Player is loading...

Embed

Copy embed code

Transcriptions

Note: this content has been automatically generated.
00:00:00
okay everybody so i just said so my name is uh where you work on the part for the manager of aussie two ways ah
00:00:08
so i don't know if you know what it means it's all about covenant control intelligence civilians not gonna sounds
00:00:14
um so basically i'm in charge of the product development of strategy and
00:00:18
the allocated digits to just reported this or that opens hydrogen
00:00:24
somebody happy to be in front of you are today uh uh to give you this final speech um which is
00:00:29
where we try to show you how we're currently integrated artificial intelligence we feel portfolio
00:00:35
i'm not and that's trying to just like actually the compete with that that technology so clothing label technologies
00:00:42
their artificial intelligence uh because there is a more more needs to for
00:00:46
massive storage message processing and this is what whatever show you today
00:00:51
so for those who don't know us us um means command so really giving the guidance the
00:00:57
joe for a t. v. to accomplish a mission objective uh you have control it's it's
00:01:02
all about controlling the process activities when when you go on on the flight operations
00:01:07
uh intelligence which is about gathering information and extracting value for this big amount of of data
00:01:14
and so it's not gonna sounds in more but uh monitoring of a specific area of of interest so this is
00:01:21
uh the the the part where products but i'm a i'm managing and you
00:01:24
will see this plantation as well but those clothes and other technologies
00:01:28
the um dries as well of business transformation for us usually we're really but from some trick
00:01:34
more more plot eccentric and no we're going to be a more service oriented so we have
00:01:39
more more services coming from from the from the pop for you from from the business
00:01:45
so the presentation would be expected to free pass first i would give you a shot of of you but i've us
00:01:50
i know sometimes not easy from outside even even from inside is not easy to understand sometimes that this organisation
00:01:56
so one sided but it never show your love it the the particular what they are what the what is the aim of this so products
00:02:02
and then when i will explain you as well the customers challenges because really
00:02:07
does not take no push but we're doing no it's really the customers
00:02:10
i need to have those technologies to be able to um to two hundred just big amount of of that though
00:02:17
um but before we are going for granted his presentation that we want to thank you possible off
00:02:21
a tall invited me uh it's a great pleasure you know egypt is really well organised uh
00:02:26
a research institute in the artificial intelligence we have already some collaboration and i'm more than
00:02:31
happy to to maintain them and even to reinforces them in the in the future
00:02:39
so it's it's a big organisation actually it's quite it's quite easy but interestingly this uh organisations but you see
00:02:47
that intelligence which is at the bottom part uh is really easy for him for for display community
00:02:53
all around this a bit that like technologies is cheesy of uh utmost importance
00:02:58
uh because you have a problem nice divisions what i'm units that more but from driven
00:03:04
and we had the one uh taking the data out of it and then providing some some services
00:03:09
so you have a best commercial aircraft which is the most famous of the vision heaviest
00:03:14
helicopters around the c. v. than military helicopters and then yeah best differences but it's
00:03:22
when i was the first page of then for a palm lines media craft
00:03:27
so or want to talk our craft of the the fighter aircraft
00:03:30
um the the topic off like a a fondue them uh you have spaces
00:03:35
stems with the launches business and the satellites business sort of observation
00:03:40
a telecommunications satellites overall missions specific uh science uh uh satellites
00:03:47
and when we are within the c. i. s. o. communications intelligence and security
00:03:52
so you have to secure communications or about the communication between those platforms
00:03:57
uh to have the the the correct network the quite but we've to to
00:03:59
send around the data you have cyber security which is about protecting
00:04:05
those platforms again cyber attacks but does well securing the the the um the exchange
00:04:10
of a possible for information so we're talking there about the secure connectivity
00:04:16
and we have intelligence so which is a than the problem unit where you have all this mission system some this
00:04:22
o. c. to i support x. to hop no it's okay that sounds a bit clearer for you all
00:04:32
so we have um in intelligence so when it's really not
00:04:36
long experience sees a one c. two s. r.
