Research Computing and Computing for Research
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
Title |
| |
Alternative Title |
| |
Title of Series | ||
Number of Parts | 60 | |
Author | ||
Contributors | ||
License | CC Attribution 3.0 Germany: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
Identifiers | 10.5446/42527 (DOI) | |
Publisher | ||
Release Date | ||
Language |
Content Metadata
Subject Area | ||
Genre | ||
Abstract |
|
00:00
Directory serviceFingerprintProcess (computing)Process (computing)Machine learningMoment (mathematics)Set (mathematics)Right angleBitCASE <Informatik>Data centerSurface4 (number)Point cloudMultiplication signMathematicsSlide ruleBeat (acoustics)Cellular automatonComputing platformLocal ringProgramming paradigmWebsiteWater vaporAxiom of choicePlotterArithmetic meanGraph (mathematics)Point (geometry)Sheaf (mathematics)Cloud computingMetropolitan area networkShared memoryVirtual machinePoisson-KlammerSoftwareSummierbarkeitView (database)2 (number)Standard errorEvoluteBit rateSupercomputerAsynchronous Transfer Mode
05:58
Square numberData storage deviceVelocityMultiplication signOrder of magnitudeProcess (computing)NumberAttribute grammarRight angleSquare numberSpectrum (functional analysis)Virtual machineGraph (mathematics)Data centerBit rateBitComputer animation
07:50
Data centerElectric generatorGreatest elementScaling (geometry)Right angleSystem callServer (computing)Moving averageModul <Datentyp>Series (mathematics)
08:44
Euler anglesGoogolKeilförmige AnordnungComputer networkExpressionInternet service providerFiber (mathematics)Pascal's triangleHypermediaSoftware-defined radioRead-only memoryBit rateCoprocessorCore dumpDDR SDRAMBefehlsprozessorGraphics processing unitSupercomputerVirtual realitySet-top boxNetwork-attached storageSingle sign-onNon-volatile memoryData managementSoftware frameworkMiddlewareClient (computing)DatabaseFreewarePoint cloudWorkstation <Musikinstrument>Mathematical analysisSimulationService (economics)Computing platformEmulationElectronic meeting systemUser interfaceMassInformation and communications technology.NET FrameworkBuildingBitGoodness of fitMoment (mathematics)Limit (category theory)Gene clusterObservational studySoftware industryTwin primeMultiplication signMusical ensembleData centerWeightPhysical systemPoint (geometry)WindowSoftware developerComputer fileNeuroinformatikSelf-organizationExtension (kinesiology)System callLocal ringComputer programmingSlide ruleDemosceneRepository (publishing)Polar coordinate systemProduct (business)SoftwareProper mapDialectBefehlsprozessorMathematicsNetwork topologyPublic key certificateComputer hardwareMarginal distributionProcess (computing)Complete metric spaceGreatest elementWordGroup actionModal logicMereologyOrder (biology)Traffic reportingSet (mathematics)Punched cardVideo gamePower (physics)Source codeNatural numberNumberGoogolNear-ringConnected spaceScaling (geometry)SupercomputerView (database)Open sourceComputer clusterBand matrixProgram flowchart
19:02
Standard errorEvent horizonDisintegrationService (economics)Focus (optics)Data managementTerm (mathematics)Different (Kate Ryan album)Self-organizationAssociative propertyMathematicsSoftware industryConvolutional codeComputer programmingHidden Markov modelGroup actionPoint (geometry)Dependent and independent variablesClassical physicsShift operatorMixed realityWordWave packetBridging (networking)FreewareSpeech synthesisRevision controlEmailXMLUML
23:23
InformationOrder (biology)Wave packetSoftware developerTerm (mathematics)Staff (military)WordProjective planeSelf-organizationRepository (publishing)Computing platformSoftwareSoftware industryOnline helpWeb serviceBitSurjective functionMoving averageCycle (graph theory)Link (knot theory)Lecture/Conference
