How AI is making it easier to diagnose disease | Pratik Shah

84,509 views ใƒป 2018-08-21

TED


์•„๋ž˜ ์˜๋ฌธ์ž๋ง‰์„ ๋”๋ธ”ํด๋ฆญํ•˜์‹œ๋ฉด ์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค.

๋ฒˆ์—ญ: ํƒœ๊ฐ• ๊น€ ๊ฒ€ํ† : TJ Kim
00:13
Computer algorithms today are performing incredible tasks
0
13280
3856
์˜ค๋Š˜๋‚  ์ปดํ“จํ„ฐ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ธฐ๋Šฅ์€ ์ •๋ง ๋†€๋ž์Šต๋‹ˆ๋‹ค.
00:17
with high accuracies, at a massive scale, using human-like intelligence.
1
17160
4736
์ธ๊ฐ„์˜ ์ง€๋Šฅ๊ณผ ๋น„์Šทํ•˜์ง€๋งŒ, ์•„์ฃผ ์—„์ฒญ๋‚œ ์–‘์„ ์ •ํ™•ํ•˜๊ฒŒ ์ฒ˜๋ฆฌํ•˜์ฃ .
00:21
And this intelligence of computers is often referred to as AI
2
21920
3936
์ด๋Ÿฐ ์ปดํ“จํ„ฐ์˜ ์ง€๋Šฅ์„ ํ”ํžˆ AI ๋˜๋Š”
00:25
or artificial intelligence.
3
25880
1856
์ธ๊ณต์ง€๋Šฅ์ด๋ผ๊ณ  ํ•˜์ฃ .
00:27
AI is poised to make an incredible impact on our lives in the future.
4
27760
4200
์ธ๊ณต์ง€๋Šฅ์€ ์ด๋ฏธ ์šฐ๋ฆฌ์˜ ๋ฏธ๋ž˜์— ์—„์ฒญ๋‚œ ์˜ํ–ฅ์„ ์ค„ ํž˜์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
00:32
Today, however, we still face massive challenges
5
32880
3936
ํ•˜์ง€๋งŒ, ์˜ค๋Š˜๋‚  ์šฐ๋ฆฌ๋Š” ์—ฌ์ „ํžˆ ์ƒ๋ช…์„ ์œ„ํ˜‘ํ•˜๋Š” ์งˆ๋ณ‘๋“ค์„
00:36
in detecting and diagnosing several life-threatening illnesses,
6
36840
3496
๋ฐœ๊ฒฌํ•˜๊ณ  ์ง„๋‹จํ•˜๋Š” ๋ฐ ๋งŽ์€ ์–ด๋ ค์›€์„ ๊ฒช๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
00:40
such as infectious diseases and cancer.
7
40360
2360
์•”์ด๋‚˜ ๊ฐ์—ผ์„ฑ ์งˆ๋ณ‘ ๊ฐ™์€ ๊ฒƒ๋“ค์ด์ฃ .
00:44
Thousands of patients every year
8
44000
2296
๋งค๋…„ ์ˆ˜ ์ฒœ๋ช…์˜ ํ™˜์ž๋“ค์ด
00:46
lose their lives due to liver and oral cancer.
9
46320
2800
๊ฐ„์•”๊ณผ ๊ตฌ๊ฐ•์•”์œผ๋กœ ์ƒ๋ช…์„ ์žƒ์Šต๋‹ˆ๋‹ค.
00:49
Our best way to help these patients
10
49880
2696
์ด๋Ÿฐ ํ™˜์ž๋“ค์„ ๋„์šธ ์ˆ˜ ์žˆ๋Š”
00:52
is to perform early detection and diagnoses of these diseases.
11
52600
4320
์ตœ์„ ์˜ ๋ฐฉ๋ฒ•์€ ๋ฐ”๋กœ ์งˆ๋ณ‘์˜ ์กฐ๊ธฐ ๋ฐœ๊ฒฌ๊ณผ ์ง„๋‹จ์ด์ฃ .
00:57
So how do we detect these diseases today, and can artificial intelligence help?
12
57880
4160
์˜ค๋Š˜๋‚  ์–ด๋–ค ์‹์œผ๋กœ ์งˆ๋ณ‘์ด ๋ฐœ๊ฒฌ๋˜๊ณ , ์ธ๊ณต์ง€๋Šฅ์€ ์–ด๋–ค ๋„์›€์„ ์ฃผ๋Š”์ง€ ์•Œ์•„๋ณผ๊นŒ์š”?
