Player is loading...
Privacy by Design Age Verification & Online Child Safety
Dr. Onur Yürüten, Head of Age Assurance Solutions and Senior ML Engineer in Privately
Wednesday, Oct. 11, 2023 · 9:26 a.m. · 15m 11s
Onur is the Head of Age Assurance Solutions and a Senior Machine Learning Engineer in Privately. You can ask him about Privately's privacy-preserving, biometrics-based age estimation products or anything you like. Before joining Privately, he delivered end-to-end machine learning solutions in a diversity of domains. He holds a Computer Science Ph.D. from EPFL.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:01
okay well uh heller one uh i'm uh on or usually known in these circles as the guy
00:00:08
between you and your lunchtime but i think for today's samuel was going to do the owners uh it
00:00:14
so i'm going to the talk about uh generally a age verification and uh an up and coming
00:00:20
requirement for a business is mostly online um we do well by magic
00:00:26
space age estimation and i want to demonstrate to that it is possible to
00:00:31
do this uh kind of work in a previous preserving like my company
00:00:35
is an uh a partial to fake in the in a next if l.
00:00:40
a. inserts and we also have an office in london we've got uh some
00:00:44
uh very solid again and partnerships we've been doing projects uh with it yet
00:00:49
we've got some uh interesting clients in the past upon coming
00:00:52
very interesting uh clients in the uh in this domain now um
00:00:58
the the winter set in the queue and they uh he's uh when
00:01:02
you know he's hoping that the regulations going to enable for his business well
00:01:06
regulations now of busy enabling might business today because uh
00:01:10
uh they have already started cracking down on a company's
00:01:14
uh on on various previous violations but the common thing
00:01:18
if you can look at the the some of the uh some of the rather recent headlines
00:01:23
i've drawn from you is the children's uh pretty see online so what do i mean um
00:01:30
for instance a children's profiles should be a private by default
00:01:35
or children should not be presented for a complex uh activities such
00:01:40
as a purchasing items without uh without parental consent um uh oh
00:01:47
or the the platforms should not be exposing a data or
00:01:51
any kind of user experience that is not appropriate for children
00:01:55
and that's uh that's not only a a no usual suspects such as a
00:02:00
rubber feet but it's also about violence for politics and i'm much more cell uh
00:02:06
so the regulation has started cracking down on these companies and
00:02:12
what they need to do what they need to do a two
00:02:16
so to avoid these uh these half the
00:02:20
uh prices is to make sure that they know
00:02:23
who are the child users in their platforms
00:02:27
so that there comes the age verification now um
00:02:32
typically the this age verification domain has
00:02:35
a lot of for intersections with identity verification
00:02:39
such as the it k. y. c. submitting your id
00:02:43
or proof of address et cetera but um uh let's
00:02:47
the talk on in a layman's uh terms why
00:02:50
would you submit your idea to use a phrase book
00:02:54
instead grandma just to play uh again that your
00:02:57
lunch break so the the solution of a potential solution
00:03:01
is doing h. estimation which might be a a
00:03:05
lightweight alternative to to do the k. y. c. because
00:03:10
most of these businesses they do not need to know all of your details they just need to know
00:03:16
roughly how old you are in order to decide in order to decide
00:03:21
if you are old enough to give consent if you understand the implications
00:03:25
if you can read the terms and conditions as is or if you if you
00:03:29
need to minimise it in if they need to minimise the no child readable manner
00:03:34
uh whether they need to add at the the change their advertisements so that they
00:03:39
are not going to show you of apes uh or or in the case of forces
00:03:44
u. k. if your uh uh all younger than sixteen you shouldn't even see fast with
00:03:48
advertisements so um uh what we develop an
00:03:52
privately is a is a one such method
00:03:56
that's a that's has brought use cases you would do
00:04:00
a a quick a pace down and um you can you
00:04:04
can be permitted to visit your uh favourite e. commerce
00:04:07
website to purchase debacle or when you enter a physical shop
00:04:12
um the the cashier make may decide not to ask for your id simply because the
00:04:18
hand h. decided by our algorithm says that you are well
00:04:22
above uh the legal limit a wharf or maybe or maybe you
00:04:27
can uh again well uh i just or advertisements or you owed
00:04:33
change the experience that you craft in your online game uh so
00:04:38
um we'd we'd be use um a face or voice and
00:04:42
and and sometimes uh uh one one additional uh additional profits necessary
00:04:48
we do have also some ah id there fixation uh prostheses running
00:04:52
is one i'm not another sh uh another part of of form
00:04:59
of all of our work is is making sure that a
