Blaise Agüera y Arcas: How computers are learning to be creative
블레즈 아구에라 이 아카스 (Blaise Agüera y Arcas): 컴퓨터가 창의력을 배우는 방법
Blaise Agüera y Arcas works on machine learning at Google. Previously a Distinguished Engineer at Microsoft, he has worked on augmented reality, mapping, wearable computing and natural user interfaces. Full bio
Double-click the English transcript below to play the video.
that works on machine intelligence;
이끌고 있습니다.
of making computers and devices
공학적으로 훈련시켜
that brains do.
interested in real brains
저희는 실제 뇌와 신경과학에
in the things that our brains do
우리의 뇌가 하는 일 중에
to the performance of computers.
부분에 대한 것입니다.
has been perception,
인식이 언급돼 왔습니다.
out there in the world --
for example, that our team makes,
저희 팀에서 한 일은
on Google Photos to become searchable,
뭐가 찍혔냐에 따라
out there into the world.
바꾸는 것입니다.
our work on machine perception
저희가 해온 일들은
with the world of machine creativity
had a penetrating insight
인식과 창의성 간의
between perception and creativity.
보았다고 생각합니다.
has a statue inside of it,
조각상을 가지고 있고
is to discover it."
조각가의 과업이다."
Michelangelo was getting at
is an act of imagination
and perceiving and imagining,
with a brief bit of history
뇌에 대한 연구의
the heart or the intestines,
about a brain by just looking at it,
이야기할 게 없기 때문입니다.
of this thing all kinds of fanciful names,
이름을 붙였습니다.
뜻은 "작은 새우"입니다.
doesn't tell us very much
실제로 무슨일을 하는지
developed some kind of insight
무슨 일이 일어나는지에 대해
Santiago Ramón y Cajal,
산티아고 라몬 이 카할입니다.
or render in very high contrast
their morphologies.
할 수 있게 한 사람입니다.
that he made of neurons
만든 그림들입니다.
of different sorts of cells,
세포를 볼 수 있습니다.
was quite new at this point.
얼마 안 된 때였습니다.
very, very long distances --
뻗을 수 있는데
to some people in the 19th century;
19세기 사람들은
were just getting underway.
of Ramón y Cajal's, like this one,
that Ramón y Cajal started.
끝내려고 노력하고 있습니다.
of Neuroscience.
조명한 것 입니다.
is about one cubic millimeter in size,
대략 1 입방 밀리미터이고
very small piece of it here.
보고 계신 것입니다.
tiny block of tissue.
of hair is about 100 microns.
100 미크론입니다.
much, much smaller
머리카락 한 가닥보다
electron microscopy slices,
나눈 일련의 조각들로
in 3D of neurons that look like these.
이렇게 복원할 수 있습니다.
style as Ramón y Cajal.
어느 정도 같습니다.
be able to see anything here.
구분할 수 없을 것입니다.
one neuron to another.
ahead of his time,
뇌의 이해에 대한 연구는
over the next few decades.
전기를 이용하는 것을 알아냈고
was advanced enough
experiments on live neurons
할 수 있게 되고
when computers were being invented,
바로 이때인데
of modeling the brain --
as Alan Turing called it,
라고 불렀습니다.
looked at Ramón y Cajal's drawing
어느날 라몬 이 카할의
imagery that comes from the eye.
처리하는 피질입니다.
like a circuit diagram.
마치 회로도처럼 보였습니다.
in McCulloch and Pitts's circuit diagram
많은 세부사항이 있지만
of computational elements
일련의 계산 요소를
one to the next in a cascade,
정보를 넘긴다는 것은
visual information would need to do.
해야 하는 일에 대해서 말이죠.
이렇게 말하는 것 입니다.
for us to do with our brains.
that for a computer,
아셔야 하는 것이
just a few years ago.
이런 것이 불가능했습니다.
this task is easy to do.
있는 것이 아닙니다.
and the word "bird,"
"새"라는 단어의 관계는
connected to each other
inside our visual cortices,
생물학적인 것이나
to have the capability
on the computer.