26:37
Standard errorEvent horizonDisintegrationPoint (geometry)Sinc functionService (economics)Office suiteReal numberNeuroinformatikMereologyData centerWindowRevision controlCASE <Informatik>Point cloudAddress spaceVector spaceMultiplication signConnected spaceComputer hardwareLecture/Conference
29:31
Magnetic stripe cardMachine visionParity (mathematics)FLOPSIntelField programmable gate arrayData centerSpeech synthesisPhysical systemTranslation (relic)Self-organizationUniverse (mathematics)Standard deviationData centerVirtual machineArithmetic mean
30:05
Service (economics)Parity (mathematics)Software testingSpeech synthesisPattern recognitionFormal languageTranslation (relic)Maschinelle ÜbersetzungData conversionLecture/Conference
30:48
Physical systemMachine visionIntelSpeech synthesisField programmable gate arrayMagnetic stripe cardData miningWeb serviceCognitionProduct (business)Self-organizationMobile appSlide ruleSeries (mathematics)Service (economics)Multiplication signCodeWeb serviceCognitionRobotArithmetic meanComputer animation
31:42
Web serviceCognitionMachine visionSpeech synthesisLaw of large numbersData miningContent (media)Fast Fourier transformService (economics)Latent heatTime domainEndliche ModelltheorieFrame problemService (economics)Bit rateCategory of beingFormal languageDifferent (Kate Ryan album)Cluster analysisServer (computing)CurveInformationMachine learningPresentation of a groupFacebookInstance (computer science)DatabaseEndliche ModelltheorieData miningWeb serviceSet (mathematics)Key (cryptography)Speech synthesisTranslation (relic)MeasurementWebsiteDemo (music)MereologyRow (database)Cubic graphTerm (mathematics)Product (business)Type theoryPay televisionCodeComputer fileRobotMetreQuery languageCall centreGoodness of fitXML
36:50
Service (economics)Latent heatTime domainEndliche ModelltheorieSoftware testingData modelWave packetMach's principleOpen setVisual systemTensorLibrary (computing)Integrated development environmentCodeVideo gameMachine learningEndliche ModelltheorieSoftware developerProduct (business)Server (computing)DataflowProjective planeOperator (mathematics)Focus (optics)Combinational logic10 (number)Default (computer science)Parameter (computer programming)WindowWeb crawlerCodeFeedbackComputer configurationComputing platformPoint (geometry)Slide rulePoint cloudGoodness of fitMereologyOpen sourceFrame problemSoftware testingPresentation of a groupMathematicsWeb serviceWave packetVirtual machineRight angleCodeBefehlsprozessorVisualization (computer graphics)Software frameworkSoftware bugSet (mathematics)Disk read-and-write headIntegrated development environmentSoftwareCloud computingGraphics processing unitExtension (kinesiology)Field programmable gate arrayDiagram
41:59
Service (economics)Process (computing)Point cloudLevel (video gaming)Film editingWordPhysical systemInstance (computer science)Cloud computingBefehlsprozessorLecture/Conference
Transcript: English(auto-generated)
00:00
I have this weird job title which says director of higher education research and everyone suddenly thinks that I work for Microsoft Research and I'm not nearly bright enough to work for Microsoft Research. My role of Microsoft is to try to help the research community take advantage of
00:21
our cloud services of a software stack which kind of is exactly what you guys are doing. So one of the reasons that we are working very closely with RSE communities globally is we're kind of trying to do the same thing and I'll
00:42
kind of say right up at the front yeah I know I work for Microsoft but it doesn't I don't really kind of care about that bit because my colleagues AWS who are somewhere one of them used to work with me at Microsoft he's at the back.
