WEBVTT

1
00:00:04.865 --> 00:00:05.655
<00:00:04.865>Hello <00:00:05.576>everyone.

2
00:00:05.785 --> 00:00:09.415
<00:00:05.785>Today <00:00:05.960>I'm <00:00:06.135>going <00:00:06.310>to <00:00:06.485>introduce <00:00:06.660>data <00:00:06.835>for <00:00:07.010>Studio, <00:00:07.185>uh,

3
00:00:09.515 --> 00:00:11.775
<00:00:09.515>in <00:00:09.715>Short <00:00:09.915>Data <00:00:10.115>for <00:00:10.315>Studio <00:00:10.515>is <00:00:10.715>a <00:00:10.915>platform

4
00:00:11.925 --> 00:00:14.215
<00:00:11.925>that <00:00:12.275>helps <00:00:12.625>data <00:00:12.975>scientists <00:00:13.325>build

5
00:00:14.435 --> 00:00:16.375
<00:00:14.435>and <00:00:14.785>manage <00:00:15.135>machine <00:00:15.485>learning <00:00:15.835>models.

6
00:00:17.035 --> 00:00:18.935
<00:00:17.035>But <00:00:17.268>before <00:00:17.501>we <00:00:17.735>dive <00:00:17.968>deep <00:00:18.201>into <00:00:18.435>it,

7
00:00:19.305 --> 00:00:20.695
<00:00:19.305>let's <00:00:19.617>ask <00:00:19.930>ourselves <00:00:20.243>a <00:00:20.556>question.

8
00:00:21.275 --> 00:00:24.735
<00:00:21.275>Why <00:00:21.450>would <00:00:21.625>data <00:00:21.799>scientists <00:00:21.974>need <00:00:22.150>help <00:00:22.325>with <00:00:22.500>their <00:00:22.674>models?

9
00:00:25.545 --> 00:00:29.035
<00:00:25.545>After <00:00:25.825>all, <00:00:26.105>contradict <00:00:26.385>to <00:00:26.665>common <00:00:26.945>belief

10
00:00:30.435 --> 00:00:32.315
<00:00:30.435>building <00:00:30.634>machine <00:00:30.834>learning <00:00:31.035>models <00:00:31.235>is <00:00:31.435>not <00:00:31.634>that <00:00:31.834>hard.

11
00:00:32.455 --> 00:00:35.395
<00:00:32.455>In <00:00:32.610>principle, <00:00:32.766>it <00:00:32.921>can <00:00:33.077>be <00:00:33.232>done <00:00:33.388>in <00:00:33.543>reliance <00:00:33.699>of <00:00:33.855>code.

12
00:00:35.665 --> 00:00:39.715
<00:00:35.665>However, <00:00:35.839>building <00:00:36.015>ML <00:00:36.190>models <00:00:36.365>is <00:00:36.540>only <00:00:36.714>a <00:00:36.890>small <00:00:37.065>part

13
00:00:39.715 --> 00:00:41.075
<00:00:39.715>of <00:00:40.021>the <00:00:40.327>whole <00:00:40.633>model <00:00:40.939>lifecycle.

14
00:00:42.295 --> 00:00:45.915
<00:00:42.295>It <00:00:42.528>includes <00:00:42.761>data <00:00:42.995>preparation, <00:00:43.228>model <00:00:43.461>evaluation, <00:00:43.695>deployment,

15
00:00:45.915 --> 00:00:48.075
<00:00:45.915>and <00:00:46.265>production <00:00:46.615>monitoring <00:00:46.964>for <00:00:47.315>data

16
00:00:48.075 --> 00:00:49.875
<00:00:48.075>and <00:00:48.541>concept <00:00:49.008>drift <00:00:49.475>retraining,

17
00:00:50.835 --> 00:00:52.355
<00:00:50.835>managing <00:00:51.108>multiple <00:00:51.382>versions <00:00:51.655>and <00:00:51.929>so <00:00:52.202>on.

18
00:00:52.615 --> 00:00:55.315
<00:00:52.615>So <00:00:52.848>all <00:00:53.081>of <00:00:53.315>those <00:00:53.548>things <00:00:53.781>add <00:00:54.015>complexity,

19
00:00:55.495 --> 00:00:59.155
<00:00:55.495>making <00:00:55.650>even <00:00:55.806>the <00:00:55.961>simplest <00:00:56.117>use <00:00:56.272>cases, <00:00:56.428>not <00:00:56.583>so <00:00:56.739>simple <00:00:56.894>anymore.

