kor_sent
stringlengths
1
5.4k
โŒ€
eng_sent
stringlengths
1
17.5k
โŒ€
source
stringclasses
4 values
similarity
float64
0.12
1
โŒ€
from
stringclasses
5 values
__index_level_0__
float64
0
1.78M
โŒ€
๋‚˜๋ˆ—์…ˆ์„ ์šฐ๋ฆฌ๊ฐ€ ๋‚˜๋ˆ—์…ˆ์„ ์ผ๋Š”๋ฐ ๊ณฑ์…ˆ์„ ์—ญ์œผ๋กœ ๊ณ„์‚ฐํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ๋ฐฉ๊ธˆ ๋งํ•œ ๋Œ€๋กœ ์‹ค์ œ ์—ฐ์‚ฐ์€ ๊ณฑํ•˜๊ธฐ๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค.
So, division -- we've used division because it's the inverse to multiplication, but as I've just said, the multiplication is a bit of a lie here.
IWSLT2017
null
null
null
์ด๊ฒƒ์€ ๋งค์šฐ ๋งค์šฐ ๋ณต์žกํ•œ ๋น„์„ ํ˜• ์—ฐ์‚ฐ์ด๊ณ  ์—ญ์œผ๋กœ ๊ณ„์‚ฐํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค.
This is a very, very complicated, very non-linear operation; it has no inverse.
IWSLT2017
null
null
null
๊ทธ๋ž˜์„œ ์šฐ๋ฆฌ๋Š” ์ด ๊ณต์‹์„ ๋‚˜๋ˆ„์ง€ ์•Š๊ณ  ํ•ด๊ฒฐํ•  ๋ฐฉ๋ฒ•์„ ์ฐพ์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค.
So we have to figure out a way to solve the equation without a division operator.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ๊ทธ ๋ฐฉ๋ฒ•์€ ๋งค์šฐ ๊ฐ„๋‹จํ•ฉ๋‹ˆ๋‹ค.
And the way to do that is fairly straightforward.
IWSLT2017
null
null
null
๋Œ€์ˆ˜ํ•™์„ ์กฐ๊ธˆ ์ด์šฉํ•ด 6์„ ๊ณต์‹์˜ ์šฐ๋ณ€์œผ๋กœ ์˜ฎ๊ธฐ๊ฒ ์Šต๋‹ˆ๋‹ค.
You just say, let's play a little algebra trick, and move the six over to the right-hand side of the equation.
IWSLT2017
null
null
null
์ด๋Ÿฌ๋ฉด ๊ณฑํ•˜๊ธฐ๋งŒ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Now, we're still using multiplication.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  0์€ ์˜ค๋ฅ˜๋ผ๊ณ  ์ƒ๊ฐํ•ฉ์‹œ๋‹ค.
And that zero -- let's think about it as an error.
IWSLT2017
null
null
null
๋‹ค์‹œ ๋งํ•ด, ์šฐ๋ฆฌ๊ฐ€ w๋ฅผ ํ•ด๊ฒฐํ•ด์„œ ์ •๋‹ต์ด ๋‚˜์˜ค๋ฉด ์˜ค๋ฅ˜๊ฐ€ 0์ด ๋  ๊ฒƒ์ด๊ณ 
In other words, if we've solved for w the right way, then the error will be zero.
IWSLT2017
null
null
null
์šฐ๋ฆฌ๊ฐ€ ์ž˜๋ชป๋œ ๊ฐ’์„ ๊ตฌํ–ˆ๋‹ค๋ฉด ์˜ค๋ฅ˜๊ฐ€ 0๋ณด๋‹ค ์ปค์งˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
And if we haven't gotten it quite right, the error will be greater than zero.
IWSLT2017
null
null
null
์ด์ œ ์šฐ๋ฆฌ๊ฐ€ ์ถ”์ธกํ•ด์„œ ์˜ค๋ฅ˜๋ฅผ ์ตœ์†Œํ™”ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ด๋Ÿฐ ๊ฒƒ์€ ์ปดํ“จํ„ฐ๊ฐ€ ์•„์ฃผ ์ž˜ํ•˜๋Š” ์ผ์ด์ฃ .
So now we can just take guesses to minimize the error, and that's the sort of thing computers are very good at.
IWSLT2017
null
null
null
๊ทธ๋ž˜์„œ ์ตœ์ดˆ์˜ ์ถ”์ธก์œผ๋กœ w๊ฐ€ 0์ด๋ผ๋ฉด
So you've taken an initial guess: what if w = 0?
IWSLT2017
null
null
null
์˜ค๋ฅ˜๋Š” 6์ž…๋‹ˆ๋‹ค.
Well, then the error is 6.
IWSLT2017
null
null
null
w๊ฐ€ 1์ด๋ฉด ์˜ค๋ฅ˜๋Š” 4์ž…๋‹ˆ๋‹ค.
What if w = 1? The error is 4.
IWSLT2017
null
null
null
์ปดํ“จํ„ฐ๊ฐ€ ๊ณ„์† ๋งˆ๋ฅด์ฝ” ํด๋กœ๊ฐ™์ด ์—ฌํ–‰ํ•˜๋ฉด ์˜ค๋ฅ˜๊ฐ€ 0์— ๊ฐ€๊นŒ์›Œ์งˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
And then the computer can sort of play Marco Polo, and drive down the error close to zero.
IWSLT2017
null
null
null
๊ทธ๋Ÿฌ๋ฉด์„œ ์ปดํ“จํ„ฐ๊ฐ€ ์„ฑ๊ณต์ ์œผ๋กœ w ๊ฐ’์˜ ๊ทผ์‚ฌ์น˜๋ฅผ ์–ป์–ด๊ฐ€๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
As it does that, it's getting successive approximations to w.
IWSLT2017
null
null
null
์ „ํ˜•์ ์œผ๋กœ ์ •ํ™•ํ•œ ๊ฐ’์„ ์–ป์ง„ ๋ชปํ•˜์ง€๋งŒ ์ˆ˜์‹ญ ๋‹จ๊ณ„๊ฐ€ ์ง€๋‚˜๋ฉด w๋Š” 2.999๋ฅผ ์–ป๊ฒŒ ๋˜๊ณ  ์ด๋Š” ์ถฉ๋ถ„ํžˆ ๊ทผ์ ‘ํ•œ ๊ฐ’์ž…๋‹ˆ๋‹ค.
