​英语作文范文：Teaching Machines to Learn on Their Own

英语作文范文：Teaching Machines to Learn on Their Own

Welcome to Scientific American’ Science Talk, poted on November 10th, 2015. I’m Steve Mirky. A hort epiode today. For which I’ll turn it over now to Scientific American’ aociate tech editor Larry Greenemeier:

Computer have alway been good at doing thing that are really complicated for u human. Thing like crunching inanely large number and running complex algorithm. On the other hand, computer have a really hard time recognizing a particular voice or face in a crowd, omething mot kid learn to do before they’re even out of diaper.

But thing are changing fat. Over the next decade or o, machine will more eaily mimi

c inherently human abilitie. And they’ll learn to do it much the ame way we do—through experience.

“Experience” in thi cae mean computer will be fed data pattern over and over again until they’re able to automatically identify a particular ound or image on their own. Thi proce i called machine learning.

To better undertand the dawn of intelligent machine and what it mean for our daily live, I poke with Stephen Hoover, CEO of Xerox’ Palo Alto Reearch Center, at a recent Intelligent Aitant conference in New York City. Here’ an edited verion of our converation. We tart by talking about the ongoing rapid change in machine learning.

SH: “That’ what’ really changed over the lat five year, i that computer now have the ability to undertand in much deeper way what it i that we’re aking and trying to do…You’re actually tarting to ee it be embedded into product. So, many of your reader will be familiar with the Net thermotat. The Net thermotat i a product with an intelligent agent built into it…So I don’t program a thermotat now, right? It learn from my behavior what it i that I’m doing, undertanding my context and then delivering to me the experience that I want, o it i an intelligent agent that’ built into the product…And more and more you’re going to ee thee kind of capabilitie built into product.”

LG: What role doe machine learning play?

SH: “You don’t o much program a computer in machine learning in the way you did, which wa I broke down a tak into a erie of tep to do that. It’ that machine learning actually learn, from the data, the right anwer. The machine program itelf.”

LG: Kind of like the way human learn a kid.

SH: “You how your daughter a car in a book, you how her another one. You ay car, car, car. She ay car. They learn by labeled data…Did we program our child to recognize a car? I mean, in ome ene you did but that’ what machine learning i. You’re going to how the computer a bunch of intance and you’re going to label it, and it’ going to learn how to do it…There’ a core code which i that learning algorithm, and then that’ applied to multiple context…We’re witching from where computer helped people to people helping computer.”

LG: You can’t talk about machine learning without alo mentioning the hardware that make it poible.

SH: “Computer can do thing that ued to be hard for computer but are eay for human, like recognizing a mile. That’ becaue Moore’ Law ha enabled it…And Moore’ Law mean the doubling of computational power every 18 month. Obviouly that mean I can write more and more complex oftware.”

Like app on a martphone?

SH: “In your iPhone, right, which by the way thi i a powerful a a Kray upercomputer from 1998 that modeled the weather for the entire world. That’ what you carry around in your pocket…But one of the amazing thing in there, think about it. It’ like a $15 chip that ha GPS, an accelerometer, a preure enor. GPS. You can tell where you’re at in the world to within a meter. Anywhere in the world, for $15. One time buy. It’ phenomenal. So, hardware that ene the world to help then oftware make better deciion i really important. That’ thi whole idea of the Internet of Thing…

“The Internet of Thing, the analogy I make for people i you think about Google. What Google and earch did i, it enlarged the human memory, right? I ued to have to know everything if I wanted to acce it in le than glacial time, make a trip to the library and look up the card catalog. I don’t have to know anything today from a fact viewpoint. When I ay know anything, I mean I have to memorize a et of fact. I jut go on Google to do it. My mind ha been expanded to be a large a all human knowledge. The Internet of Thing i about Googling reality.

“What I mean i think about it a if my body i now a big a the globe. I want to know what the temperature i in Auguta, Maine. I want to know what the tate of pollution i in Beijing. I want to know i there freh fih today at Whole Food. Senor are going to be in the world that are going to tell me thoe thing. And o hardware not only beget the capability to create new kind of oftware like machine learning, but alo i creating new way to ene, meaure and control the world. And that feedback loop i again one of the big change that we’re going to ee coming.”

SM: That’ it for thi hort Science Talk. In the coming day we’ll have interview with the author of three new book about math, hore and Parkinon’ dieae, and lot more. Meanwhile, get your cience new at our webite, www.cientificamerican.com. Where you can alo check out the November iue of the magazine, including a long-planned article about how the contruction of Egypt' Great Pyramid changed civilization. Who know, it could come up in a Preidential debate. Although that might go againt the grain.

And follow u on Twitter, where you’ll get a tweet whenever a new item hit the Web ite. Our twitter name i @ciam. For Scientific American’ Science Talk, I’m Steve Mirky, thank for clicking on u.