20
00:01:00.995 --> 00:01:03.135
<00:01:00.995>So <00:01:01.195>while <00:01:01.394>it <00:01:01.595>might <00:01:01.794>be <00:01:01.995>fair <00:01:02.195>to <00:01:02.394>say

21
00:01:03.135 --> 00:01:06.455
<00:01:03.135>that <00:01:03.290>building <00:01:03.446>ML <00:01:03.601>models <00:01:03.757>is <00:01:03.912>easy, <00:01:04.068>uh, <00:01:04.223>the <00:01:04.379>reality <00:01:04.535>is

22
00:01:06.455 --> 00:01:09.135
<00:01:06.455>that <00:01:06.804>building <00:01:07.155>production <00:01:07.505>ready <00:01:07.855>systems

23
00:01:09.275 --> 00:01:10.375
<00:01:09.275>is <00:01:09.605>really, <00:01:09.935>really <00:01:10.265>hard.

24
00:01:12.675 --> 00:01:14.855
<00:01:12.675>And <00:01:12.850>one <00:01:13.024>of <00:01:13.200>key <00:01:13.375>reasons <00:01:13.550>for <00:01:13.725>this <00:01:13.899>difficulty <00:01:14.075>is

25
00:01:14.855 --> 00:01:17.375
<00:01:14.855>that <00:01:15.088>machine <00:01:15.321>learning <00:01:15.555>ecosystem <00:01:15.788>is <00:01:16.021>highly <00:01:16.255>fragmented.

26
00:01:18.285 --> 00:01:20.655
<00:01:18.285>Data <00:01:18.485>scientists <00:01:18.685>often <00:01:18.884>have <00:01:19.085>to <00:01:19.285>use <00:01:19.485>a <00:01:19.685>range

27
00:01:20.655 --> 00:01:24.255
<00:01:20.655>of <00:01:20.855>disconnected <00:01:21.055>tools, <00:01:21.255>uh, <00:01:21.455>for <00:01:21.655>data <00:01:21.855>processing, <00:01:22.055>experiments,

28
00:01:24.255 --> 00:01:25.455
<00:01:24.255>deployments, <00:01:24.615>and <00:01:24.975>so <00:01:25.335>on.

29
00:01:25.665 --> 00:01:28.495
<00:01:25.664>These <00:01:25.839>tools <00:01:26.014>typically <00:01:26.190>don't <00:01:26.365>integrate <00:01:26.539>well <00:01:26.714>for <00:01:26.889>each <00:01:27.065>other,

30
00:01:28.845 --> 00:01:31.695
<00:01:28.845>with <00:01:29.000>each <00:01:29.156>other, <00:01:29.311>create <00:01:29.467>an <00:01:29.622>lot <00:01:29.778>of <00:01:29.933>friction, <00:01:30.089>wasted <00:01:30.245>time,

31
00:01:31.835 --> 00:01:33.175
<00:01:31.835>and <00:01:32.036>a <00:01:32.237>lot <00:01:32.438>of <00:01:32.639>room <00:01:32.840>for <00:01:33.041>error.

32
00:01:34.475 --> 00:01:37.415
<00:01:34.475>So <00:01:34.630>to <00:01:34.786>address <00:01:34.941>this <00:01:35.097>data <00:01:35.252>force, <00:01:35.408>we <00:01:35.563>provide <00:01:35.719>a <00:01:35.875>set

33
00:01:37.415 --> 00:01:38.775
<00:01:37.414>of <00:01:37.823>well <00:01:38.231>integrated <00:01:38.639>components

34
00:01:39.165 --> 00:01:42.255
<00:01:39.164>that <00:01:39.339>from <00:01:39.514>a <00:01:39.690>single <00:01:39.865>unified <00:01:40.039>flow <00:01:40.214>covering <00:01:40.389>everything <00:01:40.565>from

35
00:01:42.255 --> 00:01:44.535
<00:01:42.255>model <00:01:42.535>building <00:01:42.815>to <00:01:43.095>deploying <00:01:43.375>into <00:01:43.655>production

36
00:01:44.535 --> 00:01:46.695
<00:01:44.535>and <00:01:44.884>post <00:01:45.235>deployment <00:01:45.585>deployment <00:01:45.935>monitoring.