00:05:03
week for the enhanced the previous uh preserving aspect of this whole
00:05:08
of all experience already by not submitting a a lot
00:05:12
of details we're trying to minimise the details meta data minimisation
00:05:18
right uh but we also uh instead of uh using
00:05:22
clough plays a processing most of the time we try to
00:05:25
people are solutions to the age as much as possible to
00:05:29
days um to basis more forms are capable enough to to run
00:05:33
oh i'm a and lightweight uh eh s. d. k. uh inside a nap uh to to do
00:05:40
such a h. checks um uh so the reasons why we further go down this road is uh
00:05:47
it is a sort of um uh you can you can name it they have those apparently yeah
00:05:52
uh but we but also they're also uh other concerns uh for instance um
00:05:58
that we have seen in the previous talks the the advancement of of a cyber security
00:06:05
uh on one more fucking interruptions uh secrete at the system level but somehow the weakest link still
00:06:11
remains uh to be the humans behind it uh take a look for instance the case of four obtuse
00:06:18
you might not know what because we are not in australia button uh but they were the the in a lot of uh
00:06:23
a storage of personal information for k. y. c. a lot of
00:06:28
those information more leaked uh in a data which simply because uh uh
00:06:34
some some of the engineers were compromised in the uh in the system
00:06:39
um oh also um the law doesn't like the idea that the that
00:06:44
a biometric data is uploaded unprocessed
00:06:49
uh in in servers today especially if
00:06:52
it's a if it's the biometric data of the chart because again child
00:06:56
cannot give consent or on on giving a personal data uh we will
00:07:02
uh and this is not this is not only a a european phenomena
00:07:06
we're seeing a lot of for for lawsuits settled in
00:07:09
us on the grounds of uh if he sees regulations
00:07:13
or or even noise uh the the power by
00:07:16
much information probably press a button press act uh and
00:07:21
um and we see that uh in an exemplary cases
00:07:24
on device uh data processing seems to seems to be
00:07:28
the up quite alright uh for for the regulators and look and uh and of course uh there is
00:07:34
the and there is the the reality of uh of
00:07:37
running a a company uh you you have significantly lower
00:07:42
uh uh costs uh uh if you are running most of your work loads on
00:07:47
the age instead of running a running the entire thing on the clock so um
00:07:53
i'm i'm trying to uh uh we uh let me try to paint a bit more you know
00:07:58
uh examples uh to you um uh we've we've got some interesting use case already for instance um
00:08:05
uh in in various european countries and us we have our hope that as a
00:08:10
setup in retail come first it's a physical use case uh it helps the cashiers to
00:08:17
uh have better decision taking for asking for i. d.s and
00:08:20
the greatly reducing the uh the processing time for customers uh we've
00:08:26
got um ah website access control use cases uh we've we've done
00:08:30
some of trials with that we don't know about website for instance
00:08:34
uh we are all working with um a consent management uh
00:08:38
uh platforms today to to uh to help on board hands
00:08:42
and children to to do verify patent ca can uh a
00:08:47
consent and uh and there are some also very of cutting
00:08:51
edge very interesting the uh studies i had if it succeeds uh
00:08:56
uh in the future we might have uh some ah some dates that will be only activated by
00:09:02
the voice of an about so you so even though if you don't have that uh haven't hop or
00:09:08
or even even someone buys the the weight and tries to give it the children that that those waves and
00:09:14
uh won't be activated uh so uh let me uh
00:09:19
talk about them i'm trying to the page you a
00:09:22
a beautiful picture rainbows colours et cetera but let's talk
00:09:26
about the the the challenges that we are facing we
00:09:29
are we're trying to address uh some of which uh our address some of which are uh are in progress
00:09:35
um well you know children are curious um at the phillies are also very curious and they will try to
00:09:42
uh uh to you know to circumvent your checks um
00:09:47
yeah which is the very reason why actually we live
00:09:50
shows biometric space that way because it is much easier
00:09:54
to full systems uh if you're asking just for a
00:09:57
credit card number four number um war or an email
00:10:01
address the because those can be very easily impersonate it um
00:10:08
uh but also uh in the case of by matrix a presentation attacks we've got a
00:10:13
uh we've got um people trying to show banknotes faces of people there
00:10:17
uh for tools from a another device printed a
00:10:20
printed images um believe it or not uh i
00:10:24
i had someone trying to show the face of their dog to pass the uh that's my paycheck
00:10:28
uh uh uh they're they're more advanced vectors as
00:10:32
well such as a latex masks a a manic wins
00:10:35
and of course uh today's a favour topic that we generate a. i.
