Placeholder Image

字幕表 動画を再生する

  • Throughout the morning we've heard

  • about the incredible promise of exploring intelligence.

  • The potential to deeply understand how humans learn.

  • To use that understanding together

  • with novel computational methods to create new algorithms,

  • and to apply those methods to virtually

  • every scientific discipline, into every aspect

  • of everyday life.

  • Finance, transportation, health care, social interactions.

  • One might be inclined to infer from this

  • that the impact of intelligent algorithms

  • are mostly still to come.

  • And that impact is high.

  • But in many cases, the impact is already here.

  • Tang Xiao'ou and his company sense time

  • are a wonderful example of that.

  • An MIT PhD graduate from 1996, Xiao'ou

  • has decades of experience in computer vision and machine

  • learning.

  • I had the pleasure of serving on his thesis committee.

  • His supervisor was one of my first students,

  • and I'd been delighted to stay in contact with Xiao'ou

  • ever since.

  • Currently professor at the Chinese university of Hong Kong

  • and a former lead researcher at Microsoft Asia,

  • Xiao'ou is one of the most visible and influential leaders

  • in artificial intelligence, both within China

  • and internationally.

  • Since time, Hong Kong's first unicorn

  • produced his world leading systems

  • for face recognition and verification,

  • and other video analytics.

  • By building on decades of careful academic research,

  • and serves as a great example of how robust

  • scientific development of methods

  • can have impact in real world settings.

  • Xiao'ou is not only conducting world class academic research

  • and building a cutting edge company,

  • he's also wonderfully maintaining

  • his longstanding and close ties to MIT.

  • And thus I'm pleased to note yesterday's announcement

  • of the MIT sense time alliance on artificial intelligence.

  • A partnership that will open up new avenues of discovery

  • across MIT, in areas like computer vision,

  • human intelligence inspired algorithms, medical imaging

  • and robotics.

  • Will drive technological breakthroughs in AI

  • that have the potential to confront

  • some of the great challenges of the world.

  • And empower MIT and faculty, both faculty and students,

  • to pursue interdisciplinary projects at the vanguard

  • of intelligence research.

  • I'm thus delighted to introduce to you Tang Xiao'ou.

  • Thank you.

  • Thank you Eric and thank you MIT for giving

  • me the this opportunity to speak on this stage.

  • It's truly nice to be back home.

  • It's a lot colder than in Hong Kong, but I do feel warm.

  • I mean, I talk to Eric, I talk to Anessa, I talk to Josh.

  • I have this warm feeling, it's because of the air

  • conditioning.

  • So first of all, you know it's such a great honor

  • for this time to form this alliance with MIT,

  • and to be part of this MIT intelligent quest.

  • I think together we will definitely

  • go beyond just deep learning.

  • We will go to the uncharted territory of deep thinking.

  • Before I dive into that, let me start with something less deep.

  • Well I always start my talk with this picture for two reasons.

  • One is he's my son Samuel.

  • The second is he's just handsome.

  • And today, I do have a real reason.

  • In a few years, he's going to apply to college.

  • And, you know, perhaps MIT.

  • And we have so many professors today here, I just

  • want you to remember this face.

  • Yes I'm just joking.

  • Actually his math score is not very good.

  • It's not terrible, but it's kind of scary.

  • So I don't think he will make it into MIT,

  • we will settle for Harvard.

  • So enough kidding.

  • Let me talk about something something serious.

  • Money.

  • OK.

  • So this is a global top 10 box office last year

  • for all the movies.

  • And I read they are close to $1 billion each.

  • What is amazing about this picture is

  • that there is one movie among them is from China.

  • It's Wall of Warriors is the one right in the middle.

  • It actually beat out Wonder Woman and Paris of Caribbean.

  • It's an $850 million US dollar, which is amazing numbers.

  • I mean just 20 years ago, if we have 50 million,

  • that's a big movie.

  • And now it's $850 million, and it's a Chinese movie.

  • And all the other movies on this chart is global box office.

  • And this one, even though it's also global box office, but 99%

  • is from China.

  • Why?

  • Why can this movie make so much money?

  • Is it good really a great movie?

  • I actually watched it on the airplane.

  • And I think it's OK.

  • I have to say it's reasonably reasonable.

  • Why did it make so much money?

  • Because 20 years ago, nobody in China

  • will pay for a ticket, not nobody,

  • few will pay for a ticket to go to a theater to watch a movie.

  • Because you can't find a pirate copy anywhere easily.

  • So are not paying, so why make a movie.