01:03
In patients who, unfortunately, are suspected of these diseases,
13
63920
3656
์•ˆํƒ€๊น๊ฒŒ๋„, ์ด๋Ÿฐ ์งˆ๋ณ‘์— ๊ฐ์—ผ๋œ ๊ฒƒ์œผ๋กœ ์˜์‹ฌ๋˜๋Š” ํ™˜์ž๋“ค์—๊ฒŒ,
01:07
an expert physician first orders
14
67600
2656
์ „๋ฌธ์˜๋“ค์€ ์šฐ์„ 
01:10
very expensive medical imaging technologies
15
70280
2616
๊ฐ’๋น„์‹ผ ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.
01:12
such as fluorescent imaging, CTs, MRIs, to be performed.
16
72920
4096
ํ˜•๊ด‘์˜์ƒ๋ฒ•, CT, MRI ๋“ฑ์ด์ฃ .
01:17
Once those images are collected,
17
77040
2296
์ˆ˜์ง‘๋œ ์˜์ƒ๋“ค์„
01:19
another expert physician then diagnoses those images and talks to the patient.
18
79360
4520
๋˜ ๋‹ค๋ฅธ ์ „๋ฌธ์˜๊ฐ€ ํŒ๋…ํ•˜๊ณ  ํ™˜์ž์—๊ฒŒ ์†Œ๊ฒฌ์„ ๋ณด๋ƒ…๋‹ˆ๋‹ค.
01:24
As you can see, this is a very resource-intensive process,
19
84520
3456
๋ณด๋‹ค์‹œํ”ผ, ์•„์ฃผ ์†Œ๋ชจ๊ฐ€ ์‹ฌํ•œ ๊ณผ์ •์ž…๋‹ˆ๋‹ค.
01:28
requiring both expert physicians, expensive medical imaging technologies,
20
88000
4416
์ „๋ฌธ์˜ ๋‘˜์—, ๊ฐ’๋น„์‹ผ ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ ๊นŒ์ง€ ๊ฐ–์ถฐ์•ผ ํ•˜๋‹ˆ๊นŒ
01:32
and is not considered practical for the developing world.
21
92440
3096
๊ฐœ๋ฐœ ๋„์ƒ๊ตญ๋“ค์—๊ฒŒ ์‹ค์šฉ์ ์ธ ๋ฐฉ๋ฒ•์ด ์•„๋‹ˆ์ฃ .
01:35
And in fact, in many industrialized nations, as well.
22
95560
3360
์‚ฌ์‹ค, ์—ฌ๋Ÿฌ ์„ ์ง„๊ตญ๋“ค์˜ ์‚ฌ์ •๋„ ๋งˆ์ฐฌ๊ฐ€์ง€์ž…๋‹ˆ๋‹ค.
01:39
So, can we solve this problem using artificial intelligence?
23
99760
2880
๊ทธ๋Ÿผ, ์ธ๊ณต์ง€๋Šฅ์„ ์ด์šฉํ•ด ์ด ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์„๊นŒ์š”?
01:43
Today, if I were to use traditional artificial intelligence architectures
24
103840
4056
์˜ค๋Š˜๋‚ , ์ œ๊ฐ€ ์ด ๋ฌธ์ œ์˜ ํ•ด๊ฒฐ์„ ์œ„ํ•ด
01:47
to solve this problem,
25
107920
1216
๊ธฐ์กด์˜ ์ธ๊ณต์ง€๋Šฅ๊ตฌ์กฐ๋ฅผ ์ด์šฉํ•œ๋‹ค๋ฉด,
01:49
I would require 10,000 --
26
109160
1456
๋งŒ์žฅ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
01:50
I repeat, on an order of 10,000 of these very expensive medical images
27
110640
4016
๋‹ค์‹œ ๋งํ•ด, ๋งŒ์žฅ์˜ ๊ฐ’๋น„์‹ผ ์˜๋ฃŒ์˜์ƒ์„
01:54
first to be generated.
28
114680
1376
๋จผ์ € ์ฐ์–ด์•ผ๋งŒ ํ•ฉ๋‹ˆ๋‹ค.
01:56
After that, I would then go to an expert physician,
29
116080
2896
๊ทธ ๋‹ค์Œ์—, ๊ทธ ์˜์ƒ๋“ค์„ ๋ถ„์„ํ•ด ์ค„
01:59
who would then analyze those images for me.
