Player is loading...

Embed

Embed code

Transcriptions

Note: this content has been automatically generated.
00:00:00
yeah thank you for a while invited me under it yeah i understand the last year before the the party tonight
00:00:09
it won't be too difficult and then maybe can use you can help you to
00:00:13
put your project you know all on the moon because it's the object up at least we
00:00:19
uh so i i'm the only nemesis um based engine you i'm
00:00:22
in charge of uh what some business development initiatives a local
00:00:27
so i'm i'm sure you'll on her but uh yeah but
00:00:31
what's an and uh and i didn't couldn't to technology
00:00:34
so i want to you know the the marketing stuff and so on this is not the objective and
00:00:39
um we are potholes you know every week analysts around the world
00:00:44
part of a microphone gentle to support uh such initiative
00:00:49
but each is unique which is different uh and this one's quite in it because it
00:00:54
three week long so you would have time you know to delivered over in the p. and
00:00:59
which is quite important i believe on top of any business case initiative and all activities
00:01:06
uh and the idea is to dan for me in nineteen thirteen it is that too
00:01:12
to give you the chance or to understand that what's on is not only for large companies and
00:01:16
it is maybe the problem was what some brand that's you know it's all over the us
00:01:20
a lot of marketing and so on and and people think that this is only for for much continue sentences absolutely
00:01:27
wrong
00:01:28
okay uh so i said we are supporting act upon last appointing
00:01:33
at a initiative that in the pits and we will we will do a this year uh
00:01:37
with with which we are here so to provide the technology form for those teams
00:01:42
uh and we have also and drop one of programs you know to to help you with that knows
00:01:47
yourself with marketing you know with your initiative later on what seems important to know that ah okay
00:01:56
so i was talking about so are we try not to explain you know what you can do
00:02:00
a starting to all white and today and give you some some height some things here
00:02:07
i was talking about going to the moon so just to provide you a an idea of
00:02:12
a very recent project i don't know if you know this with the rubber name simon
00:02:19
it's currently you know or in the i. s. s. you know the space
00:02:23
you know supporting uh and the astronauts there and it has been
00:02:29
developed by just for an all the all the only the machine and
00:02:33
uh the the and the f. to ditch all the the hardware
00:02:37
and we developed or the uh i would say cognitive capabilities there what's funny is that
00:02:44
every direction the robot as reason after not interaction is coming back to hurt us
00:02:49
to the cloud enough and then the answers come back so any ads
00:02:54
with several uh capabilities first it is able to uh to recognise uh uh
00:02:59
there's several astronauts you know which is much recognition of voice recognition
00:03:04
and it's also is able to answer as questions but also maybe
00:03:08
complex question bands of all the protocols they asked to follow
00:03:13
okay uh doing if they they they have been a lot of experience that a lot of maintenance that
00:03:17
any supporting then gonna by explaining what they have to do and how they have to do that
00:03:23
what is important here is
00:03:27
the use case you know yeah this is not replacing human you can in the
00:03:31
world would be reading or listening and that that's one a i is mentally
00:03:36
augmenting human supporting human okay and this is i think when you are thinking what joe use case
00:03:43
really important you know to think about it
00:03:48
just if you want about watson
00:03:51
very whole project you've certainly heard of the about this one you know whether it
00:03:56
is seven years ago uh a machine winning and joe probably in the us
00:04:02
you know and serene understanding question answering question quickly
00:04:06
and also the marketing yeah some confidence e. d. you need you need sensor which is you know that
00:04:12
we are not anymore in the dim at deterministic world but more in the statistic for which you know what with the act
00:04:18
so this ways uh this was a challenge meaning that this
00:04:21
is to demonstrate the capabilities of of the new technology
00:04:26
i'm the same term of architecture this was definitely not for business you
00:04:32
know it was you know a very monolithic stuff you know
00:04:36
really big machine because it was not connected to internet whatever that everything was what he and yet when survey we quickly
00:04:43
and we worked a lot and we develop you know a lot of capabilities there
00:04:48
to be a a bad ah no available to everybody you know
00:04:53
and this is what uh you should you you could do and we demonstrated a few weeks ago this project debater so this
00:05:00
is a machine able to date on this but specific topic with
00:05:05
unit so we'll see more about also you know understanding
00:05:09
uh a long way that's one that also generating you know language
00:05:14
unable to to have some pretty interesting logic of the conceptual oven
00:05:21
this is still a challenge meaning that this is not something that uh we should they deliver
00:05:26
you know on the market that it would come with me that'd show the the direction okay
00:05:32
and uh that this is like when we see a a a girl showing steve for
00:05:38
stuff and so one this is all with the same you know this is sh
00:05:40
they we show something and then you know that it's going to be available for for you