그릴 수 있습니다.
that actually looks like.
about as a first layer of neurons,
how it works in the eye --
눈으로 보는 과정으로 보면
after another layer of neurons,
다음 층으로 전달합니다.
of different weights.
시냅스로 모두 연결되어있습니다.
of all of those synapses.
properties of this network.
계산되는 것을 특징짓습니다.
or a small group of neurons
those three things --
in the neural network,
혹은 수조 개가 있습니다.
these synapses in the neural network.
농도를 말합니다.
is just a simple formula,
공식이라고 해봅시다.
안에 넣었습니다.
going on there, of course,
of mathematical operations.
수학적인 과정이기 때문입니다.
that if you have one equation,
한 공식에서
by knowing the other two things.
알 수 있다는 것입니다.
that the picture of a bird is a bird,
새를 구분하는 공식은
and w and x are known.
w와 x는 알려진 경우이죠
you know the pixels.
무엇인지는 알고 있습니다.
a relatively straightforward problem.
간단한 문제입니다
and you're done.
doing exactly that.
보여드리겠습니다
on a mobile phone,
실시간으로 돌아가는 것입니다.
amazing in its own right,
billions and trillions of operations
한다는 것 자체만으로도
picture of a bird,
"Yes, it's a bird,"
끝나는 것이 아니라
with a network of this sort.
종까지 분류하는 모습입니다.
and the y is the unknown.
y는 밝혀지지 않았습니다.
difficult part, of course,
얼버무리고 지나가고 있는데
do we figure out the w,
of solving for w,
with the simple equation
6=2 x w라고 하면
it's the inverse to multiplication,
곱셈을 역으로 계산한 것입니다.
very non-linear operation;
비선형 연산이고
to solve the equation
나누지 않고 해결할 방법을
is fairly straightforward.
a little algebra trick,
to the right-hand side of the equation.
about it as an error.
for w the right way,
정답이 나오면
to minimize the error,
오류를 최소화할 수 있습니다.
computers are very good at.
컴퓨터가 아주 잘하는 일이죠.
sort of play Marco Polo,
마르코 폴로같이 여행하면
successive approximations to w.
근사치를 얻어가는 것입니다.
but after about a dozen steps,
수십 단계가 지나면
which is close enough.
이는 충분히 근접한 값입니다.
a lot of known x's and known y's
through an iterative process.
알아내고 있습니다.
that we do our own learning.
과정과 같습니다.
this is not a bird."
라고 듣습니다.
for those neural connections.
신경 연결을 해결하는 것입니다.
x and w fixed to solve for y;
y를 구합니다.
Alex Mordvintsev, on our team,
알렉스 모드빈츠세프는
with what happens if we try solving for x,
실험하기로 했습니다.
조건에서 말이죠.
that you've trained on birds,
신경망이 구축된 상태에서
the same error-minimization procedure,
trained to recognize birds,
네트워크를 통해
generated entirely by a neural network
새를 인식할 수 있는
rather than solving for y,
by Mike Tyka in our group,
마이크 티카의 작품입니다.
"동물 행진"입니다.
of William Kentridge's artworks,
작품이 떠올랐습니다.
over the space of different animals,
다양한 동물들로 설정했습니다.
to recognize and distinguish
morph from one animal to another.
동물들이 변하는 그림이 나옵니다.
have tried reducing
표현했습니다.
out of the space of all things
지도를 만들었습니다.
over that entire surface,
you make a kind of map --
이런 지도를 만듭니다.
the network knows how to recognize.
모든 것의 시각적 지도입니다.
"armadillo" is right in that spot.
저기 "아르마딜로"가 있습니다.
of networks as well.
할 수 있습니다.
to recognize faces,
인식하도록 설계됬습니다.
in a y that says, "me,"
"저"를 넣었습니다.
psychedelic picture of me
사이키델릭한 제 사진을 만듭니다.
multiple points of view at once
to get rid of the ambiguity
한 모습에서 다른 모습으로
or another pose,
제거되었기 때문입니다.
another kind of lighting.
this sort of reconstruction,
of different points of view,
his own face as a guide image
가이드로 이용해
to reconstruct my own face.
