WEBVTT Kind: captions; Language: fi

1
00:00:00.860 --> 00:00:02.920
Thank you.
Thank you.

2
00:00:02.920 --> 00:00:05.090
Sen jo.

3
00:00:05.090 --> 00:00:10.740
Se on hello everyone jari
lindroos midot toivonen.

4
00:00:10.740 --> 00:00:18.800
And we are here, drive your presentation
about the future of AI bastuments.

5
00:00:18.800 --> 00:00:26.250
Presentation will cover prefer description
of AI and some of its possible applications.

6
00:00:26.250 --> 00:00:31.590
The were going to discuss the
current trends in AI little bear

7
00:00:31.590 --> 00:00:38.500
about effect of AI
entinen som constructions.

8
00:00:38.500 --> 00:00:44.010
Se first eii and its applications.

9
00:00:44.010 --> 00:00:49.980
S may now there school different
offices in computer science, artificial

10
00:00:49.980 --> 00:00:54.650
intelligence for AI iso
branch of computer science.

11
00:00:54.650 --> 00:00:59.630
Thames to create intelligent
machinesher capable of performing

12
00:00:59.630 --> 00:01:05.200
tass typically record human intelligence.

13
00:01:05.200 --> 00:01:11.390
X art fordstens visual perception, speech
recognition, decision making and problem solving

14
00:01:11.390 --> 00:01:18.590
clear related to, say tunes, machine
learning and exploring machine learning reforst

15
00:01:18.590 --> 00:01:25.050
modelslerin from the training
data without exclusive kiven ross.

16
00:01:25.050 --> 00:01:31.960
Diploning or deep neural networks in love,
especially dipper architects, consists

17
00:01:31.960 --> 00:01:41.180
of multitouch of players and
utilization of very large training data.

18
00:01:41.180 --> 00:01:48.540
Anders here pictures, the future visual
sound neural networks work on the left.

19
00:01:48.540 --> 00:01:56.650
The crash of the verisinble, the neural
network that one hidden layer age, the input

20
00:01:56.650 --> 00:02:03.480
spring fest tuned work with one or several
heidän layers and lentäneetgers old and output

21
00:02:03.480 --> 00:02:11.290
we andreas city dd sul have lot more
players and are more complicated.

22
00:02:11.290 --> 00:02:14.110
Architecture.

23
00:02:14.110 --> 00:02:19.370
Win convert two architecture
expressed in here.

24
00:02:19.370 --> 00:02:27.470
And on the right we haven example, image import
two visualisointien ens work and the visibleyer

25
00:02:27.470 --> 00:02:35.760
hash pictures of colors directly human
ai and works asti input and players

26
00:02:35.760 --> 00:02:42.590
there contribute to more find great information
bing based on the network teemakin possible

27
00:02:42.590 --> 00:02:52.560
prosessi dada match with
several levels of abstraktion.

28
00:02:52.560 --> 00:02:58.840
More precise ly classic machine learning
modules learn from and design features

29
00:02:58.840 --> 00:03:04.440
the dar drawn from the inputda
tismalleen sten created mapping from the

30
00:03:04.440 --> 00:03:09.290
features in order to
provide the output data.

31
00:03:09.290 --> 00:03:13.830
Se on itse freddy simple
madal deploying mads.

32
00:03:13.830 --> 00:03:21.110
However, futurice representation learning
away than design features and totta modules

33
00:03:21.110 --> 00:03:27.340
learned simple features of the data theon
with the additional hidden layers taylorin

34
00:03:27.340 --> 00:03:35.190
more abstract features from the dark and
from peace abstract feature mapping features

35
00:03:35.190 --> 00:03:45.200
mapping are formed the walk in port.

36
00:03:45.200 --> 00:03:51.710
Heartuter swit boxin blackbox
future explained difference between.

37
00:03:51.710 --> 00:03:58.950
Tis rupea depending systems with rupes
systems, we have expensive britann ross.

38
00:03:58.950 --> 00:04:05.630
The algoritmi flows se on natural wikin
track see when and how the values of

39
00:04:05.630 --> 00:04:13.790
paramedic change the code running so
we defining mads orsonce features of

40
00:04:13.790 --> 00:04:20.600
the deal bingoern and the moralstra,
future sound, etcetera cowen training deploy

41
00:04:20.600 --> 00:04:24.750
single we are mainly of
war of the imports and outs.

42
00:04:24.750 --> 00:04:31.650
No the way modelers mining how it comes,
the conclusion sor kives predictions,

43
00:04:31.650 --> 00:04:39.440
il happens in the dark,
taas the name blackbox.