00:04:40
i don't want to give you a really a lot of details about the products but uh i wanted
00:04:44
to illustrate admitted this products from the two cycles the first one is a common second so really
00:04:51
getting off for t. giving direction to achieve a specific admission objective
00:04:56
we have a a a suite of c. to our products really from the high
00:05:00
liverpool static levels down to the subject to this month and so those
00:05:05
well it's all about uh and situation assessment planning of the mission planning
00:05:09
of the task of and controlling the execution of that those activities
00:05:14
and you have as well that intuition cyclists huh because to take decision on this mission you need to know uh
00:05:22
exactly what what is the situation and for that we have a lot of products around this intelligence second
00:05:30
um the internal cycles restart from the um the beside us it and say
00:05:34
okay i really need this information on the specific area on specific target
00:05:39
and when it initiates the digits i couldn't with direction uh looking at the
00:05:44
available i sauce also some undefined you gets the basal souls to
00:05:47
look for for this information could be u. a. v. scooby satellites could
00:05:52
be an ear isosceles holes then is about collecting the data
00:05:57
and processing the data so really exactly value from from this data
00:06:01
i we have read different products which are hunters this type of also sees
00:06:06
i'm starting with a you mean to so human intelligence in the intelligence which is coming from from
00:06:13
it's a a bit from from a person uh we have some product around the open source intelligence so really
00:06:20
trying to extract the value to to find to select acquire the recording value out of the open source
00:06:26
um we have long expenses of course in the you much intelligence because
00:06:30
it's already obvious has his own um costello concession of satellites
00:06:34
so we have a lot of our products which uh try to to find information into this uh images
00:06:41
and then we have some other products around sixteen so signal intelligence
00:06:46
interception of of of seniors could be communication between two people and so comment or could be a
00:06:52
a on either the and the detection of uh electronic uh signals which but took also didn't
00:07:00
so this is basically the the overview of this uh of the support for
00:07:04
him and the what i wanted to to show you as well used
00:07:10
uh uh it's what what we're doing this information because it's good to extract
00:07:14
and to uh to do some report to some analysis on this complete
00:07:18
a silo stuff by uh information but then we're doing some what
00:07:23
we called the knowledge of the base where we can
00:07:26
where we can put all this information into knowledge database where we're able to do some question sheep
00:07:31
the management and we are trying to them as well as some some some decision of this information
00:07:37
some uh cross intelligence uh uh yeah of course
00:07:42
it also says uh analysis of coronation
00:07:50
on the other side where the ripping as well as some some uh a real time service applications could be in this uh on the i. works
00:07:57
or on on the some wheels of a specific or country so hey it's all
00:08:00
about uh having some products which gets its provision information from real time application
00:08:07
gathering all the services of that are right out gently i ideas be
00:08:12
i yes from from from the shapes and then having an automated processing of this virus forms of a real time data
00:08:18
and the the and the critical task of as we to do some decisions on the jesus of vision
00:08:24
be able to detect okay this is an object which has been detected
00:08:27
by two difference also but it's actually the the same object
00:08:31
and you do the search on that is is some attributes and the
00:08:34
the a misery to elaborate on a shell situational awareness picture
00:08:38
so far i often around for my time with different uh products we we should was about
00:08:44
to um to elaborate on to to display this uh this kind of uh of sure
00:08:50
so this is the the current portfolio and i want to know how to show you what what of the the challenges from from our customers
00:08:57
from our end users which are basically something else and that is the digits and i just um what what other there there there needs
00:09:05
and and actually what we are saying in this new world of intelligence but we we went from that
00:09:11
let's say a complicated environment to a complex one i'm meaning but but
00:09:17
the the targets the targets yet sure the target characteristics a challenge
00:09:23
uh traditionally that i guess we're really large big um the formations
00:09:28
uh there were a slow moving uh they had some drivers yeah just so you could really