Typically, it never quite gets there, but after about a dozen steps, we're up to w = 2.999, which is close enough.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ด๊ฒƒ์ด ํ•™์Šต ๊ณผ์ •์ž…๋‹ˆ๋‹ค.
And this is the learning process.
IWSLT2017
null
null
null
์ง€๊ธˆ๊นŒ์ง€ ์ด์•ผ๊ธฐํ•œ ๊ฒƒ์€ ์ˆ˜๋งŽ์€ x์™€ y ๊ฐ’์„ ์•Œ๊ณ  ์žˆ๊ณ  ๊ฐ€์šด๋ฐ w ๊ฐ’์„ ์ถ”๋ก  ๊ณผ์ •์—์„œ ์•Œ์•„๋‚ด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
So remember that what's been going on here is that we've been taking a lot of known x's and known y's and solving for the w in the middle through an iterative process.
IWSLT2017
null
null
null
์ด๋Š” ์šฐ๋ฆฌ์˜ ๋‡Œ๊ฐ€ ํ•™์Šตํ•˜๋Š” ๊ณผ์ •๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.
It's exactly the same way that we do our own learning.
IWSLT2017
null
null
null
์šฐ๋ฆฌ๋Š” ์–ด๋ฆด ์  ์ˆ˜๋งŽ์€ ์ด๋ฏธ์ง€๋ฅผ ์ ‘ํ•˜๊ณ  "์ด๊ฒƒ์€ ์ƒˆ๋‹ค, ์ด๊ฒƒ์€ ์ƒˆ๊ฐ€ ์•„๋‹ˆ๋‹ค" ๋ผ๊ณ  ๋“ฃ์Šต๋‹ˆ๋‹ค.
We have many, many images as babies and we get told, "This is a bird; this is not a bird."
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์‹œ๊ฐ„์ด ํ˜๋Ÿฌ ๋ฐ˜๋ณตํ•˜๋ฉด์„œ w๋ฅผ ์•Œ์•„๋‚ด์ฃ . ์‹ ๊ฒฝ ์—ฐ๊ฒฐ์„ ํ•ด๊ฒฐํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
And over time, through iteration, we solve for w, we solve for those neural connections.
IWSLT2017
null
null
null
์ด์ œ ์šฐ๋ฆฌ๋Š” ๊ณ ์ •๋œ x์™€ w๊ฐ’์œผ๋กœ y๋ฅผ ๊ตฌํ•ฉ๋‹ˆ๋‹ค. ์ด๊ฒƒ์€ ๋งค์ผ ์šฐ๋ฆฌ๊ฐ€ ํ•˜๋Š” ์ธ์‹์ž…๋‹ˆ๋‹ค.
So now, we've held x and w fixed to solve for y; that's everyday, fast perception.
IWSLT2017
null
null
null
w ๊ฐ’์„ ๊ตฌํ•˜๋Š” ๊ณผ์ •์€ ํ•™์Šต์ด๊ณ  ๋” ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ์™œ๋ƒ๋ฉด ๋งŽ์€ ํ›ˆ๋ จ ์˜ˆ์‹œ๋ฅผ ํ†ตํ•ด ์˜ค๋ฅ˜๋ฅผ ์ตœ์†Œํ™” ํ•ด์•ผ ํ•˜๊ธฐ ๋•Œ๋ฌธ์ด์ฃ .
We figure out how we can solve for w, that's learning, which is a lot harder, because we need to do error minimization, using a lot of training examples.
IWSLT2017
null
null
null
์•ฝ 1๋…„ ์ „์— ์ €ํฌ ํŒ€์˜ ์•Œ๋ ‰์Šค ๋ชจ๋“œ๋นˆ์ธ ์„ธํ”„๋Š” ์šฐ๋ฆฌ๊ฐ€ x๋ฅผ ๊ตฌํ•˜๋ฉด ์–ด๋–ป๊ฒŒ ๋˜๋Š”์ง€ ์‹คํ—˜ํ•˜๊ธฐ๋กœ ํ–ˆ์Šต๋‹ˆ๋‹ค. w์™€ y ๊ฐ’์„ ์•Œ๊ณ  ์žˆ๋‹ค๋Š” ์กฐ๊ฑด์—์„œ ๋ง์ด์ฃ .
And about a year ago, Alex Mordvintsev, on our team, decided to experiment with what happens if we try solving for x, given a known w and a known y.
IWSLT2017
null
null
null
๋‹ค์‹œ ๋งํ•˜์ž๋ฉด ์ƒˆ๋ผ๋Š” ๊ฒƒ์„ ์•Œ๊ณ  ์ƒˆ๋ผ๋Š” ๊ฒƒ์„ ์ธ์‹ํ•  ์ˆ˜ ์žˆ๋Š” ์‹ ๊ฒฝ๋ง์ด ๊ตฌ์ถ•๋œ ์ƒํƒœ์—์„œ ์ƒˆ์˜ ๋ชจ์Šต์„ ์•Œ์•„๋‚ด๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
In other words, you know that it's a bird, and you already have your neural network that you've trained on birds, but what is the picture of a bird?
IWSLT2017
null
null
null
๋˜‘๊ฐ™์€ ์˜ค๋ฅ˜ ์ตœ์†Œํ™” ๊ณผ์ •์„ ๊ฑฐ์ณ ์ปดํ“จํ„ฐ๊ฐ€ ์ƒˆ๋ฅผ ์ธ์‹ํ•  ์ˆ˜ ์žˆ๋Š” ๋„คํŠธ์›Œํฌ๋ฅผ ํ†ตํ•ด ๋งŒ๋“ค์–ด๋‚ธ ๊ฒฐ๊ณผ๋Š”
It turns out that by using exactly the same error-minimization procedure, one can do that with the network trained to recognize birds, and the result turns out to be ...
IWSLT2017
null
null
null
์ƒˆ์˜ ๊ทธ๋ฆผ์ž…๋‹ˆ๋‹ค.
a picture of birds.
IWSLT2017
null
null
null
์ด ๊ทธ๋ฆผ์€ ์ „์ ์œผ๋กœ ์ƒˆ๋ฅผ ์ธ์‹ํ•  ์ˆ˜ ์žˆ๋Š” ์‹ ๊ฒฝ ๋„คํŠธ์›Œํฌ๋ฅผ ํ†ตํ•ด y ๊ฐ’์„ ๊ตฌํ•˜๋Š” ๋Œ€์‹  x ๊ฐ’์„ ์ถ”๋ก ํ•˜์—ฌ ๊ตฌํ˜„๋ฌ์Šต๋‹ˆ๋‹ค.