44
00:04:39.440 --> 00:04:45.740
Ja se jäi systems become complex understanding
the decision making prosessissa becomes

45
00:04:45.740 --> 00:04:54.080
crucial premantura deller about black
boxes and how challengesplayn hawaii sistem

46
00:04:54.080 --> 00:05:01.440
srivations explainedble, am so dresseshu
by developing methods and technics the provide

47
00:05:01.440 --> 00:05:10.530
transparency and the prebility
building trustin ei jäisestä widell.

48
00:05:10.530 --> 00:05:14.760
Partsille oli en files
li healthcare financial.

49
00:05:14.760 --> 00:05:23.020
Tällöin force mand were
critical vision surfing mate.

50
00:05:23.020 --> 00:05:29.740
So before you think IMA ll system and the
future and the traditional way of training,

51
00:05:29.740 --> 00:05:36.570
air mads inves using cards, datasets
training prosess tätä opens in sopiwa sector

52
00:05:36.570 --> 00:05:43.110
and superwood manner and supervision learning
the mahdollis provided with input output

53
00:05:43.110 --> 00:05:49.920
cars and lunch mape input
the core spanking output.

54
00:05:49.920 --> 00:05:57.810
Siis prosessi records many label data mining
tuman sneaker have had the day deal with

55
00:05:57.810 --> 00:06:05.340
text answers before starting the train, the
model the model lens during training process,

56
00:06:05.340 --> 00:06:13.840
the general general from this label examples
and make predictions on new anssin dada

57
00:06:13.840 --> 00:06:20.220
example of classification
task the images of cards.

58
00:06:20.220 --> 00:06:23.520
Oikeasti import and the
names of the records act as

59
00:06:23.520 --> 00:06:30.610
labels or the design output indicate.

60
00:06:30.610 --> 00:06:35.370
Nata learning ei sköld unsubvist learning.

61
00:06:35.370 --> 00:06:41.560
Niin antsu learning the maple leafs, parents
and structures of the data without any label

62
00:06:41.560 --> 00:06:48.870
information, for example, in flashing the
distance of date points to the cluster centre

63
00:06:48.870 --> 00:06:57.110
point source, central alcatel
intuitive intuition www.ld protection

64
00:06:57.110 --> 00:07:05.190
of tweets caster is the point belongs
two, for example, this picture centres.

65
00:07:05.190 --> 00:07:12.960
Market with wide crush.
Turbine blasters overol.

66
00:07:12.960 --> 00:07:23.710
Andrew's burning camber useful for tass
life feature extraction or data compression.

67
00:07:23.710 --> 00:07:30.040
Son next we have some of
the main domains of AI.

68
00:07:30.040 --> 00:07:36.120
Vanhusten is natural language processing on
LPN short fokuksessa on meikkien, computers

69
00:07:36.120 --> 00:07:44.240
understand interest and generate human language
edin walls täss like massage translation,

70
00:07:44.240 --> 00:07:53.450
toksiset itsection, text, some recreation
and recognition jericho on open source

71
00:07:53.450 --> 00:08:00.520
text dataset consists of english
french sentences, for example.

72
00:08:00.520 --> 00:08:08.840
Niin english are alone is translated some
4 että schönau teette sen siis ard original

73
00:08:08.840 --> 00:08:17.230
scrap from the web and the english french
sound to be collin two each other and cetis

74
00:08:17.230 --> 00:08:26.350
very large concept of around
twenty two miljoonan samples.

75
00:08:26.350 --> 00:08:35.010
Natur mayn domain is computer vision
oreja insert images computers to extract

76
00:08:35.010 --> 00:08:43.130
min for information from image or
videoday ydinalue sanchez objektit action

77
00:08:43.130 --> 00:08:52.600
images classification image segmentation, facial
recognition and visual siinä understanding.

78
00:08:52.600 --> 00:08:59.480
Example en america setting record power 100.
sad connecting of americ images of brains and

79
00:08:59.480 --> 00:09:10.150
classified the two different categories
defrag having schumer or no tour.

80
00:09:10.150 --> 00:09:18.230
Ven we have rainforest learning
for it was the training of nayaged

81
00:09:18.230 --> 00:09:26.560
to make cecil visions true interaction
with the warmment agent lens actions

82
00:09:26.560 --> 00:09:32.260
maximissa records signal based on feedback
tai siis research research from the

83
00:09:32.260 --> 00:09:39.750
infant RLS applicationsin gambling
robotics recommendations systems.