37
00:01:48.575 --> 00:01:51.675
<00:01:48.575>But <00:01:48.808>what <00:01:49.041>exactly <00:01:49.275>makes <00:01:49.508>our <00:01:49.741>flow <00:01:49.975>unified

38
00:01:52.575 --> 00:01:54.995
<00:01:52.575>and <00:01:52.808>the <00:01:53.041>sized <00:01:53.275>in <00:01:53.508>model <00:01:53.741>centric <00:01:53.975>design,

39
00:01:55.455 --> 00:01:56.875
<00:01:55.455>unlike <00:01:55.881>a <00:01:56.307>common <00:01:56.733>approach

40
00:01:56.885 --> 00:02:00.035
<00:01:56.884>where <00:01:57.040>virtually <00:01:57.196>any <00:01:57.351>artifact <00:01:57.507>can <00:01:57.662>be <00:01:57.818>treated <00:01:57.973>as <00:01:58.129>a <00:01:58.285>model,

41
00:02:00.095 --> 00:02:01.995
<00:02:00.095>we <00:02:00.328>follow <00:02:00.561>a <00:02:00.795>different, <00:02:01.028>more <00:02:01.261>structured <00:02:01.495>path.

42
00:02:02.495 --> 00:02:04.875
<00:02:02.495>We <00:02:02.695>define <00:02:02.895>a <00:02:03.095>model <00:02:03.295>as <00:02:03.495>a <00:02:03.695>standardized <00:02:03.895>container

43
00:02:04.985 --> 00:02:07.835
<00:02:04.985>that <00:02:05.185>encapsulates <00:02:05.385>not <00:02:05.585>only <00:02:05.785>the <00:02:05.985>model <00:02:06.185>artifact <00:02:06.385>itself,

44
00:02:08.175 --> 00:02:11.315
<00:02:08.175>but <00:02:08.350>also <00:02:08.525>reach <00:02:08.700>set <00:02:08.875>of <00:02:09.050>meditator <00:02:09.225>describing <00:02:09.400>its <00:02:09.575>inputs,

45
00:02:11.385 --> 00:02:14.155
<00:02:11.385>outputs, <00:02:11.618>environment, <00:02:11.851>training <00:02:12.084>logs, <00:02:12.318>and <00:02:12.551>so <00:02:12.785>on.

46
00:02:15.495 --> 00:02:19.075
<00:02:15.495>So <00:02:15.670>this <00:02:15.845>design <00:02:16.020>allows <00:02:16.195>each <00:02:16.370>component <00:02:16.545>within <00:02:16.720>the <00:02:16.895>system

47
00:02:19.215 --> 00:02:20.275
<00:02:19.215>to <00:02:19.692>work <00:02:20.169>natively

48
00:02:20.465 --> 00:02:23.995
<00:02:20.465>with <00:02:20.665>our <00:02:20.865>model <00:02:21.065>dynamically <00:02:21.265>extracting <00:02:21.465>the <00:02:21.665>necessary <00:02:21.865>metadata

49
00:02:23.995 --> 00:02:26.075
<00:02:23.995>when <00:02:24.275>needed <00:02:24.555>without <00:02:24.835>any <00:02:25.115>additional <00:02:25.395>configuration.

50
00:02:27.375 --> 00:02:30.435
<00:02:27.375>So <00:02:27.575>using <00:02:27.775>this <00:02:27.975>standardized <00:02:28.175>self <00:02:28.375>contained <00:02:28.575>model <00:02:28.775>format

51
00:02:30.435 --> 00:02:33.995
<00:02:30.435>sounds <00:02:30.610>great <00:02:30.785>on <00:02:30.960>paper, <00:02:31.135>but <00:02:31.310>it <00:02:31.485>naturally <00:02:31.660>may <00:02:31.835>rise.

52
00:02:34.195 --> 00:02:38.275
<00:02:34.195>A <00:02:34.334>couple <00:02:34.475>of <00:02:34.614>questions. <00:02:34.755>So <00:02:34.894>the <00:02:35.035>first <00:02:35.174>one <00:02:35.315>is <00:02:35.454>this <00:02:35.595>approach

53
00:02:38.675 --> 00:02:41.755
<00:02:38.675>flexible <00:02:38.908>enough <00:02:39.141>to <00:02:39.375>support <00:02:39.608>different <00:02:39.841>model <00:02:40.075>types

54
00:02:41.815 --> 00:02:44.515
<00:02:41.815>and <00:02:42.048>use <00:02:42.281>cases <00:02:42.515>and <00:02:42.748>the <00:02:42.981>ansys <00:02:43.215>clear?