30
119000
2496
์ „๋ฌธ์˜๋ฅผ ์ฐพ์•„๊ฐ€์•ผ ๊ฒ ์ฃ .
02:01
And using those two pieces of information,
31
121520
2096
๊ทธ๋ฆฌ๊ณ , ๊ทธ ๋‘ ์ข…๋ฅ˜์˜ ์ •๋ณด๋ฅผ ๊ฐ€์ง€๊ณ 
02:03
I can train a standard deep neural network or a deep learning network
32
123640
3656
ํ‘œ์ค€ํ™”๋œ ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง ๋˜๋Š” ์‹ฌ์ธต ํ•™์Šต๋ง์„ ์ด์šฉํ•ด
02:07
to provide patient's diagnosis.
33
127320
2136
ํ™˜์ž๋“ค์—๊ฒŒ ์ง„๋‹จ์„ ๋‚ด๋ฆฝ๋‹ˆ๋‹ค.
02:09
Similar to the first approach,
34
129480
1736
์ฒซ ๋ฒˆ์งธ ๋ฐฉ์‹๊ณผ ์œ ์‚ฌํ•˜๊ฒŒ,
02:11
traditional artificial intelligence approaches
35
131240
2143
๊ธฐ์กด์˜ ์ธ๊ณต์ง€๋Šฅ์€ ๊ฐ™์€ ๋ฌธ์ œ๋กœ
02:13
suffer from the same problem.
36
133407
1449
์–ด๋ ค์›€์„ ๊ฒช๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
02:14
Large amounts of data, expert physicians and expert medical imaging technologies.
37
134880
4560
๋ฐฉ๋Œ€ํ•œ ์ž๋ฃŒ, ์ „๋ฌธ์˜, ํŠน์ˆ˜ ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ  ๋“ฑ์ด ํ•„์š”ํ•˜๊ธฐ ๋•Œ๋ฌธ์ด์ฃ .
02:20
So, can we invent more scalable, effective
38
140320
4296
๊ทธ๋ ‡๋‹ค๋ฉด, ๋” ํ™•์žฅ์„ฑ์ด ๋†’๊ณ ,
02:24
and more valuable artificial intelligence architectures
39
144640
3296
ํšจ์œจ์ ์ธ ์ธ๊ณต์ง€๋Šฅ ๊ตฌ์กฐ๋ฅผ ๋งŒ๋“ ๋‹ค๋ฉด,
02:27
to solve these very important problems facing us today?
40
147960
3056
์˜ค๋Š˜๋‚  ์šฐ๋ฆฌ์—๊ฒŒ ๋‹น๋ฉดํ•œ ์ค‘๋Œ€ํ•œ ๊ณผ์ œ๋“ค์„ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์„๊นŒ์š”?
02:31
And this is exactly what my group at MIT Media Lab does.
41
151040
3296
์ด๊ฒƒ์ด ๋ฐ”๋กœ MIT ๋ฏธ๋””์–ด ์—ฐ๊ตฌ์†Œ์—์„œ ์ €ํฌ ํŒ€์ด ํ•˜๊ณ  ์žˆ๋Š” ์ผ์ž…๋‹ˆ๋‹ค.
02:34
We have invented a variety of unorthodox AI architectures
42
154360
3856
๋‹ค์–‘ํ•œ ๋Œ€์ฒด ์ธ๊ณต์ง€๋Šฅ ๊ตฌ์กฐ๋ฅผ ๋งŒ๋“ค๊ณ  ์žˆ์ฃ .
02:38
to solve some of the most important challenges facing us today
43
158240
3176
์˜๋ฃŒ์˜์ƒ๊ณผ ์ž„์ƒ์‹œํ—˜ ๋ถ„์•ผ์—์„œ์˜
02:41
in medical imaging and clinical trials.
44
161440
2200
์ค‘๋Œ€ํ•œ ๊ณผ์ œ๋“ค์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด์„œ์ฃ .
02:44
In the example I shared with you today, we had two goals.
45
164480
3056
์˜ค๋Š˜ ์˜ˆ๋กœ ๋“  ๊ฒƒ์—์„œ, ์ €ํฌ๋Š” ๋‘ ๊ฐ€์ง€ ๋ชฉํ‘œ๋ฅผ ์ •ํ–ˆ์Šต๋‹ˆ๋‹ค.