00:09:33
identify detect those uh those for its uh for for and detection on identification uh equipment
00:09:41
and we knew the locations that was really a state or region based effect
00:09:47
and as well the web predictable so we knew that the train we knew how they behave we
00:09:52
knew how they will react to any of all of a process of uh of actions
00:09:57
and no uh we have some basic it was a lot of parameters but with a
00:10:01
single human brain does quite easy or i believe it's some some simulation support
00:10:06
it was easy to um to predict what will happen in the future on how what is the the best um
00:10:12
way of uh of acting uh to to to to win a or to to achieve a mission objective
00:10:18
um no of course uh since the beginning of twenty first century this
00:10:23
fred such a lot we don't have this large i'm informations
00:10:27
really mainly and i was never really really small but the networks
00:10:31
could be in the videos i'm back constantly changing something up
00:10:36
really i don't think then said even the way we are we are that that and uh and then
00:10:41
and that you really critically a diffuse meaning what when when you go
00:10:46
on up up and you might be for the french army
00:10:49
unlike its population like it's really room for it so really that bad diffuse it's quite difficult to
00:10:54
uh to identify how to detect uh this uh this type of of of for it
00:10:59
and of course as well about some ambiguous attacked it's uh it's not affecting which is written in in the doctrine
00:11:06
so all these parameters are quite complicated it's not a possibly more to put
00:11:11
that into a question to have a clear view what what can happen
00:11:14
there there will be some surprises so this way resting it's a it's a complex environment and
00:11:20
it's one one five so it's really the environment your fact is
00:11:24
that we have really that explosion on the flight operations so
00:11:30
really yet to the example of to uh address platforms uh as an as an example but
00:11:35
um basically uh uh she is we we will launch the the p. added new satellite
00:11:40
f. observation satellites it so much it okay solutions like forty something that uh
00:11:45
resolutions you have no twice a day or visit capacity so you'd come over the same
00:11:51
place on the of twice a day and so so you can imagine but it's yeah
00:11:57
does does images are bigger and bigger and you can you have more more information
00:12:02
and we intelligence i'm charge of the of the one segment
00:12:06
to acquire this information these images haven't to process them
00:12:10
and this is where we we make that that technology so cloud in about an uh
00:12:14
technologies and in order to um to do massive storage and a massive processing
00:12:21
another example is the uh if there's if you could just uh a bit the the record
00:12:26
a couple of days ago so you who during twenty and almost twenty six days
00:12:31
so long given any of our craft powered by by the sun and really here it's and it's in use also
00:12:38
between satellite and u. a. v. staying at the same position and having really some some really precise sows sows
00:12:44
a lack of observation was also also images and videos uh we're
00:12:49
thinking about uh overpaid also transcontinental it into some souls
00:12:53
and uh he it you can imagine as well but the time just in the same region is much
00:12:57
more logo so much more more that that uh you can sense you can see more clearly
00:13:03
and and for that we we needed those uh but that that technologies
00:13:09
uh_huh so the oppression it's basically a uh what what the customers saying was what we want
00:13:16
to tune always to discover the the unknown so really to have some tentative analysis
00:13:21
to what because the activity base intelligence so we don't have this uni are uh
00:13:25
it didn't cycle say okay i want to see this for it this region
00:13:30
at at this time so it's not a linear with stuff by different uh
00:13:34
uh intelligence uh a sense working working on specific uh source is
00:13:40
what you want to go with this to to an integrated intelligence really i'm putting all that
00:13:45
together sharing that that that that which is sometimes complicate either especially for for military intelligence
00:13:52
what we need is what is automatic processing of massive data uh they don't have enough and that is to do the job um they were not have
00:13:59
more analyst actually so they really want to have some some some specific algorithms
00:14:04
to to perform a automatic processing through a scale out a infrastructure
00:14:11
and so the the games we to detect weeks yells so it's not really funny most three weeks yes an
00:14:17
abnormal behaviours sending some other some trends that analysts can focus really on on uh added value of tasks