So this is a picture of birds generated entirely by a neural network that was trained to recognize birds, just by solving for x rather than solving for y, and doing that iteratively.
IWSLT2017
null
null
null
๋‹ค๋ฅธ ์žฌ๋ฏธ์žˆ๋Š” ์˜ˆ๋ฅผ ๋ณด์—ฌ๋“œ๋ฆฌ๋ฉด
Here's another fun example.
IWSLT2017
null
null
null
์ด๊ฒƒ์€ ์ €ํฌ ๊ทธ๋ฃน์˜ ๋งˆ์ดํฌ ํ‹ฐ์นด์˜ ์ž‘ํ’ˆ์ž…๋‹ˆ๋‹ค. ์ด ์ž‘ํ’ˆ์˜ ์ œ๋ชฉ์€ "๋™๋ฌผ ํ–‰์ง„"์ž…๋‹ˆ๋‹ค.
This was a work made by Mike Tyka in our group, which he calls "Animal Parade."
IWSLT2017
null
null
null
์ด๊ฒƒ์„ ๋ณด๊ณ  ์œŒ๋ฆฌ์—„ ์ผ„ํŠธ๋ฆฌ์ง€์˜ ์ž‘ํ’ˆ์ด ๋– ์˜ฌ๋ž์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ์Šค์ผ€์น˜๋ฅผ ๊ทธ๋ ธ๋‹ค๊ฐ€ ์ง€์šฐ๊ณ  ๊ทธ๋ ธ๋‹ค๊ฐ€ ์ง€์›Œ๊ฐ€๋ฉฐ ์ด๋Ÿฐ ์‹์œผ๋กœ ์˜์ƒ์„ ๋งŒ๋“ค์ฃ .
It reminds me a little bit of William Kentridge's artworks, in which he makes sketches, rubs them out, makes sketches, rubs them out, and creates a movie this way.
IWSLT2017
null
null
null
์ด ๊ฒฝ์šฐ์—๋Š” ๋งˆ์ดํฌ๊ฐ€ ํ•œ ๊ฒƒ์€ ๋ณ€์ˆ˜ y๋ฅผ ๋‹ค์–‘ํ•œ ๋™๋ฌผ๋“ค๋กœ ์„ค์ •ํ–ˆ์Šต๋‹ˆ๋‹ค. ์„œ๋กœ ๋‹ค๋ฅธ ๋™๋ฌผ๋“ค์„ ๊ตฌ๋ถ„ํ•  ์ˆ˜ ์žˆ๋„๋ก ์„ค๊ณ„๋œ ๋„คํŠธ์›Œํฌ ์•ˆ์—์„œ ๋ง์ด์ฃ .
In this case, what Mike is doing is varying y over the space of different animals, in a network designed to recognize and distinguish different animals from each other.
IWSLT2017
null
null
null
๊ทธ๋ ‡๊ฒŒ ์ด๋Ÿฐ ํฌ์•ˆํ•œ ์—์…” ํ’์˜ ๋™๋ฌผ๋“ค์ด ๋ณ€ํ•˜๋Š” ๊ทธ๋ฆผ์ด ๋‚˜์˜ต๋‹ˆ๋‹ค.
And you get this strange, Escher-like morph from one animal to another.
IWSLT2017
null
null
null
์—ฌ๊ธฐ์„œ ๋งˆ์ดํฌ์™€ ์•Œ๋ ‰์Šค๋Š” y ๊ฐ’์„ ์ค„์—ฌ 2์ฐจ์› ํ‰๋ฉด์— ํ‘œํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ ‡๊ฒŒ ์ด ๋„คํŠธ์›Œํฌ๊ฐ€ ์ธ์‹ํ•  ์ˆ˜ ์žˆ๋Š” ๋ชจ๋“  ์ข…๋ฅ˜๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” ์ง€๋„๋ฅผ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
Here he and Alex together have tried reducing the y's to a space of only two dimensions, thereby making a map out of the space of all things recognized by this network.
IWSLT2017
null
null
null
์ด๋Ÿฐ ์ข…๋ฅ˜์˜ ์ด๋ฏธ์ง€ ํ†ตํ•ฉ ํ˜น์€ ์ƒ์„ฑ์€ ํ‘œ๋ฉด ์ „๋ฐ˜์— ๊ฑธ์ณ y๋ฅผ ๋‹ค๋ฅด๊ฒŒ ํ•ด์„œ ์ด๋Ÿฐ ์ง€๋„๋ฅผ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๋„คํŠธ์›Œํฌ๊ฐ€ ์ธ์‹ํ•˜๋Š” ๋ชจ๋“  ๊ฒƒ์˜ ์‹œ๊ฐ์  ์ง€๋„์ž…๋‹ˆ๋‹ค.
Doing this kind of synthesis or generation of imagery over that entire surface, varying y over the surface, you make a kind of map -- a visual map of all the things the network knows how to recognize.
IWSLT2017
null
null
null
๋ชจ๋“  ๋™๋ฌผ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ €๊ธฐ "์•„๋ฅด๋งˆ๋”œ๋กœ"๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
The animals are all here; "armadillo" is right in that spot.
IWSLT2017
null
null
null
์ด๊ฒƒ์„ ๋‹ค๋ฅธ ๋„คํŠธ์›Œํฌ๋กœ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
You can do this with other kinds of networks as well.
IWSLT2017
null
null
null
์ด ๋„คํŠธ์›Œํฌ๋Š” ์–ผ๊ตด์„ ์ธ์‹ํ•˜๋„๋ก ์„ค๊ณ„๋ฌ์Šต๋‹ˆ๋‹ค. ์„œ๋กœ ๋‹ค๋ฅธ ์–ผ๊ตด์„ ๊ตฌ๋ถ„ํ•˜๋„๋ก ๋ง์ด์ฃ .
This is a network designed to recognize faces, to distinguish one face from another.
IWSLT2017
null
null
null
์—ฌ๊ธฐ์„œ ์ €ํฌ๊ฐ€ y์— "์ €"๋ฅผ ๋„ฃ์—ˆ์Šต๋‹ˆ๋‹ค. ์ œ ์–ผ๊ตด์„ ๋ณ€์ˆ˜๋กœ ๋ง์ด์ฃ .
