Player is loading...
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
Wednesday, Oct. 11, 2023 · 8:40 a.m. · 03m 26s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:04
the all the problems related to the person the machine learning but what i'm curious is the anyone
00:00:10
practically speaking like all these got ice like who did mine i don't openly i
00:00:15
phrase book anyone of them are doing anything actually to preserve privacy in to do
00:00:20
these frameworks you or they just kind of talk on the web page and that's
00:00:24
as much of the two that's what i wonder because it feels like the second
00:00:29
okay well thank you very much for an answer your questions in fact uh
00:00:35
it's a very important question and the very important trend trend nowadays and unfortunately
00:00:40
observes it more and more these people not like five ten years ago where
00:00:44
they publish and they do what they're doing they become becoming more and more cost
00:00:49
and it will be very difficult to say what will happen next years and how it is done actually
00:00:54
a lot of hampers especially from will was about the federated learning thinking
00:00:58
and they wanted to introduce it a different devices so it is done
00:01:02
as far as i know some groups uh in this companies are
00:01:05
working on homework encryption and whether it's possible to do machine learning based
00:01:08
on the on working encrypted data but okay i'm not an expert and just to say about all the issues related to complexity et cetera
00:01:16
and general trend uh uh what this right now of course for them papers and security uh l.
00:01:22
e. generated data and there is a strong reason for that if i may just answer very quickly
00:01:27
as you know many many companies especially opening i stable and and
00:01:32
the data for these guys produce a lot of lot of data
00:01:34
from one side it is cute very cool floss as as a
00:01:38
service but the same time it's created very very it's used engines
00:01:42
be me personally i believe that if you create more data synthetic data in
00:01:46
eighty engage more and more data into the training of the systems but it costs
00:01:50
apparently it's not the keys especially for the noisy diffusion models that is the right now hopes and
00:01:56
he actually including syntactic base that into the training process and with the old data the observer and
00:02:03
distribution shaft means that system if you do first uh part include the thing that you generated data
00:02:09
second system deteriorating performance it can't explain from different
00:02:13
perspective especially of the diffusion of the manifold right by
00:02:16
by by by by by adding there isn't it to do for example the do not represent real well
00:02:21
and the general trend would be observe right now is the uh in this company is is how to uh
00:02:27
detect how to prevent inclusion obvious synthetic you generated data into their models
00:02:32
and that will this solution what is right now we turn from twenty is set years old
00:02:37
technology is what the also contributed before it is digital watermark into what they try to do
00:02:42
they try to add invisible marks who would be generated content to distinguish use more ideas generated content from this
00:02:49
in terms of for applying for could to graphic principle personally i'm not aware
00:02:54
about that a lot of the airports are down along 'cause specially a grappling yeah
00:02:59
i'm from his early days appeased in now the senior researcher so basically a lot of our present on how to fight was ever
00:03:06
settle example so security but privacy in this sense complete not addressed
00:03:11
up my best knowledge okay i didn't know what's exactly going to school
00:03:15
thank you very much i think that's all the questions we have time for so if you're
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.
2514 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.
139 views