55
00:02:44.615 --> 00:02:47.195
<00:02:44.615>Yes. <00:02:46.015>Data

56
00:02:47.215 --> 00:02:49.955
<00:02:47.215>for <00:02:47.448>studio <00:02:47.681>supports <00:02:47.915>everything <00:02:48.148>from <00:02:48.381>traditional <00:02:48.615>machine

57
00:02:50.075 --> 00:02:54.315
<00:02:50.075>learning <00:02:50.274>models <00:02:50.475>on <00:02:50.674>tabular <00:02:50.875>data <00:02:51.075>to <00:02:51.274>modern <00:02:51.475>multi-component,

58
00:02:54.365 --> 00:02:57.675
<00:02:54.365>large <00:02:54.598>language <00:02:54.831>model <00:02:55.065>pipelines <00:02:55.298>and <00:02:55.531>agent-based <00:02:55.765>workflows.

59
00:02:57.975 --> 00:03:00.165
<00:02:57.975>So <00:02:58.208>we <00:02:58.441>really <00:02:58.674>have <00:02:58.908>no <00:02:59.141>limitations <00:02:59.375>here.

60
00:03:02.215 --> 00:03:05.315
<00:03:02.215>Uh, <00:03:02.448>but <00:03:02.681>what <00:03:02.915>about <00:03:03.148>the <00:03:03.381>vendor <00:03:03.615>lockin?

61
00:03:05.695 --> 00:03:09.115
<00:03:05.695>And, <00:03:05.975>uh, <00:03:06.255>we <00:03:06.535>do <00:03:06.815>not, <00:03:07.095>um,

62
00:03:10.025 --> 00:03:13.315
<00:03:10.025>lock <00:03:10.180>our <00:03:10.336>users <00:03:10.491>because <00:03:10.647>our <00:03:10.802>platform <00:03:10.958>is <00:03:11.113>open <00:03:11.269>by <00:03:11.425>design

63
00:03:13.415 --> 00:03:17.475
<00:03:13.415>and <00:03:13.554>it <00:03:13.695>is <00:03:13.834>also <00:03:13.975>open <00:03:14.114>source <00:03:14.255>under <00:03:14.394>the <00:03:14.535>Apache <00:03:14.674>tool <00:03:14.815>license.

64
00:03:19.055 --> 00:03:20.875
<00:03:19.055>In <00:03:19.288>addition <00:03:19.521>to <00:03:19.755>that, <00:03:19.988>the <00:03:20.221>core <00:03:20.455>module

65
00:03:20.935 --> 00:03:23.715
<00:03:20.935>of <00:03:21.135>our <00:03:21.335>platform <00:03:21.535>called <00:03:21.735>Orbits <00:03:21.935>was <00:03:22.135>built <00:03:22.335>specifically

66
00:03:23.715 --> 00:03:26.035
<00:03:23.715>to <00:03:23.948>avoid <00:03:24.181>vendor <00:03:24.415>lock <00:03:24.648>orbits <00:03:24.881>allow <00:03:25.115>you

67
00:03:26.035 --> 00:03:27.555
<00:03:26.035>to <00:03:26.263>bring <00:03:26.491>your <00:03:26.719>own <00:03:26.947>storage <00:03:27.175>and <00:03:27.403>compute.

68
00:03:27.855 --> 00:03:31.355
<00:03:27.855>So <00:03:27.982>data <00:03:28.109>for <00:03:28.236>Studio <00:03:28.364>X <00:03:28.491>as <00:03:28.618>a <00:03:28.745>thin <00:03:28.873>orchestrator <00:03:29.000>rather <00:03:29.127>than <00:03:29.255>a

69
00:03:31.955 --> 00:03:34.395
<00:03:31.954>a <00:03:32.154>heavy <00:03:32.355>platform <00:03:32.554>that <00:03:32.755>you're <00:03:32.954>tied <00:03:33.154>to, <00:03:33.355>uh,

70
00:03:34.395 --> 00:03:36.795
<00:03:34.395>all <00:03:34.628>your <00:03:34.861>data <00:03:35.095>stays <00:03:35.328>in <00:03:35.561>your <00:03:35.795>environment

71
00:03:36.795 --> 00:03:38.555
<00:03:36.795>and <00:03:37.028>nothing <00:03:37.261>is <00:03:37.495>sent <00:03:37.728>to <00:03:37.961>our <00:03:38.195>servers.