02:47
Our first goal was to reduce the number of images
46
167560
2976
์ฒซ์งธ, ์ธ๊ณต์ง€๋Šฅ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ํ•™์Šต์— ํ•„์š”ํ•œ
02:50
required to train artificial intelligence algorithms.
47
170560
3256
์˜์ƒ์˜ ์ˆ˜๋ฅผ ์ค„์ด๋Š” ๊ฒƒ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
02:53
Our second goal -- we were more ambitious,
48
173840
2096
๋‘˜์งธ, ์ €ํฌ๊ฐ€ ์ข€ ๋” ์š•์‹ฌ์„ ๋‚ธ ๊ฒƒ์ธ๋ฐ
02:55
we wanted to reduce the use of expensive medical imaging technologies
49
175960
3736
ํ™˜์ž๋“ค์„ ์„ ๋ณ„ํ•˜๋Š” ๊ฐ’๋น„์‹ผ ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ ์˜ ์‚ฌ์šฉ์„
02:59
to screen patients.
50
179720
1216
์ค„์ด๋Š” ๊ฒƒ์ด์—ˆ์ฃ .
03:00
So how did we do it?
51
180960
1200
๊ณผ์—ฐ ์–ด๋–ป๊ฒŒ ๋์„๊นŒ์š”?
03:02
For our first goal,
52
182920
1216
์ฒซ ๋ฒˆ์งธ ๋ชฉํ‘œ๋ฅผ ์œ„ํ•ด,
03:04
instead of starting with tens and thousands
53
184160
2056
๊ธฐ์กด์˜ ์ธ๊ณต์ง€๋Šฅ๊ณผ ๊ฐ™์ด
03:06
of these very expensive medical images, like traditional AI,
54
186240
3016
์•„์ฃผ ๋งŽ์€ ๋น„์šฉ์„ ๋“ค์—ฌ ์ดฌ์˜ํ•˜๋Š” ์˜์ƒ ์ˆ˜ ์ฒœ์žฅ์„ ๋Œ€์‹ ํ•ด,
03:09
we started with a single medical image.
55
189280
2056
๋‹จ ํ•œ ์žฅ์œผ๋กœ ์‹œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค.
03:11
From this image, my team and I figured out a very clever way
56
191360
3776
๋™๋ฃŒ๋“ค๊ณผ ํ•จ๊ป˜ ๊ทธ ์˜์ƒ์—์„œ ์ˆ˜ ์–ต๊ฐœ์˜ ์ •๋ณด ํŒจํ‚ท์„
03:15
to extract billions of information packets.
57
195160
2736
์ถ”์ถœํ•  ์ˆ˜ ์žˆ๋Š” ์•„์ฃผ ๊ธฐ๋ฐœํ•œ ๋ฐฉ๋ฒ•์„ ์ฐพ์•„๋ƒˆ์ฃ .
03:17
These information packets included colors, pixels, geometry
58
197920
3696
์ด ์ •๋ณด ํŒจํ‚ท๋“ค์€ ์˜์ƒ ์† ์งˆ๋ณ‘์˜ ์ƒ‰์ƒ, ํ™”์†Œ, ๊ธฐํ•˜ํ•™์  ๊ตฌ์กฐ,
03:21
and rendering of the disease on the medical image.
59
201640
2536
๋ Œ๋”๋ง ๋“ฑ์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
03:24
In a sense, we converted one image into billions of training data points,
60
204200
4336
์ฆ‰, ํ•˜๋‚˜์˜ ์˜์ƒ์„ ์ˆ˜ ์‹ญ์–ต๊ฐœ์˜ ํ•™์Šต์šฉ ์ž๋ฃŒ์ ์œผ๋กœ ๋ณ€ํ™˜ํ•ด,
03:28
massively reducing the amount of data needed for training.
61
208560
3536
ํ•™์Šต์— ํ•„์š”ํ•œ ์ž๋ฃŒ์˜ ์–‘์„ ํ˜„์ €ํžˆ ์ค„์ด๋Š” ๊ฒƒ์ด์ฃ .
