1
00:00:00,000 --> 00:00:04,471
We are
now in a new era of computing technology.

2
00:00:04,471 --> 00:00:09,142
We see artificial intelligence
in nearly all aspects of our lives,

3
00:00:09,142 --> 00:00:11,911
from helping doctors
and diagnosing diseases,

4
00:00:11,911 --> 00:00:15,348
to creating recipes
based on what's in your pantry.

5
00:00:15,348 --> 00:00:18,752
AI is quickly changing the way we live.

6
00:00:18,752 --> 00:00:23,723
Artificial intelligence is also changing
the way astronomers analyze data.

7
00:00:23,723 --> 00:00:27,961
Such advances have been used to increase
the discoveries made by NASA's

8
00:00:27,961 --> 00:00:31,297
Hubble Space Telescope
and in turn, Hubble's

9
00:00:31,297 --> 00:00:36,002
data has been used to advance
and test artificial intelligence systems.

10
00:00:36,002 --> 00:00:40,240
Such systems were critical
in discovering asteroids in Hubble data.

11
00:00:40,240 --> 00:00:44,377
To date, astronomers have found over
a million asteroids in the main

12
00:00:44,377 --> 00:00:48,815
asteroid belt, an area
between the orbits of Mars and Jupiter.

13
00:00:48,815 --> 00:00:54,087
But many smaller asteroids may not have
been discovered because they are so faint.

14
00:00:54,087 --> 00:00:57,757
They are about 1/40,000,000
the brightness of the faintest

15
00:00:57,757 --> 00:01:00,560
star a human eye can see.

16
00:01:00,560 --> 00:01:03,563
This is where Hubble and AI come in.

17
00:01:03,563 --> 00:01:07,634
Hubble's archive contains over 1 million
observations.

18
00:01:07,634 --> 00:01:11,471
Many of these smaller asteroids
may be found in the images

19
00:01:11,471 --> 00:01:13,573
because they are moving across the skies

20
00:01:13,573 --> 00:01:16,776
as Hubble's view
holds steady on an astronomical object.

21
00:01:16,776 --> 00:01:20,480
They appear as curved streaks of light
in Hubble's cameras,

22
00:01:20,480 --> 00:01:23,383
essentially photobombing the image.

23
00:01:23,383 --> 00:01:26,619
Astronomers applied machine
learning algorithms,

24
00:01:26,619 --> 00:01:32,792
a form of artificial intelligence, to hunt
for these trails in over 37,000 images.

25
00:01:32,792 --> 00:01:36,062
That exercise was conducted
with the help of 11,000

26
00:01:36,062 --> 00:01:41,101
citizen scientists,
who discovered 1701 asteroids,

27
00:01:41,101 --> 00:01:46,573
1031 of these were small bodied asteroids
not formerly known to exist.

28
00:01:46,573 --> 00:01:48,475
The results help provide insight

29
00:01:48,475 --> 00:01:52,479
into both the origin
and evolution of the asteroid belt.

30
00:01:52,479 --> 00:01:57,517
AI can also revolutionize the way
astronomers classify galaxies.

31
00:01:57,517 --> 00:02:01,254
Traditionally, astronomers
would classify them by hand

32
00:02:01,254 --> 00:02:03,790
spirals, ellipticals, irregulars,

33
00:02:03,790 --> 00:02:07,460
and so on, a time consuming process.

34
00:02:07,460 --> 00:02:11,331
To test AI concepts, Hubble
data has been used in studies

35
00:02:11,331 --> 00:02:15,168
on deep learning techniques,
which use artificial neural network

36
00:02:15,168 --> 00:02:21,407
technologies to learn complex patterns
and features from large data sets.

37
00:02:21,407 --> 00:02:26,012
In this case, the Hubble Candels data
set that contains over 10,000

38
00:02:26,012 --> 00:02:30,950
classified galaxies and required
dozens of astronomers time to classify,

39
00:02:30,950 --> 00:02:36,422
was used to teach a deep learning system
about galaxy structure and forms.

40
00:02:36,422 --> 00:02:39,125
They then applied it to the Hubble Legacy
field,

41
00:02:39,125 --> 00:02:43,596
a combination of nearly 7500 separate
Hubble exposures,

42
00:02:43,863 --> 00:02:46,733
representing 16 years of observations

43
00:02:46,733 --> 00:02:51,171
that contained over 265,000 galaxies.

44
00:02:51,171 --> 00:02:56,209
With deep learning, AI can sort
such a field at an incredible speed.

45
00:02:56,209 --> 00:02:59,179
This ability becomes even more important
when NASA's

46
00:02:59,179 --> 00:03:03,049
Nancy Grace Roman Space
Telescope images become available.

47
00:03:03,049 --> 00:03:07,987
Each of those images has over 100 times
the field of view as Hubble.

48
00:03:07,987 --> 00:03:11,191
Similarly,
Hubble data has been used to help test

49
00:03:11,191 --> 00:03:14,460
machine learning concepts to accommodate
noisy data.

50
00:03:15,528 --> 00:03:17,530
Telescope imperfections or dust

51
00:03:17,530 --> 00:03:22,168
clouds can create flaws in images
from even the best observatories.

52
00:03:22,168 --> 00:03:23,603
The machine learning systems

53
00:03:23,603 --> 00:03:27,774
can identify the noise
and reconstruct the image accordingly.

54
00:03:27,774 --> 00:03:32,545
Noisy Hubble Galaxy
images were used to test these concepts.

55
00:03:32,545 --> 00:03:35,548
Future telescopes
like the Roman Space Telescope,

56
00:03:35,548 --> 00:03:38,251
will collect more data than ever before.

57
00:03:38,251 --> 00:03:44,357
Roman itself will likely amass 20,000TB
of data over its five year mission.

58
00:03:44,357 --> 00:03:49,262
AI will be critical in processing
such vast streams of information.

59
00:03:49,262 --> 00:03:54,734
From finding exoplanets to uncovering new
clues about dark matter and dark energy.

60
00:03:54,734 --> 00:03:57,203
The next frontier of space exploration

61
00:03:57,203 --> 00:04:01,140
will not only depend on our instruments,
but also on powerful software.

62
00:04:01,140 --> 00:04:02,609
Here on Earth.

63
00:04:02,609 --> 00:04:06,913
And Hubble
data is helping with this new revolution.

64
00:04:06,913 --> 00:04:09,649
The fusion
of Hubble's vast archive of data

65
00:04:09,649 --> 00:04:14,254
with cutting edge
AI marks a new era in space exploration.

66
00:04:14,254 --> 00:04:19,158
The vast potential of machine learning
can process unimaginable amounts of data,

67
00:04:19,158 --> 00:04:21,794
unlocking discoveries faster than before

68
00:04:21,794 --> 00:04:25,164
and recovering details
we would otherwise have missed.

69
00:04:25,164 --> 00:04:29,569
The future of astronomy isn't just up in
the stars, it's in the code.

70
00:04:29,569 --> 00:04:47,253
Follow Hubble on social media @NASAHubble

71
00:04:47,253 --> 00:04:54,160
NASA Meatball