And here, we're putting in a y that says, "me," my own face parameters.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ด๊ฒƒ์ด x๋ฅผ ๊ตฌํ•˜๋ฉด ์ด๋Ÿฐ ์ƒ๋‹นํžˆ ์ •์‹ ์—†๊ณ  ์•ฝ๊ฐ„์€ ์ž…์ฒดํŒŒ, ์ดˆํ˜„์‹ค์ฃผ์˜, ์‚ฌ์ดํ‚ค๋ธ๋ฆญํ•œ ์ œ ์‚ฌ์ง„์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ์—ฌ๋Ÿฌ ๋ชจ์Šต์„ ํ•œ ๋ฒˆ์— ๋ณด์—ฌ์ฃผ๋ฉด์„œ์š”.
And when this thing solves for x, it generates this rather crazy, kind of cubist, surreal, psychedelic picture of me from multiple points of view at once.
IWSLT2017
null
null
null
์—ฌ๋Ÿฌ ๋ชจ์Šต์„ ํ•œ ๋ฒˆ์— ๋ณด์—ฌ์ฃผ๋Š” ์ด์œ ๋Š” ๋„คํŠธ์›Œํฌ์˜ ์„ค๊ณ„์—์„œ ์–ผ๊ตด์˜ ํ•œ ๋ชจ์Šต์—์„œ ๋‹ค๋ฅธ ๋ชจ์Šต์œผ๋กœ ๋„˜์–ด๊ฐ€๋Š” ๋ชจํ˜ธํ•œ ๊ณผ์ •์ด ์ œ๊ฑฐ๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ํŠน์ • ๊ฐ๋„์˜ ์–ผ๊ตด์„ ๋ณด๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
The reason it looks like multiple points of view at once is because that network is designed to get rid of the ambiguity of a face being in one pose or another pose, being looked at with one kind of lighting, another kind of lighting.
IWSLT2017
null
null
null
๊ทธ๋ž˜์„œ ์ด๊ฒƒ์„ ์žฌ๊ตฌ์„ฑํ•  ๋•Œ ๊ฐ€์ด๋“œ ์ด๋ฏธ์ง€๋‚˜ ํ†ต๊ณ„๋ฅผ ์‚ฌ์šฉํ•˜์ง€ ์•Š์œผ๋ฉด ์ด๋Ÿฐ ํ˜ผ๋ž€์Šค๋Ÿฌ์šด ์‹œ์ ๋“ค์ด ๋‚˜์˜ต๋‹ˆ๋‹ค. ๋ชจํ˜ธํ•˜๊ธฐ ๋–„๋ฌธ์ด์ฃ .
So when you do this sort of reconstruction, if you don't use some sort of guide image or guide statistics, then you'll get a sort of confusion of different points of view, because it's ambiguous.
IWSLT2017
null
null
null
์ด๊ฒƒ์€ ์•Œ๋ ‰์Šค๊ฐ€ ๋ณธ์ธ ์–ผ๊ตด์„ ๊ฐ€์ด๋“œ๋กœ ์ด์šฉํ•ด ์ตœ์ ํ™” ๊ณผ์ •์„ ๊ฑฐ์ณ ์ œ ์–ผ๊ตด์„ ๋งŒ๋“  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
This is what happens if Alex uses his own face as a guide image during that optimization process to reconstruct my own face.
IWSLT2017
null
null
null
๋ณด์‹œ๋‹ค์‹œํ”ผ ์™„๋ฒฝํ•˜์ง„ ์•Š์Šต๋‹ˆ๋‹ค.
So you can see it's not perfect.
IWSLT2017
null
null
null
์–ด๋–ป๊ฒŒ ์ตœ์ ํ™”๋ฅผ ํ•ด์•ผ ํ• ์ง€ ์•„์ง๋„ ๊ฐˆ ๊ธธ์ด ๋ฉ‰๋‹ˆ๋‹ค.
There's still quite a lot of work to do on how we optimize that optimization process.
IWSLT2017
null
null
null
ํ•˜์ง€๋งŒ ์ œ ์–ผ๊ตด์„ ๊ฐ€์ด๋“œ๋กœ ์“ฐ๋ฉด ๋” ์ผ๊ด€๋œ ์–ผ๊ตด์„ ๊ตฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
But you start to get something more like a coherent face, rendered using my own face as a guide.
IWSLT2017
null
null
null
๊ตณ์ด ๋นˆ ์บ”๋ฒ„์Šค๋กœ ์‹œ์ž‘ํ•˜์ง€ ์•Š์•„๋„ ๋ฉ๋‹ˆ๋‹ค. ํ˜น์€ ๋ฐฑ์ƒ‰ ์žก์Œ์œผ๋กœ์š”.
You don't have to start with a blank canvas or with white noise.
IWSLT2017
null
null
null
x๋ฅผ ๊ตฌํ•  ๋•Œ ์ด๋ฏธ ๊ทธ๋ ค์ง„ ๊ทธ๋ฆผ ์œ„์— x๋ฅผ ๊ตฌํ•ด๋„ ๋ฉ๋‹ˆ๋‹ค.
When you're solving for x, you can begin with an x, that is itself already some other image.
IWSLT2017
null
null
null
์ด๊ฒƒ์ด ๋ฐ”๋กœ ๊ทธ ์˜ˆ์ž…๋‹ˆ๋‹ค.
That's what this little demonstration is.
IWSLT2017
null
null
null
์ด ๋„คํŠธ์›Œํฌ๋Š” ์˜จ๊ฐ– ๋ฌผ์ฒด๋ฅผ ๊ตฌ๋ถ„ํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ธ์กฐ๋ฌผ์ด๋‚˜ ๋™๋ฌผ ๋“ฑ์„ ๋ง์ด์ฃ .
This is a network that is designed to categorize all sorts of different objects -- man-made structures, animals ...
IWSLT2017
null
null
null
์—ฌ๊ธฐ์„œ ์ €ํฌ๋Š” ๊ตฌ๋ฆ„ ์‚ฌ์ง„์„ ์ด์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๊ฐ€ ์ตœ์ ํ™”๋ฅผ ํ•˜๋ฉด ๊ธฐ๋ณธ์ ์œผ๋กœ ์ด ๋„คํŠธ์›Œํฌ๋Š” ๊ตฌ๋ฆ„์—์„œ ๋ฌด์—‡์ด ๋ณด์ด๋Š”์ง€ ๊ตฌ๋ถ„ํ•ฉ๋‹ˆ๋‹ค.