00:10:39
x. are also a present we have some uh internal solutions though to them
00:10:45
uh but we are also for for more advanced levels of attacks we are conducting a
00:10:50
uh and uh for the research and development studies um one of which
00:10:55
will will uh start a very soon with a yep as well uh
00:11:01
and um the the other challenge is the is also um something that
00:11:06
we hear a lot from our clients which on which are not uh
00:11:11
uh necessarily equate it with a i but nevertheless they had a very valid point when base when they when we
00:11:16
say okay we check your age used looking at your face
00:11:20
they would say ah so your sh doing face recognition well
00:11:24
um the points i in face recognition is you can
00:11:28
identify people uh using the embedding generated and and face recognition
00:11:33
algorithms at least that's how you would oh optimise your system
00:11:36
your or you're trying to optimise the system such that that
00:11:40
uh on your today at one or two years from now on will generate more or less similar and
00:11:46
that takes right uh on oh on the contrary
00:11:50
uh what's what a phase space age estimation wore similar
00:11:54
a face analysis methods do they try to opt to mice in bathing suits such that that it
00:12:01
is the picking up age related features which which
00:12:05
work uh which were ignored in in face recognition
00:12:09
uh approaches so um uh so that's uh that's um
00:12:15
that's so you know um as long as you have
00:12:18
uh of systems that are up to my isn't improving towards the or a respective tasks
00:12:24
that should the that should become more and more apparent hopefully yeah that uh that they are
00:12:29
a separate uh uh tasks um and uh and the last one is of course i'm uh
00:12:36
not everyone who wants to put age uh
00:12:40
checks to their businesses uh because um i mean
00:12:44
we're living in twenty twenty three right uh you enter to a new web
00:12:48
second yet again you are being asked uh please keep my cookies in or a
00:12:53
browser right uh why why why why would a business try to you know uh
00:12:59
add another layer awful complexity to it and that's a very the a valid question
00:13:05
it is up to us to optimise forces are data capture uh to
00:13:10
to reduce the process in in in your seconds and and uh and also
00:13:16
they should be there should be a some carrot and not just the the stick
00:13:20
as well uh imagine having i imagine being able to uh delivers uh uh experience
00:13:27
it's a little bit based on age increasing your your your customers engagement with your
00:13:33
part for optimising or maybe maybe had really news from are from ethically source the way
00:13:40
uh of of uh eh you know that age
00:13:43
ranges of your of your clients so um uh it
00:13:49
well last last but not least uh uh uh uh i mean
00:13:53
uh i can always talk about how amazing my uh system is
00:13:58
uh but um you don't have to ask me you can ask
00:14:02
uh you can ask uh and some a certification body switch yeah
00:14:07
which is uh is up and coming a new standards are being a defined internationally
00:14:13
uh there are some of course there are some others uh nations
00:14:16
that are uh more advanced than than others once is u. k. germany
00:14:20
uh they are uh uh they have they are um the third
00:14:24
party uh what it's for a a a just a mission qualities um
00:14:29
a quick note uh not uh it seems like a human's we
00:14:33
have four apparent age estimation we haven't need major or for seven years
00:14:39
uh the the current uh system we have as the mean age or or one point three years for
00:14:45
uh uh for the age range uh
00:14:48
between seventeen to twenty and also oh we
00:14:52
the we take a pretty see a seriously we a code of conduct what it's
00:14:57
and and get the appropriate certifications work consisted if you're a while but that's uh
Conference Program
(Keynote Talk) Privacy-Preserving Machine Learning : theoretical and practical considerations
Prof. Slava Voloshynovskiy, University of Geneva, professor with the Department of Computer Science and head of the Stochastic Information Processing group
Oct. 11, 2023 · 7:55 a.m.
2509 views
5 minutes Q&A - Privacy-Preserving Machine Learning : theoretical and practical considerations
Prof. Slava Voloshynovskiy, University of Geneva, professor with the Department of Computer Science and head of the Stochastic Information Processing group
Oct. 11, 2023 · 8:40 a.m.
Enabling Digital Sovereignity With ML
Vladimir Vujovic, Senior Digital Innovation Manager, SICPA
Oct. 11, 2023 · 8:44 a.m.