  • But nowadays, people are going to the theaters.

  • They are paying.

  • So when you have the money, then creativity will follow.

  • I think that's the reason why we have such a great breakthrough.

  • And it's not just a one time thing.

  • It's actually this year, you know,

  • we are breaking the record again with several movies come out,

  • just making just a huge amount of money.

  • So when you have this protection of IP,

  • then you have the creativity and all the new things follow.

  • So let's come back to today's topic.

  • Today's topic is AI.

  • Creativity and AI.

  • So when I say the word AI, you know, what are you thinking?

  • Which company comes to your mind instantly?

  • I think we must be all thinking about save the company.

  • You know, Eric is smiling.

  • You know, great minds think alike.

  • Yes it's CSI.

  • Thank you.

  • Well instead of laughter, I'm looking for applause here.

  • [APPLAUSE]

  • Again, thank you for the sympathy.

  • So yes, it's Google.

  • Why Google?

  • Because Google did spend the money in tools.

  • 15 is spotted for R&D, is 12 Billion.

  • You know, they threw away 12 billion dollar just

  • to do something that may work or may not work.

  • And then in 2014, they bought a company called DeepMind.

  • $606 million, 12 people, no product.

  • Just play.

  • Using deep learning to play but OK.

  • They could have just hired people one by one.

  • That way is much cheaper.

  • But they did spend $606 million.

  • That's the value for the people, for the talents.

  • And if they have not done that, we will not be of our goal.

  • And Apple Go really is amazing breakthrough,

  • and it advocated everyone on AI.

  • So what do you do?

  • How do you follow up with Apple Go.

  • Well, Google followed with Apple Go 2, the Apple Go 0.

  • What about other companies?

  • I think in China, quite a number of companies

  • are trying to develop their own go playing,

  • deploying our wisdoms.

  • But if you do that, you are just following the footstep

  • of what Google is doing.

  • In the ancient time before we have glass,

  • in China on the windows they had paper, paper windows.

  • So we have a saying, you poke a hole on the paper window,

  • then you will see what is inside that room.

  • So what Apple Go did is poke a hole.

  • Then we start to see a lot of amazing things.

  • So if you follow that with another game,

  • another learning network, all you

  • are doing is just poking another hole on the window.

  • You are not seeing anything new.

  • You're just seeing it from a slightly different angle.

  • So nothing really being created.

  • But what is really important is what you follow

  • and also what you have done before Apple Go.

  • So in 2014, we did something that

  • is not the same, but similar.

  • We developed a computer vision, [INAUDIBLE]

  • that can recognize a human face better

  • than a human being's eye.

  • So we break that record.

  • We beat Facebook to the punch and did it first.

  • And in the year follows, when we beat the human eye,

  • the accuracy we did is 98.5.

  • The human eye is 97.5.

  • And then next year we improved it to 99.5

  • with 300 face training, and a year later, we

  • improved it to one over a million accuracy,

  • using 16 million data for training.

  • Then a year later, we used two billion face data for training.

  • We reached the one over a 100 million

  • accuracy, which is equivalent to eight-digit password.

  • So we worked with Qualcomm from this alliance on AI.

  • And all this great work is done by MIT alumni.

  • Professor Xiaogang Wang is a student

  • of Professor Eric Grimson.

  • Also my student, a master's student, he came by MIT

  • and went back to CUHK as a professor.

  • A professor [INAUDIBLE],, the same, did a master in CUHK,

  • then come to MIT.

  • Worked with Eric.

  • Finished his PhD.

  • Now he's a professor in CUHK.

  • Now let me just show a few work we have done in our lab.

  • The first is to use pictures, to look at the pictures.

  • Use face expression and also the gestures of the people,

  • to people in the picture, to test their relationship.

  • We can do this based on expression

  • or based on gestures.

  • What is it useful for?

  • Because we can just take your picture from on your website

  • and see which picture you're taking

  • and what is the relationship.

  • You have taken a picture with rich people,

  • then the bank will lend you money easier.

  • If you take a picture with a criminal, then bad luck.

  • So we actually can do this in real time for videos.

  • But those two do not know how to throw a party.

  • [INTERPOSING VOICES]

  • So the red one is competitive and the green one

  • is a friendly.

  • You can see when they're arguing the red one will come up.

  • When they are not then [INAUDIBLE] in real time.

  • And I ask my students to try this on my kids' photos,

  • because my kids has a lot of girlfriends.

  • I just want to find out which one is real.

  • [LAUGHTER]

  • You know, the result come back.