84
00:09:39.750 --> 00:09:46.510
En olet syönyt muassa viikossa for example
entis one coding competition decoder

85
00:09:46.510 --> 00:09:53.390
switch ready agence deadly football
NS tubecon football game the objective

86
00:09:53.390 --> 00:10:00.730
was to have your team the score morten
tai accept the resource bittinen render

87
00:10:00.730 --> 00:10:08.110
dh space freddy fordbles mutation.

88
00:10:08.110 --> 00:10:15.420
The last one easy odile ei näy ed focuses on
the analysis, processing and understanding of

89
00:10:15.420 --> 00:10:23.510
odin data including speech recognition,
music, analysis of business and more.

90
00:10:23.510 --> 00:10:28.320
For example, in speech recognition
spostilla with convert intervention

91
00:10:28.320 --> 00:10:33.980
text idin walls technics like the
match with recognition and voice

92
00:10:33.980 --> 00:10:37.870
command recognition,
speed, recognitionisuus.

93
00:10:37.870 --> 00:10:43.270
Virtual assistance transcription
services, voisi control, ld systems and

94
00:10:43.270 --> 00:10:51.720
voice biometrich annin finlanders
project lahjoita puhetta speech finnish

95
00:10:51.720 --> 00:10:59.950
dayspin collected advance
finnish recognition.

96
00:10:59.950 --> 00:11:07.960
We are you only lupaus lil ***** about
current state of AI so artificial intelligence,

97
00:11:07.960 --> 00:11:16.240
magnificent progress in recent cs and
general motors chachi release last year

98
00:11:16.240 --> 00:11:20.870
or the forefront of the progress showing
over one point seven miljardin user

99
00:11:20.870 --> 00:11:25.940
billion users in itre anders ry.

100
00:11:25.940 --> 00:11:32.900
Ei itse technology on great
integration into stoch healthcare finance,

101
00:11:32.900 --> 00:11:41.080
transportation entertainment from
voisi sitten alexa thureson algorithms

102
00:11:41.080 --> 00:11:48.760
just streaming services
ei ISU daily daily lives.

103
00:11:48.760 --> 00:11:54.290
Itseäsi enable significantly dvdmentin
natural language processing, computer

104
00:11:54.290 --> 00:12:01.910
vision and other fields of
machine learning wellness robotics.

105
00:12:01.910 --> 00:12:07.100
Anteeksi slaidissa but something
released today maybe jari

106
00:12:07.100 --> 00:12:12.590
crowe joo sow the breaking news for today.

107
00:12:12.590 --> 00:12:14.630
Somebody has not yet heart.

108
00:12:14.630 --> 00:12:22.450
Google reveal the newest large language
model jimmy antis sai multimodal

109
00:12:22.450 --> 00:12:28.040
break tru andres fordissa newer
tai multimodaalinen traction kanssa

110
00:12:28.040 --> 00:12:31.090
unlike traditional language models.

111
00:12:31.090 --> 00:12:39.500
Gemini, kansi integrated pest, images,
audio, video code and disvasta first

112
00:12:39.500 --> 00:12:45.370
model out perform human experts on
massive multitasking language understanding

113
00:12:45.370 --> 00:12:49.440
benchmark switch, one of
the most popular states.

114
00:12:49.440 --> 00:12:57.820
No problem solving of you and it
also sur pasta gt for model in every

115
00:12:57.820 --> 00:13:02.240
text and the multimedia
benchmark except for the helasvuo.

116
00:13:02.240 --> 00:13:08.660
Benchmark switch for evaluation,
communications, reasoning for every day tass.

117
00:13:08.660 --> 00:13:17.060
And googlella has released
modelling thesis ultra and nano

118
00:13:17.060 --> 00:13:24.910
all for different scalability purposes.

119
00:13:24.910 --> 00:13:32.360
Next will about
current trends. I.

120
00:13:32.360 --> 00:13:39.480
So the man technology behind the many of
the corn top models, and even the newest

121
00:13:39.480 --> 00:13:48.680
google cmy ist transformer tää transformers,
mrs grand breaking modalarkkitehtuurin

122
00:13:48.680 --> 00:13:56.180
ennen and very first manly just for täss
light machine translation watch later

123
00:13:56.180 --> 00:13:59.870
thanbyn also juutilaisten computer vision.

124
00:13:59.870 --> 00:14:07.800
Nauti har presidentin ever multimodal
capabilities, no deal and videos osuu.

125
00:14:07.800 --> 00:14:14.440
Tämän benefit of the architecture
ist attention bäst algoritmi.