03:32
For our second goal,
62
212120
1216
๋‘ ๋ฒˆ์งธ ๋ชฉํ‘œ๋ฅผ ์œ„ํ•ด์„œ๋Š”
03:33
to reduce the use of expensive medical imaging technologies to screen patients,
63
213360
3856
ํ™˜์ž์˜ ์„ ๋ณ„์— ์“ฐ์ด๋Š” ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ ์˜ ์‚ฌ์šฉ์„ ์ค„์ด๊ธฐ ์œ„ํ•ด,
03:37
we started with a standard, white light photograph,
64
217240
2856
์ผ๋ฐ˜์ ์ธ ๋ฐฑ์ƒ‰๊ด‘ ์‚ฌ์ง„์„ ์‚ฌ์šฉํ•˜๊ธฐ ์‹œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค.
03:40
acquired either from a DSLR camera or a mobile phone, for the patient.
65
220120
4336
DSLR ์‚ฌ์ง„๊ธฐ๋‚˜ ํœด๋Œ€์ „ํ™”์—์„œ๋„ ๊ฐ€๋Šฅํ•œ ๊ฒƒ์ด์ฃ .
03:44
Then remember those billions of information packets?
66
224480
2456
์ˆ˜ ์‹ญ์–ต๊ฐœ์˜ ์ •๋ณด ํŒจํ‚ท ๊ธฐ์–ตํ•˜์‹œ์ฃ ?
03:46
We overlaid those from the medical image onto this image,
67
226960
3536
์˜๋ฃŒ์˜์ƒ์—์„œ ๋นผ๋‚ธ ์ •๋ณด๋“ค์„ ๋ฐ”๋กœ ์ด ์˜์ƒ ์œ„์— ์ž…ํ˜€
03:50
creating something that we call a composite image.
68
230520
2520
ํ•ฉ์„ฑํ•˜๋Š” ๊ฑฐ์ฃ .
03:53
Much to our surprise, we only required 50 --
69
233480
3296
๋†€๋ž๊ฒŒ๋„, ์ €ํฌ๋Š” ๋‹จ 50์žฅ,
03:56
I repeat, only 50 --
70
236800
1336
ํ•œ๋ฒˆ ๋” ๋งํ•˜์ฃ , ๋‹จ 50์žฅ์ž…๋‹ˆ๋‹ค.
03:58
of these composite images to train our algorithms to high efficiencies.
71
238160
3840
์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ํ•™์Šต์ด ๋‹จ 50์žฅ์˜ ์˜์ƒ์œผ๋กœ ๊ฐ€๋Šฅํ–ˆ๋˜ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:02
To summarize our approach,
72
242680
1336
๊ทธ ๋ฐฉ์‹์„ ์š”์•ฝํ•ด ๋ณด๋ฉด,
04:04
instead of using 10,000 very expensive medical images,
73
244040
3176
๋งŒ์žฅ์ด๋‚˜ ๋˜๋Š” ์•„์ฃผ ๊ฐ’๋น„์‹ผ ์˜๋ฃŒ์˜์ƒ์„ ๋Œ€์‹ ํ•ด,
04:07
we can now train the AI algorithms in an unorthodox way,
74
247240
3016
์ด์ œ๋Š” ์ƒ‰๋‹ค๋ฅธ ๋ฐฉ์‹์œผ๋กœ ์ธ๊ณต์ง€๋Šฅ์˜ ํ•™์Šต์ด ๊ฐ€๋Šฅํ•˜๊ฒŒ ๋œ ๊ฒƒ์ด์ฃ .
04:10
using only 50 of these high-resolution, but standard photographs,
75
250280
4256
์ผ๋ฐ˜์ ์ธ DSLR ์‚ฌ์ง„๊ธฐ๋‚˜ ํœด๋Œ€์ „ํ™”๋ฅผ ์ด์šฉํ•ด ์–ป์„ ์ˆ˜ ์žˆ๋Š”
04:14
acquired from DSLR cameras and mobile phones,
76
254560
2496
๋‹จ 50์žฅ์˜ ๊ณ ํ™”์งˆ ์˜์ƒ์œผ๋กœ
04:17
and provide diagnosis.
77
257080
1536
์ง„๋‹จ๊นŒ์ง€ ๊ฐ€๋Šฅํ•ด ์ง„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
04:18
More importantly,
78
258640
1216
๋” ์ค‘์š”ํ•˜๊ฒŒ
04:19
our algorithms can accept, in the future and even right now,
79
259880
3136
์ €ํฌ ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์•ž์œผ๋กœ, ์‹ฌ์ง€์–ด ์ง€๊ธˆ ๋‹น์žฅ์ด๋ผ๋„
04:23
some very simple, white light photographs from the patient,
80
263040
2816
ํ™˜์ž๋“ค์ด ์ œ๊ณตํ•˜๋Š” ๊ฐ„๋‹จํ•œ ๋ฐฑ์ƒ‰๊ด‘ ์‚ฌ์ง„์„ ์ด์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
04:25
instead of expensive medical imaging technologies.