  • They're all real.

  • But I just tell him that no matter how many friends you

  • have in the daytime, in the evening,

  • you can only take one home.

  • [LAUGHTER]

  • So I think it is a sensitive subject for dating.

  • Let me switch to another application, which

  • is the less sensitive--

  • politics.

  • [LAUGHTER]

  • I'm not from Russia.

  • So I want to help, not interfere.

  • I think in America, I've observed that sometimes people

  • will agree with an idea, but they don't like the messenger.

  • They like the message, not the messenger.

  • So with available technology that can help--

  • so if you don't like the messenger,

  • we can just do the face swap.

  • Change you to someone you like, then perhaps you will help.

  • And of course, we do it for both sides.

  • So if you don't like this messenger,

  • then we can switch you to the other messenger.

  • So this is all done in real time.

  • We can also even use your own face.

  • And the next application--

  • [MUSIC PLAYING]

  • We're watching the Olympics.

  • I think that for a long time, you just cannot see the real

  • action.

  • But what we can do using this technology

  • is to really pick up the highlights so

  • that you don't have to sit there for three hours.

  • Then you just-- nine minutes, then you watch other things.

  • And this one is--

  • we can

  • [MUSIC PLAYING]

  • search for other kinds of videos that you like.

  • For example, this is for disaster videos, and

  • [INTENSE MUSIC]

  • --this is for martial arts and for different types of movies.

  • And this one is trying to use natural language processing

  • to describe the scene and then come up

  • with a scene of the movie.

  • I apologize for this particular movie I am using.

  • I don't have time to change.

  • So this is Francis.

  • He's wearing a [INAUDIBLE] suit and a tie,

  • and he's seating himself beside Claire.

  • So let's pick up the scenes from the movie.

  • The next one is using natural language to describe the scene.

  • So basically it's like people are describing sports events

  • using machine.

  • A player in the blue shoots.

  • A player in white knocks the ball with a [INAUDIBLE]..

  • And this one is a movie.

  • It's called, Once Upon a Time in America.

  • So in this movie, we can detect everything

  • in the scene, including the tables, the players,

  • [MUSIC PLAYING]

  • --the actors, actress, almost everything in this.

  • So you this technology we can--

  • [FIGHTING SOUNDS]

  • --analyze the movie.

  • [INAUDIBLE] with description of what is happening and also

  • is it drama?

  • Is it action?

  • And everything.

  • [DARK MUSIC]

  • And this is, when we are watching the movie,

  • the machines tells what kind of action this is.

  • Is this action?

  • This part is action?

  • Oh, this part is a romantic.

  • [RANDOM SOUNDS]

  • And this one is picking out exciting moments in the movie.

  • [FIGHTING SOUNDS]

  • Because of the time, I just skip this.

  • [MUSIC PLAYING]

  • So basically what we are doing, we

  • are teaching the machines to watch movies for humans.

  • So the mind is teaching machine to play golf

  • and we have people teaching a machine

  • to play poker, play game, and we are now

  • teaching them to watch a movie.

  • So the machine is doing all the fun stuff.

  • What we do-- we just do the hard work of research.

  • I think something is wrong here.

  • [LAUGHTER]

  • So let me come back just to today's topic.

  • I think instead of teaching--

  • the machine we have now, it's really capable of learning.

  • It's not thinking.

  • So I was with Josh, he talked today,

  • he said, in our lifetime, in his lifetime,

  • perhaps we will never have a machine who can think.

  • That maybe 20 years or maybe even in my lifetime,

  • in 50 years I can probably never see this.

  • But you start off really to come up

  • with a machine who can think, who is intelligent.

  • Perhaps we can think by ourself.

  • We can think to to build an environment that can really

  • help scientists to solve the problem in AI.

  • And how to help scientists-- what scientists need the most.

  • As a scientist myself, I can tell you the only thing

  • we need is the money.

  • So just give us the money, and we will solve the problem.

  • So thank you, MIT, for giving me the opportunity

  • to learn from you when I was a student.

  • And thank you for giving me the opportunity

  • to think with you together in this IQ campaign.

  • Thank you very much.

  • [APPLAUSE]

Throughout the morning we've heard

字幕と単語

ワンタップで英和辞典検索 単語をクリックすると、意味が表示されます

A2 初級

MITのインテリジェンス・クエストがローンチ注目のイノベーター (MIT Intelligence Quest Launch: Featured Innovator)

  • 9 2
    Jacob Mei に公開 2021 年 01 月 14 日
動画の中の単語