Here we're starting with just a picture of clouds, and as we optimize, basically, this network is figuring out what it sees in the clouds.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ด๊ฒƒ์„ ๋” ์ž์„ธํžˆ ๋ณด์‹œ๋ฉด ๊ตฌ๋ฆ„์—์„œ ๋” ๋‹ค์–‘ํ•œ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
And the more time you spend looking at this, the more things you also will see in the clouds.
IWSLT2017
null
null
null
์—ฌ๊ธฐ์„œ ์–ผ๊ตด์„ ์ธ์‹ํ•˜๋Š” ๋„คํŠธ์›Œํฌ๋กœ ํ™˜๊ฐ์„ ๋งŒ๋“ค๋ฉด ๊ฝค๋‚˜ ์ •์‹ ์—†๋Š” ๊ทธ๋ฆผ์ด ๋‚˜์˜ต๋‹ˆ๋‹ค.
You could also use the face network to hallucinate into this, and you get some pretty crazy stuff.
IWSLT2017
null
null
null
ํ˜น์€ ๋งˆ์ดํฌ๊ฐ€ ๋‹ค๋ฅธ ์‹œ๋„๋ฅผ ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฐ”๋กœ ๊ตฌ๋ฆ„ ๊ทธ๋ฆผ์„ ์ด์šฉํ•ด ํ™˜๊ฐ์„ ๋งŒ๋“ค๊ณ  ํ™•๋Œ€ํ•˜๊ณ  ํ™˜๊ฐ์„ ๋งŒ๋“ค๊ณ  ํ™•๋Œ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
Or, Mike has done some other experiments in which he takes that cloud image, hallucinates, zooms, hallucinates, zooms hallucinates, zooms.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ด๋ ‡๊ฒŒ ๋ฐฉํ™ฉํ•˜๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ด๋Š” ๋„คํŠธ์›Œํฌ๋‚˜ ์ž์œ  ์—ฐ์ƒ์˜ ์ผ์ข…์œผ๋กœ ๋„คํŠธ์›Œํฌ๊ฐ€ ์Šค์Šค๋กœ ๊ผฌ๋ฆฌ๋ฅผ ๋ฌผ๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.
And in this way, you can get a sort of fugue state of the network, I suppose, or a sort of free association, in which the network is eating its own tail.
IWSLT2017
null
null
null
๊ทธ๋ž˜์„œ ๋ชจ๋“  ์ด๋ฏธ์ง€์˜ ๊ธฐ๋ณธ์€ ์ด๋ ‡์Šต๋‹ˆ๋‹ค. "๋‹ค์Œ์—๋Š” ๋ฌด์—‡์ด ๋ณด์ด์ง€?
So every image is now the basis for, "What do I think I see next?
IWSLT2017
null
null
null
๋‹ค์Œ์—๋Š” ๋ฌด์—‡์ด ๋ณด์ด์ง€? ๋‹ค์Œ์—๋Š” ๋ฌด์—‡์ด ๋ณด์ด์ง€?"
What do I think I see next? What do I think I see next?"
IWSLT2017
null
null
null
์ด๊ฒƒ์„ ์ตœ์ดˆ๋กœ ๊ณต๊ฐœํ•œ ๊ณณ์€ ์‹œ์• ํ‹€์˜ "๊ณ ๋“ฑ ๊ต์œก"๊ทธ๋ฃน์˜ ๊ฐ•์—ฐ์—์„œ์˜€์Šต๋‹ˆ๋‹ค. ๋งˆ๋ฆฌํ™”๋‚˜๊ฐ€ ํ•ฉ๋ฒ•ํ™” ๋œ ์งํ›„์— ๋ง์ด์ฃ .
I showed this for the first time in public to a group at a lecture in Seattle called "Higher Education" -- this was right after marijuana was legalized.
IWSLT2017
null
null
null
๊ทธ๋ž˜์„œ ์ •๋ฆฌ๋ฅผ ์งง๊ฒŒ ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์— ์ œ์•ฝ์ด ์—†๋‹ค๋Š” ๊ฒƒ์„ ๋งํ•˜๋ฉด์„œ ๋ง์ด์ฃ .
So I'd like to finish up quickly by just noting that this technology is not constrained.
IWSLT2017
null
null
null
์ˆœ์ „ํžˆ ์‹œ๊ฐ์ž๋ฃŒ๋ฅผ ๋ณด์—ฌ๋“œ๋ฆฐ ์ด์œ ๋Š” ํฅ๋ฏธ๋ฅผ ์œ ๋ฐœํ•˜๊ธฐ ์œ„ํ•ด์„œ ์ž…๋‹ˆ๋‹ค.
I've shown you purely visual examples because they're really fun to look at.
IWSLT2017
null
null
null
์ด๊ฒƒ์€ ์ˆœ์ „ํžˆ ์‹œ๊ฐ ๊ธฐ์ˆ ๋งŒ์€ ์•„๋‹™๋‹ˆ๋‹ค.
It's not a purely visual technology.
IWSLT2017
null
null
null
์ €ํฌ์™€ ํ•จ๊ป˜ ์ผํ•˜๋Š” ์•„ํ‹ฐ์ŠคํŠธ ๋กœ์Šค ๊ตฟ์œˆ์€ ์‹คํ—˜์„ ํ–ˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์ง„์„ ์ฐ๋Š” ์‚ฌ์ง„๊ธฐ์™€ ๋“ฑ์— ๋งค๊ณ  ์žˆ๋Š” ์ปดํ“จํ„ฐ๋กœ ์‹ ๊ฒฝ ๋„คํŠธ์›Œํฌ๋ฅผ ์ด์šฉํ•ด ์‹œ๋ฅผ ์ผ์Šต๋‹ˆ๋‹ค. ์‚ฌ์ง„์— ์ฐํžŒ ๋‚ด์šฉ์„ ๋ณด๊ณ  ๋ง์ด์ฃ .
Our artist collaborator, Ross Goodwin, has done experiments involving a camera that takes a picture, and then a computer in his backpack writes a poem using neural networks, based on the contents of the image.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์‹œ์ธ ์‹ ๊ฒฝ ๋„คํŠธ์›Œํฌ๋Š” 20์„ธ๊ธฐ ์‹œ์˜ ์ง‘๋Œ€์„ฑ์œผ๋กœ ํ›ˆ๋ จ๋ฌ์Šต๋‹ˆ๋‹ค.