00:14:25
the challenge behind of assistant to to break the the silos from which is also fusion
00:14:31
uh as i say for intelligence for some reason it is sometimes difficult to
00:14:36
to to accept or it's it's really a computer can show to to protect the data
00:14:40
i'm breaking the silos and putting all the data together in a in a big digit
00:14:44
platform you beat 'em right back sometime it's uh it is some some change management
00:14:50
give a challenge the data quality under that adversity something sure but the data
00:14:55
what the what the data but we're putting into the artificial intelligence the
00:14:59
training of of these algorithms ah correct because if you put some garbage in
00:15:04
it you will be the year to get summer get much out
00:15:07
it's a a big concern for this or uh for customers as well as the data sovereignty into the data of the of
00:15:13
the ability so basically you don't want to put onto that thing a good cloud or in the c. v. o'clock platform
00:15:20
for them it's real important to know okay where where my data stored in which a location where where is that doesn't though
00:15:27
so it's something but but we need so we're talking more but this
00:15:31
private load let's say a phone for storage of this data
00:15:36
another challenge is as well to secure connectivity uh with the the right to been we've on that flight of operations
00:15:43
uh because basically it's really is some constraint the networks there it's quite difficult to uh to exchange
00:15:47
data uh we we're going to keep going to that but it's uh it's still a challenge
00:15:53
the channel as well actually the two last bullet points are quite telling together is the position of this uh of reasons
00:15:59
uh when you say okay uh eighty percent of the time you were detector ship the
00:16:04
twenty percent or not uh it's not acceptable for customers but really want to have
00:16:08
precise algorithms uh and they really want to to to to train so
00:16:13
to to work to achieve the the the perfect a decision right
00:16:17
and so that s. points man machine to mean hold them and how the analyst
00:16:22
contrast information information which is coming out of this uh affectionate digits of reasons
00:16:27
all about this and my machine to mean we're doing a lot of of studies around around this
00:16:31
point and yeah because they have to actually to to trust what the the machine it's in
00:16:39
what about cheating isn't so since we uh we use cut technologies it's
00:16:43
we use of this actually then it's this a micro services
00:16:46
uh to to do massive uh processing to do this scale out uh
00:16:51
we we are going to to new business uh more than this
00:16:54
so the say no we're thinking about or we're already delivering some some
00:16:58
services uh on on the clothes so really i mean assessment
00:17:03
and and of all purchases as well was talking about that during the the break is what we are going elsewhere no partnership ecosystem
00:17:09
for innovation because we're allies but actually alone to be more more difficult to achieve that even got the idea of a switch
00:17:17
great engine competencies it's going so fast that we we need to be open for
00:17:21
for innovation and there's more more inserted around town from this point uh_huh
00:17:28
he can example because there when we say okay let's put the the that that together in the in the big just about phone
00:17:35
um yeah sometimes our customers element is it really possible
00:17:39
and we had the same appointment to the before we've uh we've skyway so we we launched isa
00:17:44
skyward digital phones okay let's put all these data that are coming from the engineering design
00:17:50
uh from the oppression um the problem though version of the of the of the flight
00:17:55
as with the data from the suppliers let's put all that together and on out of it to be able
00:18:01
to to um to uh to be more efficient and to create shock value for each of the stakeholders
00:18:07
and when we talk about shared the bedrooms for example for as opposed to have better product
00:18:13
is for the operators to have for example productive mountains so really we we lies the the
00:18:19
the the airlines have to put all does that um that to agree to give
00:18:22
up to put all does that put together if it's only one airline walking to not
00:18:26
be good enough so we are creating shock value for for every second those
00:18:34
and actually so this is what we're currently you're doing within a year to just program you
00:18:39
need a it sort of did did the same uh approach to say okay let's put
00:18:45
or does that that together coming from you know that you meet the you mean to c. d. into posing and and say right into
00:18:51
having a massive data collection doing some formatting uh all about this uh u. t.