00:05:47
uh this is something i'm done i i explain
00:05:52
uh
00:05:54
quickly uh the main focus of of what's an initially is not you know
00:05:59
pew national owning it leverage machine learning and provide capabilities for business
00:06:06
meaning that you don't have to code or nothing you just use service
00:06:10
either a. p. i. to analyse a video to understand text
00:06:14
to understanding uh ought to do a a chance what was talking about chad button someone so you
00:06:20
you just and ends those capabilities for with your expertise and with your data
00:06:26
as in all your your your projects which are they i base
00:06:31
you need expertise but very important and you need data and if you have only one of us over this
00:06:37
it won't work you know or if you have good big that but well wrong expertise we not trained the data efficiently
00:06:45
so it's quite important that and and
00:06:48
and and that's an capabilities helps you to go fast you know living in that
00:06:52
not focusing on a few you know a a machine not may not
00:06:58
be planning a capabilities okay you can also do to do that
00:07:02
uh with with a platform and i think it can be interesting in such in seven project for you
00:07:07
but you can see you can choose those capabilities uh to go fast
00:07:14
so just a few project or use cases we do to give you some ideas
00:07:20
a million that
00:07:23
health industry is it was the first one after the job and
00:07:28
uh maybe call all doctor oncologist solar game at the t.
00:07:33
v. and understand this is exactly what do what they need they have so much data and many we should this is
00:07:40
uh and such kind of of studies over of a military because or whatever are are a lot of politicians
00:07:48
and they can't follow the wreaths of publications you know and then you turn the up to date
00:07:53
when you have uh when when you have a patient you need to provide the best treatment
00:07:59
based on the like the latest literature okay and these kind of to be providing that in that
00:08:05
you provide all information about your patient can be instructor instructor they in for
00:08:09
a format and then the system would be able to provide uh
00:08:16
a different uh in medication or treatment plant
00:08:21
in always confidence calling physical or okay
00:08:24
uh based on the seated at you and if and it is updated every
00:08:28
month or every week and this is this is the value value
00:08:33
here think oh always about the value value is not to
00:08:36
be the the the the final um treatment this is
00:08:43
the and the the value wheezing the confidence and in if you didn't it
00:08:49
because if it's if you show so that to you know a very uh from third world uh on colleges that with
00:08:56
a lot of experience he knows the answer okay because he's aware of a lot of things and so on
00:09:04
so what what is interesting is that you would see that they always a document because he published maybe
00:09:10
two weeks ago with this is the value you know not to to have a wrong wrong stuff
00:09:17
so this was the first project and then we split to the the the what's and capabilities in
00:09:23
in three category i was the first one is language so we for the thigh pull here
00:09:29
uh so the the language everything about language understanding and
00:09:33
coming up with a uh also personality a test
00:09:37
language understanding conversation conversation or the capabilities uh
00:09:43
turn allies are everything about language okay
00:09:46
then you have what we call signal signal that would voice in h. videos
00:09:52
uh and sounds uh we did a lot of work with some for uh i don't know uh
00:09:58
uh make me machine quality at best you know based only on the some okay it works for what
00:10:05
and then we put each machine running and keep running as a service on the crowd and i would just
00:10:11
explain that later i think it could be quite useful for you if you want to go fast
00:10:19
so you may age but okay just
00:10:22
just to give you
00:10:25
i think this last with this one is that it's getting a good so if the group picture
00:10:31
you know but is that power from here and it's a bit but mainly it's it's interesting we were talking about uh
00:10:37
you know user use case and so when they decided to implement a i not for the
00:10:44
customer but for the agents for the employees you know to better work and today after
00:10:52
uh it's alive since nearly one year not not get nine once they do would but rising with their
00:10:59
own employees to explain that a i and what's an expert is helping them to better work
00:11:06
okay and that's uh you know in france it's quite a great success to
00:11:09
do so because employee and employer is is relationship is not very easy
00:11:16
uh but and and the the result what quite good so this is not complain
00:11:21
but
00:11:23
what i want to to to show you when interest for example is that very small companies and started by
00:11:29
using what's and this one is a ski resorts fifty kilometres from here maybe less you know the implementing
00:11:37
an assistant to answer a lot of question about the ski resort in winter and summer
00:11:44
and if it works on on face the messenger and also here but to use
00:11:48
uh what's than a agent to answer question whether question or whatever questions okay
00:11:56
and this is what small company and uh this is five of six people and not a lot of money this one is funny
00:12:04
you much recognition for circulate a people you know you you know new but there was kind
00:12:11
of chuckle it you you never know what's inside so you just take a picture
00:12:17