81
265880
2440
๊ณ ๊ฐ€์˜ ์˜๋ฃŒ์˜์ƒ๊ธฐ์ˆ ์„ ๋Œ€์‹ ํ•ด์„œ ๋ง์ด์ฃ .
04:29
I believe that we are poised to enter an era
82
269120
3096
์ €๋Š” ์ธ๊ณต์ง€๋Šฅ์ด ์šฐ๋ฆฌ์˜ ๋ฏธ๋ž˜์—
04:32
where artificial intelligence
83
272240
1936
์—„์ฒญ๋‚œ ์˜ํ–ฅ์„ ์ค„
04:34
is going to make an incredible impact on our future.
84
274200
2536
์‹œ๋Œ€์˜ ๋ฌธํ„ฑ์— ์ด๋ฏธ ์™€ ์žˆ๋‹ค๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค.
04:36
And I think that thinking about traditional AI,
85
276760
2456
๊ธฐ์กด์˜ ์ธ๊ณต์ง€๋Šฅ์€
04:39
which is data-rich but application-poor,
86
279240
2776
์ž๋ฃŒ๋Š” ๋ฐฉ๋Œ€ํ•˜์ง€๋งŒ ํ™œ์šฉ๋„๊ฐ€ ๋‚ฎ๊ธฐ ๋•Œ๋ฌธ์—,
04:42
we should also continue thinking
87
282040
1536
์†Œ๋Ÿ‰์˜ ์ž๋ฃŒ๋งŒ์œผ๋กœ๋„
04:43
about unorthodox artificial intelligence architectures,
88
283600
3016
์šฐ๋ฆฌ์—๊ฒŒ ๋‹น๋ฉดํ•œ ์ค‘๋Œ€ํ•œ ๊ณผ์ œ๋“ค์„ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ๋Š”
04:46
which can accept small amounts of data
89
286640
1936
๋Œ€์ฒด ์ธ๊ณต์ง€๋Šฅ ๊ตฌ์กฐ๋ฅผ
04:48
and solve some of the most important problems facing us today,
90
288600
2936
๊ณ„์† ์—ฐ๊ตฌํ•ด์•ผ ํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.
04:51
especially in health care.
91
291560
1256
ํŠนํžˆ ์˜๋ฃŒ์™€ ๊ด€๋ จํ•ด์„œ ๋ง์ด์ฃ .
04:52
Thank you very much.
92
292840
1216
๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
04:54
(Applause)
93
294080
3840
(๋ฐ•์ˆ˜)
์ด ์›น์‚ฌ์ดํŠธ ์ •๋ณด

์ด ์‚ฌ์ดํŠธ๋Š” ์˜์–ด ํ•™์Šต์— ์œ ์šฉํ•œ YouTube ๋™์˜์ƒ์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ „ ์„ธ๊ณ„ ์ตœ๊ณ ์˜ ์„ ์ƒ๋‹˜๋“ค์ด ๊ฐ€๋ฅด์น˜๋Š” ์˜์–ด ์ˆ˜์—…์„ ๋ณด๊ฒŒ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ฐ ๋™์˜์ƒ ํŽ˜์ด์ง€์— ํ‘œ์‹œ๋˜๋Š” ์˜์–ด ์ž๋ง‰์„ ๋”๋ธ” ํด๋ฆญํ•˜๋ฉด ๊ทธ๊ณณ์—์„œ ๋™์˜์ƒ์ด ์žฌ์ƒ๋ฉ๋‹ˆ๋‹ค. ๋น„๋””์˜ค ์žฌ์ƒ์— ๋งž์ถฐ ์ž๋ง‰์ด ์Šคํฌ๋กค๋ฉ๋‹ˆ๋‹ค. ์˜๊ฒฌ์ด๋‚˜ ์š”์ฒญ์ด ์žˆ๋Š” ๊ฒฝ์šฐ ์ด ๋ฌธ์˜ ์–‘์‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์˜ํ•˜์‹ญ์‹œ์˜ค.

https://forms.gle/WvT1wiN1qDtmnspy7