And that poetry neural network has been trained on a large corpus of 20th-century poetry.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ๊ฒฐ๊ณผ๋กœ ๋‚˜์˜จ ์‹œ๋Š” ๋ง์ด์ฃ  ์‚ฌ์‹ค ์ œ ์ƒ๊ฐ์—” ๋‚˜์˜์ง€ ์•Š์•„ ๋ณด์ž…๋‹ˆ๋‹ค.
And the poetry is, you know, I think, kind of not bad, actually.
IWSLT2017
null
null
null
๋งˆ์ง€๋ง‰์œผ๋กœ ์ €๋Š” ๋ฏธ์ผˆ๋ž€์ ค๋กœ์˜ ์ƒ๊ฐ์ด ์˜ณ์•˜๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ธ์‹๊ณผ ์ฐฝ์˜์„ฑ์€ ๋งค์šฐ ๋ฐ€์ ‘ํ•˜๊ฒŒ ์—ฐ๊ฒฐ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
In closing, I think that per Michelangelo, I think he was right; perception and creativity are very intimately connected.
IWSLT2017
null
null
null
์ง€๊ธˆ๊นŒ์ง€ ๋ณด์‹  ๊ฒƒ์€ ์‹ ๊ฒฝ ๋„คํŠธ์›Œํฌ ์ž…๋‹ˆ๋‹ค. ์ „์ ์œผ๋กœ ํ›ˆ๋ จ์ด ๋˜์–ด ๊ตฌ๋ถ„ํ•˜๊ฑฐ๋‚˜ ํ˜น์€ ๋‹ค๋ฅธ ๊ฒƒ๋“ค์„ ์ธ์‹ํ•˜๊ฑฐ๋‚˜ ๋ฐ˜๋Œ€๋กœ ์ ์šฉํ•˜์—ฌ ๋งŒ๋“ค์–ด ๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
What we've just seen are neural networks that are entirely trained to discriminate, or to recognize different things in the world, able to be run in reverse, to generate.
IWSLT2017
null
null
null
์ด๊ฒƒ์„ ๋ณด๊ณ  ๋Š๋‚€ ์  ์ค‘์— ํ•˜๋‚˜๋Š” ๋ฏธ์ผˆ๋ž€์ ค๋กœ๊ฐ€ ์ •๋ง๋กœ ๋ณธ ๊ฒƒ์€ ๋Œ๋ฉ์ด ์•ˆ์— ์žˆ๋Š” ์กฐ๊ฐ์ƒ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์–ด๋–ค ์ƒ๋ฌผ, ์ƒ๋ช… ์‹ฌ์ง€์–ด ์™ธ๊ณ„์ธ๋„ ์ธ์‹ํ–‰์œ„๋ฅผ ํ•  ์ˆ˜ ์žˆ์œผ๋ฉด ์ฐฝ์กฐํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ ์ž…๋‹ˆ๋‹ค. ๋‘ ๊ฒฝ์šฐ ๋ชจ๋‘ ๊ฐ™์€ ์กฐ์ž‘๊ณผ์ •์„ ์‚ฌ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์ด์ฃ .
One of the things that suggests to me is not only that Michelangelo really did see the sculpture in the blocks of stone, but that any creature, any being, any alien that is able to do perceptual acts of that sort is also able to create because it's exactly the same machinery that's used in both cases.
IWSLT2017
null
null
null
๋˜ํ•œ ์ €๋Š” ์ธ์‹๊ณผ ์ฐฝ์˜์„ฑ์€ ๊ฒฐ์ฝ” ์ธ๊ฐ„์— ๊ตญํ•œ๋˜์ง€ ์•Š๋Š”๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.
Also, I think that perception and creativity are by no means uniquely human.
IWSLT2017
null
null
null
์ €ํฌ๋Š” ๋˜‘๊ฐ™์€ ์ผ์„ ํ•  ์ˆ˜ ์žˆ๋Š” ์ปดํ“จํ„ฐ ๋ชจ๋ธ์„ ๋งŒ๋“ค์—ˆ๊ณ 
We start to have computer models that can do exactly these sorts of things.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ๊ทธ ๋‡Œ๊ฐ€ ์ปดํ“จํ„ฐ๋กœ ๋งŒ๋“ค์–ด ์กŒ๋‹ค๋Š” ๊ฒƒ์€ ๋†€๋ž„ ์ผ๋„ ์•„๋‹™๋‹ˆ๋‹ค.
And that ought to be unsurprising; the brain is computational.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ๋งˆ์ง€๋ง‰์œผ๋กœ ์ปดํ“จํ„ฐ๋Š” ์ง€๋Šฅ์  ๊ธฐ๊ณ„๋ฅผ ์„ค๊ณ„ํ•˜๋ฉด์„œ ์‹œ์ž‘๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
And finally, computing began as an exercise in designing intelligent machinery.
IWSLT2017
null
null
null
์ด๊ฒƒ์€ ์ด๋Ÿฐ ์ƒ๊ฐ์„ ๋”ฐ๋ผ ๋งŒ๋“ค์–ด์กŒ์Šต๋‹ˆ๋‹ค. ์–ด๋–ป๊ฒŒ ํ•˜๋ฉด ์šฐ๋ฆฌ๊ฐ€ ๊ธฐ๊ณ„๋ฅผ ๋˜‘๋˜‘ํ•˜๊ฒŒ ๋งŒ๋“ค์ง€ ๋ง์ด์ฃ .
It was very much modeled after the idea of how could we make machines intelligent.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ด์ œ ์„ ๊ตฌ์ž๋“ค๊ณผ ํ•œ ์•ฝ์† ์ค‘์— ์ผ๋ถ€๋ฅผ ์ด๋ค„๊ฐ€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠœ๋ง, ํฐ ๋…ธ์ด๋งŒ ๋งค์ปฌ๋กœํฌ ๊ทธ๋ฆฌ๊ณ  ํ”ผํŠธ์—๊ฒŒ ๋ง์ด์ฃ .
And we finally are starting to fulfill now some of the promises of those early pioneers, of Turing and von Neumann and McCulloch and Pitts.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์ €๋Š” ์ปดํ“จํ„ฐ๋Š” ํšŒ๊ณ„๋‚˜ ๊ฒŒ์ž„ ํ•  ๋•Œ๋งŒ ์“ฐ๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.
And I think that computing is not just about accounting or playing Candy Crush or something.