so you recognise the that circulate and then you know what's the and and and you can you can maybe
00:12:22
but i just chuckle ain't or whatever other so it's a new way to engage you know your your customer very funny thing
00:12:30
not so easy you know you know we do some image recognition for very
00:12:34
specific things in maybe car but for truck related not so easy
00:12:39
because sometimes i exactly the same
00:12:42
okay and this one is uh it was just it was always to the stuff in the u.
00:12:47
s. ros and they provide a i based uh advice is legal advice is so you can
00:12:55
send your case or describe your case maybe sometimes in chat but
00:13:00
way okay and then it will dig in a lot of you know uh
00:13:05
a lot of resources you know been expect resources
00:13:09
regulations uh exist in places and so
00:13:13
want to provide advice is for legal people you know to be deaf case
00:13:20
okay what interesting especially in the us or with important to think about the market because in in the market
00:13:28
in this market you know they're they're they're the they're always based their
00:13:33
their cases on on previous cases okay it's really really important
00:13:39
would be any different in in europe uh it was the the the gift that it different from how they do that
00:13:46
but the average approximately yeah they are fifteen uh
00:13:51
services base basic services that are available
00:13:55
when you would you can go now you know and use that for free
00:14:02
to try it and to implement them to use it and and and quite easy uh
00:14:09
for different use 'cause we were talking about chad but under for for the for the hotel okay it you have
00:14:15
and that is that you you can be your understanding bands in the
00:14:20
promenade diodes uh finding information is someone that you can and aunts
00:14:25
we still rising but also phone under and language understanding is you have to do you mean a lot of information
00:14:30
and extract the relevant information provided uh to to the
00:14:34
assistant okay you just combine those services very easy
00:14:40
and you it's always based on training when you train your data you need data okay
00:14:47
and it's not coding you know and then you integrated them so when you want for example
00:14:52
trying to go fast you know uh for the the the just the result of before
00:14:58
i did the first you know approval valuing three days you know wheeze integration
00:15:03
of or a a face to make a quilt or a google on
00:15:08
and so on and the it was you could get a fork was to integrate with this that
00:15:13
not to train the system because i had the data i get the the train it so it's quite quite quite easy
00:15:22
got again but some uh it was not to the visual recognition
00:15:26
so you can and ends existing but it
00:15:29
okay good training data
00:15:32
and then i that's i think which is important because i think you are or you have
00:15:36
some specific uh a strict or instructor information and you want to be your own models and you want to deploy them
00:15:44
and today uh yes you can you you have tens of go you are a lot of things you have to write your script and so on
00:15:50
so we
00:15:52
we provide some tools where you can
00:15:55
i'm with the design young people and uh i'll go with
00:16:01
white you know easy and and manage your method parameters
00:16:05
okay
00:16:07
and then you can download the code if you want to run it on a
00:16:11
on your local machine and so on but you can also the probably
00:16:15
on what some machine on o. u. penn train one euro on the low moderate
00:16:19
on what some machine on it and then you can deploy it automatically
00:16:23
meaning that it will be available for your for your on the court for for it for god and for your business
00:16:30
so what you own only need you know your data and and you can and and that's what that's that's really
00:16:36
you know if you know a machine learning and be planning on going and you know how to combine them you
00:16:41
need for that you know and you need you need to have the skills anyway but you can build that
00:16:48
run it transmits fifty and then deploy you know without any infrastructure management and i think when you have one week two weeks three weeks
00:16:56
depends where you wonder project it would help you to go faster you know and to focus on
00:17:02
your business going to space you know and validation what you can do with your bit
00:17:10
we can arrange if it's if it's something that uh some scenes
00:17:15
wants someone coming explaining you know how to do so uh
00:17:21
for this one off or another one uh another a. p. i. during the three weeks so afford we will do that uh
00:17:28
uh with pleasure to find the the time but really if this is something that you want
00:17:32
to do uh we can support uh if it helps you to go faster and
00:17:39
so
00:17:41
i'm i'm finished
00:17:43
so just tripping that what's on for you that you can use it first one is
00:17:48
the first uh one of the possible often you know go by the end
00:17:52
uh if if it was question that would be also here tonight if uh if it had
00:17:58
and then we can provide you know its report if you want to
00:18:02
use those kind of technology for for for doing this great

Share this talk: 


Conference program

Watson for You
Jérôme de Nomazy, GM & IBM Watson
28 Aug. 2018 · 4:38 p.m.

Recommended talks

Presentation of the «Natural Language Understanding» research group
HENDERSON, James, Idiap Senior Researcher
29 Aug. 2018 · 2:03 p.m.
Non-Stationary Signal Processing and its Application in Speech Recognition
Zoltán Tüske, RWTH Aachen University
7 Sept. 2012 · 2:29 p.m.