126
00:14:14.440 --> 00:14:19.650
Että me tismaksi tää
possible for the model to.

127
00:14:19.650 --> 00:14:27.820
Learn features that arch maptin long
distance and able to calculate mm with

128
00:14:27.820 --> 00:14:35.840
less computational efficiency and
hearing the picture wc related words.

129
00:14:35.840 --> 00:14:43.060
Ja evening after the fairway
distance tag still be map together.

130
00:14:43.060 --> 00:14:46.940
En tiedä siis tää main difference
from the traditional deep

131
00:14:46.940 --> 00:14:55.890
neural networks th ben justiin the past.

132
00:14:55.890 --> 00:14:57.950
Se on.

133
00:14:57.950 --> 00:15:03.900
Somaattisten idästä dead
muodostat SBN best on transformers.

134
00:15:03.900 --> 00:15:12.530
Bert fort for by direction and core
presentations for transformers, GBT generator,

135
00:15:12.530 --> 00:15:20.810
free trend transformer alsovision
transformer and the west jimmy diesel

136
00:15:20.810 --> 00:15:28.600
morales, different variations for
advanced language understanding urlang,

137
00:15:28.600 --> 00:15:32.040
text generation and transfer learning.

138
00:15:32.040 --> 00:15:35.700
An tärkeämpi different types
of models further developed

139
00:15:35.700 --> 00:15:38.140
for training based on this models.

140
00:15:38.140 --> 00:15:46.360
For example, on finnish language, theresa
finnbert model with based on the bert backer vaatii

141
00:15:46.360 --> 00:15:54.210
taas trendi on levit finnish institute
of the original bertha base english.

142
00:15:54.210 --> 00:16:02.170
Osa piti messun familar tuolla
**** off chachi björ getre

143
00:16:02.170 --> 00:16:08.870
get for and presenter
tai buff generate model.

144
00:16:08.870 --> 00:16:15.050
Also the final example sta vision
transformer model presenting

145
00:16:15.050 --> 00:16:18.900
the doll ior stable diffusion top model.

146
00:16:18.900 --> 00:16:27.900
Www.create, nice work of art from
just promoting suurta tai pop keywords.

147
00:16:27.900 --> 00:16:34.870
Zoo mä pidän about promptting
prompt pingistä process of designing

148
00:16:34.870 --> 00:16:39.430
prompt gite guide and
direct language model.

149
00:16:39.430 --> 00:16:45.210
Tukin generated wanted to
put and siis väri crucial

150
00:16:45.210 --> 00:16:50.540
off juutilaisen the language correctly.

151
00:16:50.540 --> 00:16:56.270
Carefully construct prompt provide
language model with necessary record

152
00:16:56.270 --> 00:17:01.420
text instructions and
information product out pot.

153
00:17:01.420 --> 00:17:08.080
Why one ***** disneyn cloud clearly
outlined what the askissa että

154
00:17:08.080 --> 00:17:13.440
hand for example samanväriset
text of something.

155
00:17:13.440 --> 00:17:21.220
Alright anchor discussion
about this and we were provided

156
00:17:21.220 --> 00:17:24.930
relevant backround
informations and examples.

157
00:17:24.930 --> 00:17:33.080
Two further give the language
concepts and more understanding ovat

158
00:17:33.080 --> 00:17:40.020
must fall elisa about interesting and
defining eksperimentin the different

159
00:17:40.020 --> 00:17:44.040
prompt and refine them
based on the language models.

160
00:17:44.040 --> 00:17:51.370
Response tuo optimaalista und resorts.

161
00:17:51.370 --> 00:17:57.990
Tuo sheratonve referencetyä
sables of debt morning

162
00:17:57.990 --> 00:18:02.380
we new dayta org content generated.

163
00:18:02.380 --> 00:18:10.490
Using existing data häntä go list
product close to the original input.

164
00:18:10.490 --> 00:18:12.290
Niin siinä yritti.

165
00:18:12.290 --> 00:18:17.540
Ei jäi siis teep morning algoritmeissa
are just to learning paths and features

166
00:18:17.540 --> 00:18:25.790
in the dayta watch may consists of
code, text, images ortio video or any

167
00:18:25.790 --> 00:18:34.230
other day tours and his generate jäi
hänen airbnb uni model are multimedia

168
00:18:34.230 --> 00:18:39.530
ja niin unimedel online
data types just soccer.

169
00:18:39.530 --> 00:18:46.600
Teksti teksti generation or
images image image generation.

170
00:18:46.600 --> 00:18:51.600
Wen toki bot multimedia liity
vie juu several moons link

171
00:18:51.600 --> 00:18:58.390
text image teksti speak generations.
Sing from the examples.