01:00
We're all kind of in the same boat and trying to achieve the same things at the moment so we have much more in common than we do against each other. There's kind of three things we're gonna try and I'm gonna try and do in the next 45 minutes. One is tell you stuff that you already know but I think maybe it'll be ammunition for some of you to help in the process of building
01:23
the RSE community with people outside of this room. The second is talk about some of the things that we're doing with RSE communities around the globe and then Sven's gonna do a bit of a deeper dive into our AI platform and just kind of cruise along to give you a bit of a view about other things that
01:43
you can do hopefully hopefully it'll be interesting to you. Last thing I'm a Microsofty so I've tons of slides I've seen them all did they're really good but I don't need to see them so I'm happy to take questions I it's much
02:03
more important to me that you get questions answered than I get through my slides yeah it's probably more important the same gets through his slides but I really don't care so if you got a question at some point stick your hand up makes it way more interesting to me because I'm on Seattle time still and it's just that afternoon bit where I'm starting to fall
02:24
asleep so I'll go into kind of autopilot on the slides if you're not careful so it's your choice you can ask interesting questions or I'll go into autopilot mode right there you go I warned you as they say I'm lying on this
02:45
slide yeah first slide I'm lying already why am I lying because the we doesn't apply to us the intersection that everyone's out at the moment is the move to cloud computing and whilst there's been a lot of movement we're
03:06
basically not even scratching the surface yet and when I say we are not even scratching the surface actually I mean you and the institutions that you work with we ie Microsoft and in the case of Amazon are we don't have a
03:27
concept anymore of on-prem stuff we don't have a set of cloud data centers that we develop against and then some local data centers that we use for something else we just have this set over here that doesn't we closed our
03:43
last kind of on-campus data center about three years ago so we so we're not an intersection we're way past that intersection but you guys still are and a lot of your institutions are not even at that point yet and the
04:00
amount of I'll be provocative the amount of new HPC on-prem things that I see coming out tenders I'm like seriously you know what year are we in and I still see people doing this and and they're gonna keep doing it for a while
04:22
yeah it's when you get evolution it doesn't mean that everyone evolves at the same rate yeah but actually you've just got to look where the graphs are going and it's it's kind of disappointing to me in a way that people don't plot the graphs ahead far enough and it's disappointing to me out
04:41
of not you as individuals but some of the institutions that I deal with because there are really bright people there whose job is to plot graphs for everyone else in the world to say this is what's happening and yet there seems to be some intransigence about looking at their own stuff so
05:01
there is it there is an intersection we're not at it we're kind of past it but we still need to get people through it and that's going to take a while and you guys all know what's going on yeah so this is the stuff that you know big data cloud processing machine learning I hate the
05:21
brackets AI yeah it's machine learning still to me we'll get to AI I'm sure at some point but it's machine learning but it's really when you can combine that together it kind of feels almost like AI to the man in the street it may as well be AI people in this room notes machine learning and that's
05:45
changing the way the research is being done yeah again this isn't kind of new stuff we've been going through this change for a long time fourth paradigm was written a long time ago if you come a bit more up to speed you got people
06:01
like Tony hey used to work for us you know saying you really have to look at the rate of acceleration and you then have to look at those graphs again to see what how this is going to affect everybody it's kind of interesting to look back slightly and say wow weren't those numbers really
06:21
small you know 750 Meg seemed like a lot a while ago I was I reading Apple did their new launch yesterday and you can put 1.5 I'm gonna get this wrong because it sounds wrong but 1.5 I want to say terribly a terabits of of ram
06:45
into the thing I mean I'm like really you know that the numbers are just crazy the LHC is still spitting out more data than anyone knows what to do with it's a good job we throw two orders of magnitude away before you try and do
07:04
anything with it but really it would be nice to look at those two orders of magnitude you know today today we're led on a path that says go look at this tiny little spot in the spectrum because we think there's something there interesting we have data for all this other bit but we have to throw that
07:22
away to look at this single spot it would be nice to run some machine learning against the rest of it to say wow that's interesting there's something over here that the theoreticians haven't actually discovered yet thought about and that's what's not happening and then when we
07:40