00:18:57
and then having some enhancer melted some of each of the so sauces
00:19:03
to create to to put this information to to packet and to put it in the knowledge database
00:19:08
and then out of that doing fusion correlation to be able to to raise some some uh a trend some islands
00:19:15
uh some uh let's depending on the abnormal behaviours so but then we're able to to
00:19:20
come to this code of predictive analysis so it's all around me just sauces
00:19:25
a massive volumes or automatic processing which is really important knowledge capitalisation
00:19:31
to be able to do this correlation because they're really them
00:19:35
we want to or which information coming from open source which information coming from
00:19:39
uh for example you mentioned the regions um so the infusion unproductive analysis
00:19:49
yeah but that's where some example of a of a artificial intelligence like wasn't but
00:19:54
we we already have so you will see it uh having a couple
00:19:58
of examples like six examples it's reached from maturity or down to uh to
00:20:03
low chair where we want to improve uh those are the algorithms
00:20:07
uh it here that's what we call the forced from examining such a product which exploits
00:20:11
the the data coming from open source and here we already have some artificial intelligence
00:20:16
such as much image up assisting optical character recognition uh being able to um
00:20:23
to to detect whether the speaker doing some language translations speech to text
00:20:29
a knowledge extraction something want to understand that the context and uh well about which people
00:20:34
we're talking about and as well doing some opinion and analysis on on social major
00:20:40
so this is one of our products which is what my true
00:20:46
as well as some some quite much you algorithms but the cloud detection so
00:20:51
uh we we use redeployed owning a to do this cultivation because sometimes like a peach or there
00:20:57
it's not so easy to know is it as no one the mounting size it is the cloud and when we sell images to our customers
00:21:04
yeah it's been really want to have perfect features without any clothes so we we train over a a
00:21:10
option to insert prisms to detect those clout i really we came to some uh really good accuracy
00:21:17
less than three percent a fellow rates so much much better than than the than the human
00:21:26
another example she's the automatic change detection
00:21:30
so basically he is to is from from from the the images coming from satellites
00:21:35
for probably from the ladies at the same the return of the same position
00:21:40
uh being able to detect even if any change happen on the match or not
00:21:44
and from that uh trying as when fall onto a on
00:21:48
which this information beef anywhere that unlike like open source
00:21:53
and uh and she's you know this is the the kind of aggressive that we are developing
00:21:58
you can imagine the the different around of applications uh so read for example the setting
00:22:04
some services for deforestation to see the impact of industry and deforestation so looking up
00:22:09
forrest how how to grow how many degrees uh we have studies around to your bands that there
00:22:16
have been studies i can control of um surveys so or one descent automatic change detection
00:22:24
another example um uh for for customers actually quite important for them it's to to detect the
00:22:29
object into the image so really be able to uh to extract those those um
00:22:35
the subjects could be vehicles could be planes could be shapes uh as well some ground image processing algorithms
00:22:41
to be able to count how many houses to s. c. uh on this site you manage
00:22:45
and really to to accelerate uh the the the processing them to be able to derive from that
00:22:51
over over information so this is the type of person that we're developing
00:22:58
i know we're coming to some algorithms which are we follow a g. i. l. but which are really
00:23:04
uh the important for custom those it's or want to uh object extraction from those to the majority and be able
00:23:10
to uh to compare two images so to for example to be here able to to understand the semantic context
00:23:17
uh from from the speech uh on something that is a flag is a big conveys a
00:23:22
building and then being able to do some some are and some costs comparison we
00:23:27
uh over images where is you see the something like this this and buildings is
00:23:31
and location and he as well it's really important for for for customers
00:23:40
the same uh can be applied on on videos so we have a
00:23:44
lot of uh videos going from the or you with these
00:23:47
and actually it's really difficult for the analyst to to look at all
00:23:50
this uh this uh these videos uh it it's really long time
00:23:55
couldn't we could take years to look at all these video so what we want to do is we to extract
00:23:59
out of this uh uh videos the them the best part
00:24:03
of it so we do some indexing of um
00:24:07
so first we detect the metadata of the the videos to understand the motion got tested
00:24:13
um the do ga localisation our pete and we had some imaginary it uh
00:24:17
amenities doesn't okay what is the the accuracy to do you much fuzzy
00:24:22
uh so what we're able to to read to select the best part out of the videos but analysts can can concentrate on
00:24:33
concerning at optimal behaviour so
00:24:37