5 minutes Q&A - Enabling Digital Sovereignity With ML
Vladimir Vujovic, Senior Digital Innovation Manager, SICPA
Oct. 11, 2023 · 8:58 a.m.
Privacy-Enhanced Computation in the Age of AI
Dr. Dimitar Jechev, Co-founder and CTO of Inpher
Oct. 11, 2023 · 9:01 a.m.
138 views
5 minutes Q&A - Privacy-Enhanced Computation in the Age of AI
Dr. Dimitar Jechev, Co-founder and CTO of Inpher
Oct. 11, 2023 · 9:20 a.m.
Privacy by Design Age Verification & Online Child Safety
Dr. Onur Yürüten, Head of Age Assurance Solutions and Senior ML Engineer in Privately
Oct. 11, 2023 · 9:26 a.m.
5 minutes Q&A - Privacy by Design Age Verification & Online Child Safety
Dr. Onur Yürüten, Head of Age Assurance Solutions and Senior ML Engineer in Privately
Oct. 11, 2023 · 9:41 a.m.
(Keynote Talk) Biometrics in the era of AI: From utopia to dystopia?
Dr. Catherine Jasserand, KU Leuven (Belgium), Marie Skłodowska-Curie fellow at Biometric Law Lab
Oct. 11, 2023 · 11:06 a.m.
5 minutes Q&A - Biometrics in the era of AI: From utopia to dystopia?
Dr. Catherine Jasserand, KU Leuven (Belgium), Marie Skłodowska-Curie fellow at Biometric Law Lab
Oct. 11, 2023 · 11:42 a.m.
AI and Privacy
Alexandre Jotterand, CIPP/E, CIPM, attorney-at-law, partner at id est avocats
Oct. 11, 2023 · 11:48 a.m.
5 minutes Q&A - AI and Privacy
Alexandre Jotterand, CIPP/E, CIPM, attorney-at-law, partner at id est avocats
Oct. 11, 2023 · 12:06 p.m.
Preliminary Pperspectives on the Ethical Implications of GenAI
Julien Pache, A Partner at Ethix and Venture Partner at Verve Ventures
Oct. 11, 2023 · 12:12 p.m.
5 minutes Q&A - Preliminary Pperspectives on the Ethical Implications of GenAI
Julien Pache, A Partner at Ethix and Venture Partner at Verve Ventures
Oct. 11, 2023 · 12:30 p.m.
AI & Media: Can You Still Trust Information
Mounir Krichane, Director of the EPFL Media Center
Oct. 11, 2023 · 12:32 p.m.
5 minutes Q&A - AI & Media: Can You Still Trust Information
Mounir Krichane, Director of the EPFL Media Center
Oct. 11, 2023 · 12:54 p.m.
(Keynote Talk) Unlocking the Power of Artificial Intelligence for Precision Medicine with Privacy-Enhancing Technologies
Prof. Jean Louis Raisaro, CHUV-UNIL, assistant professor of Biomedical Informatics and Data Science at the Faculty of Biology and Medicine and the head of the Clinical Data Science Group at the Biomedical Data Science Center
Oct. 11, 2023 · 1:22 p.m.
5 minutes Q&A - Unlocking the Power of Artificial Intelligence for Precision Medicine with Privacy-Enhancing Technologies
Prof. Jean Louis Raisaro, CHUV-UNIL, assistant professor of Biomedical Informatics and Data Science at the Faculty of Biology and Medicine and the head of the Clinical Data Science Group at the Biomedical Data Science Center
Oct. 11, 2023 · 1:50 p.m.
Genomics, AI and Privacy
Julien Duc, Co-Founder and Co-CEO of Nexco Analytics
Oct. 11, 2023 · 2:01 p.m.
5 minutes Q&A - Genomics, AI and Privacy
Julien Duc, Co-Founder and Co-CEO of Nexco Analytics
Oct. 11, 2023 · 2:18 p.m.
How trust & transparency lead the success of an Idiap student's Master's project in fraud detection
Raphaël Lüthi, Machine Learning and AI Lead at Groupe Mutuel
Oct. 11, 2023 · 2:22 p.m.
5 minutes Q&A - How trust & transparency lead the success of an Idiap student's Master's project in fraud detection
Raphaël Lüthi, Machine Learning and AI Lead at Groupe Mutuel
Oct. 11, 2023 · 2:38 p.m.
Recommended talks
L.I.F.E. - express your truth through your accurate data
Dario Ossola, PhD, Algorithm R&D Coordinator, L.I.F.E. Corporation
June 3, 2016 · 2:52 p.m.
105 views