get into the square kilometer array things get really wild in fact they get so wild that people are saying how are we really going to cope with this data big data centers is probably the answer so it's always nice to show you pictures here's a small data center you can tell the scale because that's a
08:05
car at the bottom running along the road this is actually two you're looking at three generations of data center in here you get it the right way around the right hand side as the left hand side as you look at it is one of our
08:24
kind of C series data centers then you can see it evolve into these kind of modular blobs and then kind of at the top you can see this backbone piece which is the next series so you can see as we build these things out we're constantly evolving them but actually this this is quite an old one and if we
08:44
look at this and you know look at the top of this the top bit of buildings is the bit in the previous picture this is quite big this is about two miles long and about three-quarters of a mile wide and this
09:04
is one of dozens that are being built the the biggest challenge to us today the biggest thing that slows us down is how fast concrete will set you know you're basically you still have to pour a concrete base to put something on
09:23
trust me people have been looking at doing you know what could we do this quicker than that and then you fill these things we'll get on to what we fill them with in a minute but the point of putting this up is just to re-emphasize the scale thing we all work your work in an industry that is
09:47
using ever more amounts of data ever more amounts of CPU ever more amounts of bandwidth and so you're you're pretty close to the coalface but actually you still have to step back occasionally and just remind yourself
10:01
this is where we're going and this is an accelerating path so if we're building these things today and sinking things off the Orkney's underwater like we did last year what you have to look at is where we're going to be in three or four or five years time UKRI doing a ten-year study
10:25
I'm like good luck if you can predict ten years out at the moment I mean you're doing really well and then we're not just doing this in one place obviously we have a few of these scattered around so I think this slide
10:42
normally needs updating when I do it but I think I'm still accurate that there's 54 current regions and there aren't 54 current regions because how nice it is for us to put a region in Brazil so that Brazilian software developers can use local resources no the reason we have a region in Brazil
11:03
is that European companies sell products into Brazil and they want that connected thing car coffee maker freezer whatever it is we used to laugh when we said that and you can buy all of those things today they want that to
11:20
work locally against a local data center and they need it to work locally because you may have restrictions in that country that say you can dial home but not with all the data only with some of it yeah and that's why we have data centers in China and India and South America and South Africa and
11:44
go on and on and yes you know the ultimate extension of this is one at the bottom of your garden because we're literally rolling these things I think we were France was the last set that we went live on oh no I like UAE was the last set that we went live on so you know this is happening more and
12:02
more and more and more so again think about the scale and think about what that means to you and think about what that means for the resources the same with the network you know we're not just using the networks that exist we're buying building and laying the new networks that are out there because
12:24
you have to when you're moving data at this kind of scale and when you're failing over between regions if necessary because it's not just failing over between those data centers or failing over between a twin set of data centers in one region and another one in the same US region it's about
12:41
failing an entire region over to another region if you need to do so you have a tsunami in Japan you fail everything over to the US it's what you have to do assuming your customers say that it's okay for you to fail over to the US a lot of people wouldn't be happy about that so what
13:01
do we put in them everything that you would imagine frankly proper hpc's within Finiband FPGA systems as well and you know probably only 18 months ago I used to put this slide up and talk about FPGA systems and even an audience like
13:22
yourselves would say wow that's pretty esoteric there's not many people doing FPGA programming you know for the systems that they're working on and even in the last 18 months this is becoming more and more and more mainstream and the guys at CERN are using these because it's the only way
13:41
for them to get the throughput that they need to do some of the work they're going to do equally I've been working with a research group in the Netherlands who are doing neonatal work and they were doing genome processing and they've really needed to get their processing times down into
14:02
their kind of couple of hours margin because you don't have that amount of time with neonatal patients you just don't have hours and hours and hours or days to wait you have to do it quickly so even in the last 18 months what was really esoteric 18 months ago is starting to be more mainstream
14:25
