im really looking looking forward to seeing what you built thank you so much.


Comments

Leave a Reply

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