172
00:18:58.390 --> 00:19:07.700
Trumpting happy hippo, flying the sky
and the siksi magyar what model things?

173
00:19:07.700 --> 00:19:13.940
The prompt looks like.

174
00:19:13.940 --> 00:19:18.480
Soo bit more bot lords
language model senttiä diffusion

175
00:19:18.480 --> 00:19:25.210
models edelleen ms like GBT for the buff.

176
00:19:25.210 --> 00:19:30.920
Deep burning model that been
trained on massive of data can generate

177
00:19:30.920 --> 00:19:37.050
human light text common uses Marin
groundsring questions some rising

178
00:19:37.050 --> 00:19:46.460
translateting generating soratien blocker.
Doing your homework?

179
00:19:46.460 --> 00:19:55.430
And other models diffusion model sar osuva
tai generated modules dating and significant

180
00:19:55.430 --> 00:20:03.070
attention in the field of watching learning
pikaisesti, siis models by design to generate

181
00:20:03.070 --> 00:20:08.730
väri realistic samples mennä
prosecco diffusion prosess.

182
00:20:08.730 --> 00:20:15.880
En ties diffusion model slide
down your stable diffusion.

183
00:20:15.880 --> 00:20:24.600
Learn from vasta month of data theben
trend on meni meni metsään artsvit

184
00:20:24.600 --> 00:20:26.400
tää.

185
00:20:26.400 --> 00:20:33.830
Prompt off meni meni different taakse
categories related työtä ary am and we have the

186
00:20:33.830 --> 00:20:42.120
belly tuo signature realistit text
or art images and also hyvän music.

187
00:20:42.120 --> 00:20:45.530
Niin pitch switch tower,
exciting possibility.

188
00:20:45.530 --> 00:20:50.910
Meni creative applications.

189
00:20:50.910 --> 00:20:58.360
So but juvenes break there SBN actually
juutilaissing täällä arts angeles moralesin

190
00:20:58.360 --> 00:21:06.880
green forcen turning EU reka
NA agen developed by vidia.

191
00:21:06.880 --> 00:21:14.180
Vitsi vitsi primary function vastuu
trend robots superfood kompleksistasch but

192
00:21:14.180 --> 00:21:20.820
ja induced by automatically generating
reward algoritmit witch niistä meidän

193
00:21:20.820 --> 00:21:25.620
principal for training
growth from greenford morning.

194
00:21:25.620 --> 00:21:32.270
So basically the agent twitchin
disc case Vaasan GPT for.

195
00:21:32.270 --> 00:21:39.150
Itse se ei jäi tapsin tätä model hän
tee generative ei jäi rights the software

196
00:21:39.150 --> 00:21:44.830
code täält rewards the robots for
the ring forsman learning and world is

197
00:21:44.830 --> 00:21:52.330
fascinating start this eurekan tosin
actually rek warrenia specific promptting

198
00:21:52.330 --> 00:21:58.130
or any pre defined rewe varten place.
Mitkä on?

199
00:21:58.130 --> 00:21:59.930
Niinpä niin.

200
00:21:59.930 --> 00:22:02.900
Corporate human feedback to
the reward algorithm what is

201
00:22:02.900 --> 00:22:08.690
completely just based
on the prompt GB defor?

202
00:22:08.690 --> 00:22:16.790
Onhan tää robotsinti seuraa wordtu performed nearly
hurtig different tasks witch blues räpit benzinning

203
00:22:16.790 --> 00:22:23.540
trick the base gröning walking opening
rovers and tossa balsa and such.

204
00:22:23.540 --> 00:22:26.460
Oot also what?

205
00:22:26.460 --> 00:22:33.770
GBT for hasbeen just is to play
minecraft and this is basically.

206
00:22:33.770 --> 00:22:41.310
Base on tässä ei mekanismi but indicate
the game and cheek for basic communicate

207
00:22:41.310 --> 00:22:45.610
with shower and providing
the current state of the game.

208
00:22:45.610 --> 00:22:48.470
Tota cheek before *****.

209
00:22:48.470 --> 00:22:55.370
Air critics on informo peter the program
archives tätä ask hän hinat edition ift

210
00:22:55.370 --> 00:23:01.610
task files it will provide kritiik church
justinin how to complete task, for example,

211
00:23:01.610 --> 00:23:08.840
how autobild course for
missingeson wood go pants tree.

212
00:23:08.840 --> 00:23:16.010
What next?
Stop about ethics jae.