IWSLT2017
null
null
null
์‹œ์ž‘๋ถ€ํ„ฐ ์ธ๊ฐ„์„ ๋ณธ๋”ฐ ์ปดํ“จํ„ฐ๋ฅผ ๋งŒ๋“ค์—ˆ๊ณ 
From the beginning, we modeled them after our minds.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ๊ทธ ๊ณผ์ •์—์„œ ์ธ๊ฐ„์˜ ๋งˆ์Œ์„ ๋” ์ž˜ ์ดํ•ดํ•˜๊ณ  ๋” ๋„“ํžˆ๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
And they give us both the ability to understand our own minds better and to extend them.
IWSLT2017
null
null
null
๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
Thank you very much.
IWSLT2017
null
null
null
์ž์œ ๋กœ์šด ๋ฏธ๊ตญ์—์„œ ์šฐ๋ฆฌ๋Š” ๋“ฃ๊ณ  ๋ฐฐ์šฐ๋Š” ์—ฌํ–‰๋ฅผ ํ–ˆ์Šต๋‹ˆ๋‹ค.
At Free America, we've done a listening and learning tour.
IWSLT2017
null
null
null
๊ฒ€์ฐฐ๊ด€ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ๊ตญํšŒ์˜์›๋„ ๋งŒ๋‚ฌ๊ณ  ์—ฌ๋Ÿฌ ์ง€์—ญ์˜ ์ฃผ๋ฆฝ ๊ต๋„์†Œ์˜ ์ˆ˜๊ฐ์ž๋“ค๋„ ๋งŒ๋‚ฌ๊ณ 
We visited not only with prosecutors but with legislators, with inmates in our state and local prisons.
IWSLT2017
null
null
null
์ด๋ฏผ์ž ์ˆ˜์šฉ์†Œ์—๋„ ๋ฐฉ๋ฌธํ•˜๋ฉฐ
We've gone to immigration detention centers.
IWSLT2017
null
null
null
๋งŽ์€ ์‚ฌ๋žŒ๋“ค์„ ๋งŒ๋‚ฌ์Šต๋‹ˆ๋‹ค.
We've met a lot of people.
IWSLT2017
null
null
null
๊ทธ๋ฆฌ๊ณ  ์šฐ๋ฆฌ๋Š” ์•Œ๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์›๊ณผ ๋ณ€ํ™”๊ฐ€ ๊ต๋„์†Œ์—์„œ, ๊ฐ์˜ฅ์—์„œ ์ด๋ฏผ์ž ์ˆ˜์šฉ์†Œ์—์„œ ์ผ์–ด๋‚  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ์„์š”. ๋ณต์—ญ๊ธฐ๊ฐ„์ด ๋๋‚œ ํ›„์— ๋” ๋‚˜์€ ์‚ถ์„ ๋งŒ๋“ค๊ณ  ์‹ถ์–ดํ•˜๋Š” ๊ทธ ์‚ฌ๋žŒ๋“ค์—๊ฒŒ ํฌ๋ง์„ ์ฃผ๋ฉด์„œ ๋ง์ด์ฃ .
And we've seen that redemption and transformation can happen in our prisons, our jails and our immigration detention centers, giving hope to those who want to create a better life after serving their time.
IWSLT2017
null
null
null
์šฐ๋ฆฌ๊ฐ€ ์ˆ˜๊ฐ๊ธฐ๊ฐ„์˜ ๋๋„ ๋ฐ›์•„๋“ค์—ฌ์ค€๋‹ค๋ฉด ์–ด๋–จ๊นŒ์š”.
Imagine if we also considered the front end of this prison pipeline.
IWSLT2017
null
null
null
์šฐ๋ฆฌ๊ฐ€ ๋งŒ์•ฝ ์žฌํ™œ์„ ํ•ต์‹ฌ ๊ฐ€์น˜๋กœ ์—ฌ๊ธฐ๊ณ  ์‚ฌ๋ž‘๊ณผ ์—ฐ๋ฏผ์„ ํ•ต์‹ฌ ๊ฐ€์น˜๋กœ ์—ฌ๊ธด๋‹ค๋ฉด ์–ด๋–จ๊นŒ์š”?
What would it look like if we intervened, with rehabilitation as a core value -- with love and compassion as core values?
IWSLT2017
null
null
null
์šฐ๋ฆฌ ์‚ฌํšŒ๋Š” ๋” ์•ˆ์ „ํ•˜๊ณ  ๋” ๊ฑด๊ฐ•ํ•˜๊ณ  ์•„์ด๋“ค์ด ์‚ด๊ธฐ ์ข‹์€ ๊ณณ์ด ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
We would have a society that is safer, healthier and worthy of raising our children in.
IWSLT2017
null
null
null
์ œ์ž„์Šค ์บ๋น—์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.
I want to introduce you to James Cavitt.
IWSLT2017
null
null
null
์ œ์ž„์Šค๋Š” ์ƒŒ ํ€œํ‹ด ์ฃผ๋ฆฝ ๊ต๋„์†Œ์—์„œ 12๋…„๊ฐ„ ๋ณต์—ญ ๋์— 18๊ฐœ์›”์•ˆ์— ์„๋ฐฉ๋ฉ๋‹ˆ๋‹ค.
James served 12 years in the San Quentin State Prison and is being released in 18 months.
IWSLT2017
null
null
null
์ž, ์ œ์ž„์Šค๋Š” ์ €์™€ ์—ฌ๋Ÿฌ๋ถ„๋“ค์ฒ˜๋Ÿผ ๊ทธ๊ฐ€ ์ €์ง€๋ฅธ ์ผ๋ณด๋‹ค ๋” ๋งŽ์€ ์˜๋ฏธ๋ฅผ ์ง€๋…”์Šต๋‹ˆ๋‹ค.
Now James, like you and me, is more than the worst thing he's done.
IWSLT2017
null
null
null
๊ทธ๋Š” ์•„๋ฒ„์ง€์ด๋ฉฐ, ๋‚จํŽธ์ด์ž, ์•„๋“ค์ด๋ฉฐ, ์‹œ์ธ์ž…๋‹ˆ๋‹ค.
He is a father, a husband, a son, a poet.
