Page 23 - ARUBA TODAY

A23
TECHNOLOGY Tuesday 22 augusT 2017

Why AI visionary Andrew Ng teaches humans to teach computers

By RYAN NAKASHIMA times.
AP Technology Writer Geoffrey Hinton, whose
PALO ALTO, Calif. (AP) — University of Toronto team
Andrew Ng has led teams wowed peers by using a
at Google and Baidu that neural network to win the
have gone on to create prestigious ImageNet com-
self-learning computer pro- petition in 2012, credits Ng
grams used by hundreds of with persuading him to use
millions of people, includ- the technique. That win
ing email spam filters and spawned a flurry of copy-
touch-screen keyboards cats, giving birth to the rise
that make typing easier by of modern AI.
predicting what you might “Several different people
want to say next. suggested using GPUs,”
As a way to get machines Hinton says by email. But
to learn without supervi- the work by Ng’s team, he
sion, he has trained them says, “was what convinced
to recognize cats in You- me.”
Tube videos without be- TEACHING HOW TO TEACH
ing told what cats were. COMPUTERS
And he revolutionized this Ng’s fascination with AI
field, known as artificial in- was paralleled by a de-
telligence, by adopting sire to share his knowledge
graphics chips meant for In this Friday, July 14, 2017, photo, computer scientist Andrew Ng, right, works with others at his with students. As online ed-
video games. office in Palo Alto, Calif. ucation took off earlier this
To push the boundaries of Associated Press decade, Ng discovered a
artificial intelligence fur- natural outlet.
ther, one of the world’s changed is his hairdresser’s a “neural network” — the Ng’s standout AI work in- His “Machine Learning”
most renowned research- — to which a friend of his core computing engine of volved finding a new way course, which kicked off
ers in the field says many responded that in fact, she artificial intelligence mod- to supercharge neural net- Stanford’s online learn-
more humans need to get could get a robot to do his eled on the human brain. works using chips most of- ing program alongside
involved. So his focus now hair. “It seemed really amaz- ten found in video-game two other courses in 2011,
is on teaching the next At the end of a 90-minute ing that you could write a machines. immediately signed up
generation of AI specialists interview in his sparse office few lines of code and have Until then, computer scien- 100,000 people without
to teach the machines. in Palo Alto, California, he it learn to do interesting tists had mostly relied on any marketing effort.
Nearly 2 million people reveals what’s partially be- things,” he said. general-purpose proces- A year later, he co-found-
around the globe have hind his ambition. After graduating high sors — like the Intel chips ed the online-learning
taken Ng’s online course “Life is shockingly short,” school from Singapore’s that still run many PCs. Such startup Coursera. More re-
on machine learning. In his the 41-year-old computer Raffles Institution, Ng made chips can handle only a cently, he left his high-pro-
videos, the lanky, 6-foot-1 scientist says, swiveling the rounds of Carnegie few computing tasks simul- file job at Baidu to launch
Briton of Hong Kong and his laptop into view. He’s Mellon, MIT and Berkeley taneously, but make up for deeplearning.ai , a startup
Singaporean upbringing calculated in a Chrome before taking up residence it with blazing speed. Neu- that produces AI-training
speaks with a difficult- browser window how many as a professor at Stanford ral networks, however, work courses.
to-place accent . He of- days we have from birth to University. much better if they can run Every time he’s started
ten tries to get students death: a little more than There, he taught robotic thousands of calculations something big, whether
comfortable with mind- 27,000. “I don’t want to helicopters to do aerial ac- simultaneously. That turned it’s Coursera, the Google
boggling concepts by waste that many days.” robatics after being trained out to be a task eminently Brain deep learning unit, or
acknowledging up front, BUILDING BRAINS AS A TEEN by an expert pilot. The work suited for a different class Baidu’s AI lab, he has left
in essence, that “hey, this An upstart programmer by was “inspiring and excit- of chips called graphics once he felt the teams he
stuff is tough.” age 6, Ng learned cod- ing,” recalls Pieter Abbeel, processing units, or GPUs. has built can carry on with-
Ng sees AI as a way to ing early from his father, a then one of Ng’s doctoral So when graphics chip out him.
“free humanity from repeti- medical doctor who tried students and now a com- maker Nvidia opened up “Then you go, ‘Great. It’s
tive mental drudgery.” He to program a computer puter scientist at Berkeley. its GPUs for general purpos- thriving with or without
has said he sees AI chang- to diagnose patients using Abbeel says he once es beyond video games me,’” says Ng, who con-
ing virtually every industry, data. “At his urging,” Ng crashed a $10,000 helicop- in 2007, Ng jumped on the tinues to teach at Stanford
and any task that takes less says, he fiddled with these ter drone, but Ng brushed technology. while working in private in-
than a second of thought concepts on his home it off. “Andrew was always His Stanford team began dustry.
will eventually be done computer. At age 16, he like, ‘If these things are too publishing papers on the For Ng, one of his next chal-
by machines. He once wrote a program to calcu- simple, everybody else technique a year later, lenges might include hav-
said famously that the late trigonometric functions could do them.’” speeding up machine ing a child with his roboti-
only job that might not be like sine and cosine using THE MARK OF NG learning by as much as 70 cist wife, Carol Reiley.q