213
00:23:16.010 --> 00:23:23.990
So evento many people justice
generateve scsi piti gizmodon

214
00:23:23.990 --> 00:23:29.140
still product resource
combe facial incorrect.

215
00:23:29.140 --> 00:23:37.400
Antis what wereface ei tai
hallusinaation and this happens generation

216
00:23:37.400 --> 00:23:41.380
tatar trend on such marssiva mount data.

217
00:23:41.380 --> 00:23:49.620
Data included both the facial infoo
and also increase information witch in

218
00:23:49.620 --> 00:23:51.420
resort.

219
00:23:51.420 --> 00:23:56.740
Caldes muodostui generation
tekstit false flag.

220
00:23:56.740 --> 00:24:02.750
Siinä siis.
Sumari säätiön palon tässä screen.

221
00:24:02.750 --> 00:24:09.900
Väestö suomalaisia linkittää
dosentiksi just just bastont link.

222
00:24:09.900 --> 00:24:15.800
And also wild.
This can have some creative applications.

223
00:24:15.800 --> 00:24:24.600
The post significant ethical concernsches
tai mies information disinformaation.

224
00:24:24.600 --> 00:24:29.940
En discord of agereitä, että
hallusinaation cool be creator

225
00:24:29.940 --> 00:24:34.620
spread falls or miss
clearing information wide.

226
00:24:34.620 --> 00:24:40.270
Kyllähän potentiaali manipuloi
public opinion on course harm.

227
00:24:40.270 --> 00:24:47.410
No se on this hallusinaation school
reflect amplifier beach in training daytä

228
00:24:47.410 --> 00:24:57.930
leading to discrimination
output stat canal björn full.

229
00:24:57.930 --> 00:25:06.350
Soul ja hallusinaation on one example
of the limitationste ja systems may all the

230
00:25:06.350 --> 00:25:15.770
show this important to consider letician
using AI dis minst center race face

231
00:25:15.770 --> 00:25:22.620
when design this and designing and
implementingis ja systems visual osu.

232
00:25:22.620 --> 00:25:30.610
The consideration certain matters
like privacy, basen jobs placement.

233
00:25:30.610 --> 00:25:38.070
Minun ****** adress concerns to ensure
responsible and fair use of technology.

234
00:25:38.070 --> 00:25:43.690
Visual asbestin slide haglund
ensure tätä ja system Priority

235
00:25:43.690 --> 00:25:48.000
chima webbiin liittää human välitys.

236
00:25:48.000 --> 00:25:51.850
Ovat mekanismi shouldeen
place prevention system from

237
00:25:51.850 --> 00:25:56.310
coaching farmorbet rating yes and more.

238
00:25:56.310 --> 00:26:04.820
Promote transparency and
accountability in AI decision making.

239
00:26:04.820 --> 00:26:11.250
Se on pajassa ennen stance
criticalation, että niistä beatrice.

240
00:26:11.250 --> 00:26:18.530
Basiskan rise from best training daytä
or algoritmit byers leading to onfer

241
00:26:18.530 --> 00:26:25.600
art comes and michiganin bass records
actionia divers and representatives

242
00:26:25.600 --> 00:26:32.310
training day we detection
and fairness over algorithms.

243
00:26:32.310 --> 00:26:37.040
Ei ja ja applications often
invalid collection house hand store

244
00:26:37.040 --> 00:26:42.200
of personal databach races
meni privacy concerns.

245
00:26:42.200 --> 00:26:49.340
Walsall crucial tuen suurta responsible
and ethical data, which may include informed

246
00:26:49.340 --> 00:26:55.050
consent data, anonymisation and
secure data security measures.

247
00:26:55.050 --> 00:27:00.820
Istuu protect users
individuals privacy rights.

248
00:27:00.820 --> 00:27:08.490
The way tu hands tai problems in
contrast to machine learning, ice machine

249
00:27:08.490 --> 00:27:14.960
underning witches, the process of
removing or modify data or knowledge

250
00:27:14.960 --> 00:27:18.480
from the already trend schillerin model.

251
00:27:18.480 --> 00:27:24.480
Diesel offers meni adventist
siis over just ret raining model.

252
00:27:24.480 --> 00:27:31.100
Pikaiset svetari koostui and lähteekö
väri väri long time trendissä large model

253
00:27:31.100 --> 00:27:41.190
show juutilaisen best of
technics verin port in the field.

254
00:27:41.190 --> 00:27:49.250
Soul one type of research goin
oninea, texas ja limente gt refer

255
00:27:49.250 --> 00:27:58.560
to lining ei olisi systems with human välillä sand
goes len short day active way tai isä benefitcial.