we we we are running here some some uh some services uh for the protecting the the mightn't
00:24:41
traffic so all the ideas that awful for use on over there the complete uh uh
00:24:46
and being able to detect some abnormal behaviour so for the machine known what is norman
00:24:52
uh behaviours and after that you can detect i really concentrate on on specific seductively
00:24:58
which looks up norman i mean that's the job of done that is to to to
00:25:02
concentrate on on on this part of the name of the of the information
00:25:11
another example so we have the same kind of reasons for for ideas be so this so
00:25:16
the information of the latitude longitude of the um of the aircraft and we do as well some collection of a craft objectively
00:25:23
he he well we're in discussion with some part of that was on airport the
00:25:26
centre so that you can predict the estimated the time uh of a one
00:25:31
really based on all all the data the message that that we are collecting them but to yeah we on the list
00:25:42
and this is really something uh i think which is which is the chief of our intelligence
00:25:47
analyst is to be able to to cross correlate a two types of a source is
00:25:53
so we're working of course when you're not on the open source plus uh you mean
00:25:57
because we we have the data so open source is publicly available and you need to have a
00:26:02
tones of uh of images to to work on a really to to learn from that
00:26:07
to do some the ontology alignment and then be able to do some semantic analysis
00:26:11
to be able to do our which both information one coming from the image and the
00:26:15
other one coming from from open source so there's a lot of work on that
00:26:19
the conduct and she didn't correlation and fusion the channel is not a high enough
00:26:24
up at the moment but it's uh it's really key for for end users
00:26:32
how much to do last slide with the um the way that we are uh as the machine open innovation
00:26:39
um but because we re lies but uh we have to be up and so as i said
00:26:44
that was so that would not be able to two hundred or does that uh this uh this information we need
00:26:49
that uh we need an antiques uh we need help to uh just like child as whether the pasta show
00:26:55
so we are really the last year was a establishing some open innovation
00:27:00
initiatives so we'd even access to all of the status of the something like that or you have some uh you had to have access to all
00:27:06
decent images and we organise a lot of a backup to challenges uh
00:27:11
to be able to to uh to walk we we've all images
00:27:15
uh so we we think that that's the way to to walk to give us so we
00:27:18
assertions that you'd like a huge up with a start up with a small companies
00:27:23
uh we we have some what we call the elapsed do it on one level where
00:27:27
every move over to move some stuff about coming in showing the des capacities
00:27:32
and then we integrate that and walk to give a wanna see what to give us
00:27:36
not the old fashioned way from a bus like at the command a given
00:27:39
is contracted specification really being a strong this about that it's really a and hand
00:27:44
in hand so we we we channel where it's more in a general way
00:27:48
um but but we uh so we we open really the the
00:27:51
innovation and it said uh do this a consistent because as
00:27:56
i said a lot will be able to to challenge all all all to to uh to copy for just strangers

Share this talk: 


Conference Program

Introduction by Hervé Bourlard
BOURLARD, Hervé, Idiap Director, EPFL Full Professor
Aug. 29, 2018 · 9:03 a.m.
916 views
Presentation of the «Speech & Audio Processing» research group
MAGIMAI DOSS, Mathew, Idiap Senior Researcher
Aug. 29, 2018 · 9:22 a.m.
16949 views
Presentation of the «Robot Learning & Interaction» research group
CALINON, Sylvain, Idiap Senior Researcher
Aug. 29, 2018 · 9:43 a.m.
9721 views
Presentation of the «Machine Learning» research group
FLEURET, François, Idiap Senior Researcher, EPFL Maître d'enseignement et de recherche
Aug. 29, 2018 · 10:04 a.m.
14111 views
Presentation of the «Uncertainty Quantification and Optimal Design» research group
GINSBOURGER, David, Idiap Senior Researcher, Bern Titular Professor
Aug. 29, 2018 · 11:05 a.m.
3210 views
Presentation of the «Perception and Activity Understanding» research group
ODOBEZ, Jean-Marc, Idiap Senior Researcher, EPFL Maître d'enseignement et de recherche
Aug. 29, 2018 · 11:24 a.m.
5620 views
Presentation of the «Computational Bioimaging» research group
LIEBLING, Michael, Idiap Senior Researcher, UC Santa Barbara Adjunct Professor
Aug. 29, 2018 · 11:45 a.m.
4130 views
Presentation of the «Natural Language Understanding» research group
HENDERSON, James, Idiap Senior Researcher
Aug. 29, 2018 · 2:03 p.m.
8976 views
Presentation of the «Biometrics Security and Privacy» research group
MARCEL, Sébastien, Idiap Senior Researcher
Aug. 29, 2018 · 2:19 p.m.
6512 views
Presentation of the «Biosignal Processing» research group
RABELLO DOS ANJOS, André, Idiap Researcher
Aug. 29, 2018 · 2:43 p.m.
4027 views
Presentation of the «Social Computing» research group
GATICA-PEREZ, Daniel, Idiap Senior Researcher, EPFL Adjunct Professor
Aug. 29, 2018 · 2:59 p.m.
7158 views

Recommended talks

Big Data with Health Data
Sébastien Déjean
Sept. 5, 2019 · 9:20 a.m.
Parades technologiques et humaines à "CheatGPT"?
Maxime Gabella, Directeur, MAGMA Learning
March 22, 2023 · 5:55 p.m.