IWSLT2017
null
null
null
๊ทธ๋Š” ๋ฒ”์ฃ„๋ฅผ ์ €์งˆ๋ €๊ณ  ๊ทธ ์ฃ—๊ฐ’์„ ์น˜๋ฅด๊ณ  ์žˆ์œผ๋ฉฐ ๋‹ค์‹œ ์‚ฌํšŒ๋กœ ๋Œ์•„์™”์„ ๋•Œ ์ƒ์‚ฐ์ ์ธ ์‚ถ์„ ๋˜์ฐพ๊ธฐ ์œ„ํ•ด ๋…ธ๋ ฅํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
He committed a crime; he's paying his debt, and working hard to build the skills to make the transition back to a productive life when he enters the civilian population again.
IWSLT2017
null
null
null
์ž, ์ œ์ž„์Šค๋Š” ์ˆ˜๊ฐ ์ค‘์ธ ์ˆ˜๋ฐฑ๋งŒ ๋ช…์˜ ์‚ฌ๋žŒ๋“ค์ฒ˜๋Ÿผ ์šฐ๋ฆฌ์˜ ์‹คํŒจ๊ฐ€ ์šฐ๋ฆฌ๋ฅผ ์ •์˜ํ•˜์ง€ ์•Š๋Š”๋‹ค๊ณ  ๋ฏฟ๋Š”๋‹ค๋ฉด ์–ด๋–ค ์ผ์ด ์ผ์–ด๋‚ ์ง€ ๋ณด์—ฌ์ฃผ๋Š” ์˜ˆ์‹œ์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ ๋ชจ๋‘๊ฐ€ ๊ตฌ์›๋ฐ›์„ ๊ฐ€์น˜๊ฐ€ ์žˆ๊ณ  ๋Œ€๊ทœ๋ชจ์˜ ํˆฌ์˜ฅ์„ ๊ฒช์€ ์‚ฌ๋žŒ๋“ค์„ ๋„์™€์ค€๋‹ค๋ฉด ๋ชจ๋‘๊ฐ€ ์น˜์œ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Now James, like millions of people behind bars, is an example of what happens if we believe that our failings don't define who we are, that we are all worthy of redemption and if we support those impacted by mass incarceration, we can all heal together.
IWSLT2017
null
null
null
์ œ์ž„์Šค๋ฅผ ์ง€๊ธˆ ์†Œ๊ฐœํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค. ๊ทธ๊ฐ€ ๊ฒช์€ ๊ตฌ์›์˜ ์—ฌ์ •์„ ๋งํ•ด์ค„ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ œ์ž„์Šค ์บ๋น—: ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค, ์กด.
I'd like to introduce you to James right now, and he's going to share his journey of redemption James Cavitt: Thanks, John.
IWSLT2017
null
null
null
TED, ์ƒŒ ํ€œํ‹ด์— ์˜ค์‹  ๊ฑธ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค.
TED, welcome to San Quentin.
IWSLT2017
null
null
null
๊ต๋„์†Œ์˜ ๋‹ด๋ฒผ๋ฝ ๋’ค์—๋Š” ๋งŽ์€ ์žฌ๋Šฅ๋“ค์ด ์žˆ์Šต๋‹ˆ๋‹ค.
The talent is abundant behind prison walls.
IWSLT2017
null
null
null
๋ฏธ๋ž˜์˜ ์†Œํ”„ํŠธ์›จ์–ด ๊ธฐ์ˆ ์ž ์‚ฌ์—…๊ฐ€ ๊ณต์˜ˆ๊ฐ€ ์Œ์•…๊ฐ€ ๊ทธ๋ฆฌ๊ณ  ์˜ˆ์ˆ ๊ฐ€๊นŒ์ง€.
Future software engineers, entrepreneurs, craftsmen, musicians and artists.
IWSLT2017
null
null
null
์ด ์ž‘ํ’ˆ์€ ์ด ๊ณณ ์•ˆ์—์„œ ์ˆ˜๊ฐ๊ธฐ๊ฐ„์ด ๋๋‚œ ํ›„ ๋” ๋‚˜์€ ์‚ถ๊ณผ ๋ฏธ๋ž˜๋ฅผ ์ด๋ฃจ๊ธฐ ์œ„ํ•ด ๋…ธ๋ ฅํ•˜๊ณ  ์žˆ๋Š” ์‚ฌ๋žŒ๋“ค์—๊ฒŒ ์˜๊ฐ์„ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค. ์ด ์ž‘ํ’ˆ์˜ ์ œ๋ชฉ์€ "๋‚ด๊ฐ€ ์‚ฌ๋Š” ๊ณณ" ์ž…๋‹ˆ๋‹ค.
This piece is inspired by all of the hard work that men and women are doing on the inside to create better lives and futures for themselves This piece is entitled, "Where I Live."
IWSLT2017
null
null
null
๋‚˜๋Š” ๋Œ€๋ถ€๋ถ„์˜ ์‚ฌ๋žŒ๋“ค์ด ๊ฐ€๊ธฐ ๋„ˆ๋ฌด ๋‘๋ ค์›Œํ•˜๋Š” ์„ธ์ƒ์— ์‚ฐ๋‹ค.
I live in a world where most people are too afraid to go.
IWSLT2017
null
null
null
๋†’์€ ์ฝ˜ํฌ๋ฆฌํŠธ ๋ฒฝ๊ณผ ๊ฐ•์ฒ ๋กœ ๋œ ์ฒ ์ฐฝ์œผ๋กœ ๋‘˜๋Ÿฌ์‹ธ์ธ ๋” ๋ฐ์€ ๋ฏธ๋ž˜์— ๋Œ€ํ•œ ํฌ๋ง์„ ์ž˜๋ผ๋ฒ„๋ฆฌ๋Š” ์ฒ ์กฐ๋ง์ด ์žˆ๋Š” ๊ณณ.
Surrounded by tall, concrete walls, steel bars, where razor wire have a way of cutting away at the hopes for a brighter tomorrow.
IWSLT2017
null
null
null
๋‚˜๋Š” ์‚ฌ๋žŒ์„ ์ฃฝ์ด๋Š” ๊ฒƒ์ด ๋‚˜์˜๋‹ค๋Š” ๊ฒƒ์„ ๊ฐ€๋ฅด์น˜๊ธฐ ์œ„ํ•ด์„œ ์‚ฌ๋žŒ์„ ์ฃฝ์ธ ์‚ฌ๋žŒ์„ ์ฃฝ์ด๋Š” ์„ธ์ƒ์— ์‚ฐ๋‹ค.
I live in a world that kill people who kill people in order to teach people that killing people is wrong.
IWSLT2017
null
null
null
์ƒ์ƒํ•ด๋ณด๋ผ.
Imagine that.
IWSLT2017
null
null
null