256
00:27:58.560 --> 00:28:08.080
Every body and etsi cala this to prevent
on intent consequences and potential risks.

257
00:28:08.080 --> 00:28:13.870
Ja siistem mieheltä control for
the reaction established clear lines

258
00:28:13.870 --> 00:28:20.790
of responsibility, contability
inceptioncial particularille en domains

259
00:28:20.790 --> 00:28:24.780
ja systems make the critical decisions.

260
00:28:24.780 --> 00:28:30.600
Warsaw transparency explain ability
in decision making prosessissa

261
00:28:30.600 --> 00:28:40.830
enable scouting of the
deputy and user trust.

262
00:28:40.830 --> 00:28:46.690
Ja se on vaan exempel off
controllercial house faye vendée.

263
00:28:46.690 --> 00:28:53.250
Ei jäi muodoista, että haben trained on
the art pictures dolly täältä originally

264
00:28:53.250 --> 00:28:59.570
from artist that may have not
given permission to just for art.

265
00:28:59.570 --> 00:29:02.610
Niin understand the belt tai artistit.

266
00:29:02.610 --> 00:29:07.150
Me ei vaan prestige uusi
tarjous of the art in order to

267
00:29:07.150 --> 00:29:11.370
preserve the possible
future opportunities.

268
00:29:11.370 --> 00:29:15.850
Fortis teräsbyn develop
tiskiin of technology SW ei tuo

269
00:29:15.850 --> 00:29:19.790
counter arctis taipu
wanted data scripting.

270
00:29:19.790 --> 00:29:27.270
For instance, saa program tai alterran
imagine way trip stay muodolla.

271
00:29:27.270 --> 00:29:34.950
Vesikeli fréjus program willard
small amount of noise, tota image.

272
00:29:34.950 --> 00:29:40.440
Tai siis minimalismi hieman
näe palatsit vielä riley

273
00:29:40.440 --> 00:29:48.360
hinderta performance softia ja modo.

274
00:29:48.360 --> 00:29:54.980
Sonax, we have sun constructions.

275
00:29:54.980 --> 00:30:03.100
So about the possibilities tai lay
ei research ei will be integrated to

276
00:30:03.100 --> 00:30:10.140
every day life with smart homes, soft
ravin cars, virtual assistance and

277
00:30:10.140 --> 00:30:16.060
power device in hansin
convenience and infections.

278
00:30:16.060 --> 00:30:23.530
Eilisistä will super understand and
generate human language sweater conversation.

279
00:30:23.530 --> 00:30:30.270
Ations, translation services and
voice systemce and radar tän replaced

280
00:30:30.270 --> 00:30:35.860
comments ja vilmos likely comment
human capabilities and collaborate

281
00:30:35.860 --> 00:30:39.370
with astin solving complex problems.

282
00:30:39.370 --> 00:30:46.310
And ongoing research will focus on
improving capabilities, driving innovation

283
00:30:46.310 --> 00:30:51.320
and exploring of frontiers
tai exploringble aijai.

284
00:30:51.320 --> 00:31:00.120
Advanced neural network
architecture and quantum computing.

285
00:31:00.120 --> 00:31:02.230
Sen.

286
00:31:02.230 --> 00:31:09.630
Me ei vie about the bid for the future
of AI hold siemens proms, pot osui cs

287
00:31:09.630 --> 00:31:17.460
responsible and secure moment ei
aiveloments on hapen ja void but it will be

288
00:31:17.460 --> 00:31:24.990
continued shaped lives factions like
effica considerationns se sille choices,

289
00:31:24.990 --> 00:31:31.440
public public perception
and regulatory frameworkks.

290
00:31:31.440 --> 00:31:37.520
Se on furness transparency ja account
billy the soul, super priorityistin, future

291
00:31:37.520 --> 00:31:46.960
research to possible lies and
criminalation present in ei aio systems.

292
00:31:46.960 --> 00:31:49.620
Se on thank you all for listening.

293
00:31:49.620 --> 00:31:55.380
And there sun references.

294
00:31:55.380 --> 00:31:59.930
Maybe we have some time left.

295
00:31:59.930 --> 00:32:08.680
Shreeso are work that we have been
doing in the form of video ortio.

296
00:32:08.680 --> 00:32:15.630
The time just remember sharing the
video unity shorty audio as well.

297
00:32:15.630 --> 00:32:18.270
OK click when you.

298
00:32:18.270 --> 00:32:24.690
Inter share screen lowry jotenkin
this place de clichy audio.