72
00:03:39.185 --> 00:03:40.675
<00:03:39.185>This <00:03:39.408>means <00:03:39.632>that <00:03:39.855>you <00:03:40.079>get <00:03:40.302>the <00:03:40.526>benefit

73
00:03:40.815 --> 00:03:44.195
<00:03:40.815>of <00:03:40.970>a <00:03:41.126>unified <00:03:41.281>clean <00:03:41.437>workflow <00:03:41.592>while <00:03:41.748>staying <00:03:41.903>fully <00:03:42.059>in <00:03:42.215>control

74
00:03:44.195 --> 00:03:45.635
<00:03:44.195>of <00:03:44.519>your <00:03:44.843>data <00:03:45.167>and <00:03:45.491>infrastructure.

75
00:03:47.265 --> 00:03:50.085
<00:03:47.265>So <00:03:47.464>if <00:03:47.665>you're <00:03:47.864>interested <00:03:48.065>to <00:03:48.265>try <00:03:48.464>the <00:03:48.665>platform,

76
00:03:50.105 --> 00:03:51.245
<00:03:50.105>you <00:03:50.310>can <00:03:50.515>scan <00:03:50.720>the <00:03:50.925>Q <00:03:51.131>code.

77
00:03:51.245 --> 00:03:52.525
<00:03:51.245>It's <00:03:51.629>already <00:03:52.013>live. <00:03:52.397>Some

78
00:03:52.525 --> 00:03:54.205
<00:03:52.525>of <00:03:52.758>the <00:03:52.991>modules <00:03:53.225>are <00:03:53.458>still <00:03:53.691>under <00:03:53.925>construction,

79
00:03:54.225 --> 00:03:55.365
<00:03:54.225>but <00:03:54.430>we <00:03:54.635>are <00:03:54.840>planning <00:03:55.045>a <00:03:55.251>lot

80
00:03:55.365 --> 00:03:57.125
<00:03:55.365>of <00:03:55.565>releases <00:03:55.765>in <00:03:55.965>a <00:03:56.165>couple <00:03:56.365>of <00:03:56.565>next <00:03:56.765>months.

81
00:03:57.785 --> 00:04:01.485
<00:03:57.785>Uh, <00:03:57.924>we <00:03:58.065>also <00:03:58.204>have <00:03:58.345>a <00:03:58.484>public <00:03:58.625>roadmap <00:03:58.765>on <00:03:58.905>our <00:03:59.045>GitHub <00:03:59.185>repository,

82
00:04:01.545 --> 00:04:02.805
<00:04:01.545>so <00:04:01.707>you <00:04:01.869>can <00:04:02.031>check <00:04:02.192>it <00:04:02.355>out <00:04:02.517>as <00:04:02.679>well.

83
00:04:03.385 --> 00:04:05.205
<00:04:03.385>Uh, <00:04:03.560>and <00:04:03.734>if <00:04:03.910>you <00:04:04.084>want <00:04:04.260>to <00:04:04.435>have <00:04:04.609>a <00:04:04.785>chat,

84
00:04:05.345 --> 00:04:07.005
<00:04:05.345>our <00:04:05.625>booth <00:04:05.905>is <00:04:06.185>right <00:04:06.465>over <00:04:06.745>there.

85
00:04:07.065 --> 00:04:08.325
<00:04:07.065>We <00:04:07.348>have <00:04:07.632>really <00:04:07.915>cool <00:04:08.198>stickers

86
00:04:08.705 --> 00:04:11.645
<00:04:08.705>and <00:04:08.845>also <00:04:08.985>a <00:04:09.125>giveaway, <00:04:09.265>so <00:04:09.405>feel <00:04:09.545>free <00:04:09.685>to <00:04:09.825>visit <00:04:09.965>us <00:04:10.105>there.

87
00:04:11.775 --> 00:04:12.245
<00:04:11.775>Thank <00:04:12.198>you.