제 얼굴을 만든 것입니다.
that optimization process.
more like a coherent face,
with a blank canvas
시작하지 않아도 됩니다.
that is itself already some other image.
x를 구해도 됩니다.
that is designed to categorize
man-made structures, animals ...
인조물이나 동물 등을 말이죠.
with just a picture of clouds,
구름 사진을 이용했습니다.
what it sees in the clouds.
무엇이 보이는지 구분합니다.
you spend looking at this,
will see in the clouds.
볼 수 있습니다.
to hallucinate into this,
네트워크로 환각을 만들면
zooms hallucinates, zooms.
환각을 만들고 확대했습니다.
of the network, I suppose,
is eating its own tail.
기본은 이렇습니다.
What do I think I see next?"
다음에는 무엇이 보이지?"
called "Higher Education" --
강연에서였습니다.
marijuana was legalized.
is not constrained.
말하면서 말이죠.
because they're really fun to look at.
흥미를 유발하기 위해서 입니다.
아티스트 로스 굿윈은
a camera that takes a picture,
사진을 찍는 사진기와
writes a poem using neural networks,
신경 네트워크를 이용해 시를 썼습니다.
has been trained
are very intimately connected.
연결되어 있습니다.
신경 네트워크 입니다.
things in the world,
만들어 낼 수 있습니다.
Michelangelo really did see
any being, any alien
perceptual acts of that sort
machinery that's used in both cases.
사용하기 때문이죠.
and creativity are by no means
결코 인간에 국한되지 않는다고
that can do exactly these sorts of things.
컴퓨터 모델을 만들었고
the brain is computational.
졌다는 것은 놀랄 일도 아닙니다.
in designing intelligent machinery.
설계하면서 시작되었습니다.
똑똑하게 만들지 말이죠.
of those early pioneers,
is not just about accounting
게임 할 때만 쓰는 것이
we modeled them after our minds.
to understand our own minds better
인간의 마음을 더 잘 이해하고
ABOUT THE SPEAKER
Blaise Agüera y Arcas - Software architectBlaise Agüera y Arcas works on machine learning at Google. Previously a Distinguished Engineer at Microsoft, he has worked on augmented reality, mapping, wearable computing and natural user interfaces.
Why you should listen
Blaise Agüera y Arcas is principal scientist at Google, where he leads a team working on machine intelligence for mobile devices. His group works extensively with deep neural nets for machine perception and distributed learning, and it also investigates so-called "connectomics" research, assessing maps of connections within the brain.
Agüera y Arcas' background is as multidimensional as the visions he helps create. In the 1990s, he authored patents on both video compression and 3D visualization techniques, and in 2001, he made an influential computational discovery that cast doubt on Gutenberg's role as the father of movable type.
He also created Seadragon (acquired by Microsoft in 2006), the visualization technology that gives Photosynth its amazingly smooth digital rendering and zoom capabilities. Photosynth itself is a vastly powerful piece of software capable of taking a wide variety of images, analyzing them for similarities, and grafting them together into an interactive three-dimensional space. This seamless patchwork of images can be viewed via multiple angles and magnifications, allowing us to look around corners or “fly” in for a (much) closer look. Simply put, it could utterly transform the way we experience digital images.
He joined Microsoft when Seadragon was acquired by Live Labs in 2006. Shortly after the acquisition of Seadragon, Agüera y Arcas directed his team in a collaboration with Microsoft Research and the University of Washington, leading to the first public previews of Photosynth several months later. His TED Talk on Seadragon and Photosynth in 2007 is rated one of TED's "most jaw-dropping." He returned to TED in 2010 to demo Bing’s augmented reality maps.
Fun fact: According to the author, Agüera y Arcas is the inspiration for the character Elgin in the 2012 best-selling novel Where'd You Go, Bernadette?
Blaise Agüera y Arcas | Speaker | TED.com