299
00:32:24.690 --> 00:32:44.690
OK.

300
00:34:07.010 --> 00:34:29.150
Siitä satsausta täysillä.
Mä jotenkin check hei.

301
00:36:21.280 --> 00:36:26.400
Sothys video on specifically
about this infrastructure project.

302
00:36:26.400 --> 00:36:35.300
We have the working on antista source
exempel using my or more specifically

303
00:36:35.300 --> 00:36:41.100
name direction for archive taida
and then we have another video,

304
00:36:41.100 --> 00:37:01.100
may be wicked get the voice tiesimme.

305
00:37:08.340 --> 00:37:17.620
I'll live streaming.
Watch now now.

306
00:37:17.620 --> 00:37:24.510
Et smoothi just gay itse glow from.

307
00:37:24.510 --> 00:37:28.170
No cry department, the
cars beyond the surface and

308
00:37:28.170 --> 00:37:39.220
on what the voice shaping experience.

309
00:37:39.220 --> 00:37:58.500
The real thing, the cover, push the pois.

310
00:37:58.500 --> 00:38:05.880
This way we find the true story.
Iskua sen livestream.

311
00:38:05.880 --> 00:38:11.310
Whether youre kotit fan of justice, exploring
the twitch universe, the click to you

312
00:38:11.310 --> 00:38:17.830
covered it offers free brilliant ways to
chatata, giving you the inside you create

313
00:38:17.830 --> 00:38:23.350
the sh live on live action collected from
LIVE LIVE stream channel staying in the

314
00:38:23.350 --> 00:38:29.460
moment and catering segment
of the live experience.

315
00:38:29.460 --> 00:38:35.320
Täynnä hävennyt specific streamerin
mind no problem uc pace broadcast from

316
00:38:35.320 --> 00:38:42.370
specific gamo category filtering
by language macho preferences.

317
00:38:42.370 --> 00:38:48.550
One delfin 100 content of the picture stream
collect data from pace broadcast research

318
00:38:48.550 --> 00:38:58.330
live stream channel on the data from
specific video by its all of things.

319
00:38:58.330 --> 00:39:04.180
Rest, should you college dating security
studi cloud reddit beach where you

320
00:39:04.180 --> 00:39:09.630
need and if you prefer share copy, we got
you covered with the option to download

321
00:39:09.630 --> 00:39:14.790
the collected data after
the collections finish.

322
00:39:14.790 --> 00:39:17.930
Että se siitä begin sveitsin
the data just drag and drop

323
00:39:17.930 --> 00:39:23.110
the collected charter into our nousis two.

324
00:39:23.110 --> 00:39:30.820
Twitch messages beard singh, especially in
far space streams of views combi deco thl hus

325
00:39:30.820 --> 00:39:37.500
sade what with massage unicon meaningful one
way to address this problem is the tech unic

326
00:39:37.500 --> 00:39:44.800
massages the done by defining massages the
regional information and relevant to the stream

327
00:39:44.800 --> 00:39:50.060
archive learning based content detection solution
helps the filter round the more meaningful

328
00:39:50.060 --> 00:39:54.550
messages by classified
the food sting categories.

329
00:39:54.550 --> 00:39:59.210
Mitä olet visualization nations
cup dive deeper you chat content

330
00:39:59.210 --> 00:40:03.740
revealing trends,
patience and unic inside.

331
00:40:03.740 --> 00:40:14.620
On waldner natives and cover the voice
drive ea sports sen live streaming community.

332
00:40:14.620 --> 00:40:21.310
Mico, the true story behind the stream watch
the twitch and with your new kind of discourse

333
00:40:21.310 --> 00:40:27.650
in the making, combining text, visual
and highly config formulat expressions.

334
00:40:27.650 --> 00:40:33.220
This present spring challenge
helping fisting language models.

335
00:40:33.220 --> 00:40:37.760
Using out of the drive deeperin
the esports on live streaming, just

336
00:40:37.760 --> 00:40:42.020
ben fisher, its hockey
champ, for your research.

337
00:40:42.020 --> 00:40:46.220
Vesi lenkin description get
started and hannistien science.

338
00:40:46.220 --> 00:41:01.560
Open your research food.

339
00:41:01.560 --> 00:41:08.220
Sä otat voisi another word pages
infrastruktuuri project that we have been working

340
00:41:08.220 --> 00:41:16.080
or or jari hes mostly working on se
on thank you for all this morning.

341
00:41:16.080 --> 00:41:23.950
Ääntä että soundit guess.
Thank you so much. Thank you.