well common let's put it that way and then if you really want to go the whole hog you can stick Cray systems or suppose I should say HP systems these days in as well and again you know the interesting thing about this is most
14:44
organizations in most parts of the world just can't afford to put that kind of infrastructure into their own organizations but now they don't have to you know one of the things that we're doing here and again not just us my
15:01
colleagues at AWS are doing exactly the same thing is that you're democratizing the capability to do research you're saying to someone in a research institute in South Africa you can have access to all of the compute you could ever want without any upfront payment at all and you can buy it by
15:23
the minute if you'd like to do that and you can buy the kind of scale systems that you dream about and so you can attract researchers to your institution not based on the capabilities of that institution but based on what the rest of the institution can offer them or great
15:43
surfing if you're near Durban so you know this is changing the nature of some of the institutions that you work with and it's going to change the nature of how things get done but it's a democratization process and that's a good thing in my view I'll stick this slide up you all these slides are
16:05
available so you're not meant to read them in brief because everyone is rolling out these new systems literally weekly which is again another point it's like we and our friends Amazon and if they were here our friends at
16:21
Google are investing in the hardware in a way that nobody else is so it means that you're getting access to new systems far as basically as fast as they're available which means that you are not compelled to buy something and sweat the assets over a number of years we have to do that as our
16:42
problem and we will continue to do that but it stops becoming your problem and this is just the latest bunch that have rolled out now okay the hardware is great what about the stuff that runs on it because let's face it I was
17:00
I've been at Microsoft a long time convicted monopolist lived through many years of DOJ trials not a pleasant thing frankly as an employee I like to think we've learned from it actually I think we've really learned from it and when I look at how we work now I mean I'm it's kind of weird to me
17:27
that the world's largest installed base of Windows systems is at AWS and we run more Linux clusters than anybody else and we make more contributions to Linux
17:41
not just Linux actually to open source repositories than any other quote proprietary software company out there so it's not the Microsoft I grew up with that's for sure I like to think we've evolved past that but I you know I have to make the point because I know most people still don't see that and
18:05
it's not just you know a one layer we have to do it at all layers to be able to do this we have to really make sure that everything that you want to do you can just do and that's of complete change from us we have to
18:22
make sure we're trusted as well so every certification that's out there we've got to make sure that it's available because a lot of the organizations that are doing the research with you or funding the research need us to meet these requirements so you saw us roll out
18:43
FedRAMP to every region in the US last week because surprise surprise the US government would quite like us to be compliant with that so we have to be open and we have to be compliant in order to be trusted and that's stuff that you know but you can use with others who don't know so what about you
19:03
guys you know why are we why are we here and what are we doing well the first thing is really when we thought about this few years ago the classic Microsoft response would have been we'll go and set up a Microsoft
19:22
community we'll do this Microsoft thing and then we'll invite people to join it yeah and some of you probably would have joined so okay there's probably free beer so I'll go along and it would have been okay it wouldn't have been great and I persuaded a lot of people at Microsoft that that is not
19:42
the way that we should do things actually what we should do is look at the way that the UK had built its RSC community and then work with other organizations around the world to kind of go to you guys to turn up where you are on your terms which is why we're here speaking at your conference not a
20:04
Microsoft thing and why we're working with the South Africans the Nordics the Dutch the US guys to try and get the same approach going there I have to tell you it's harder yeah it's a it's a lot harder and some of my senior
20:23
managers kind of look at me and think I'm a bit mad but I think it's the right way to do things the other thing that it does is it means you don't have to worry about us losing focus because like most software companies you know the
20:40
focus shifts dramatically sometimes if we had set something up the danger is at some point somebody decides to defund it and then you guys go what about us yeah you don't have to wait it's your thing you never have to worry about that with us you you may decide you want a different sponsor
21:03
next year great that's that that's a pretty minor thing to have to worry about what you don't have to worry about is our suddenly disappearing in your organization going up in flames so we want to come to where you are and we're doing that by investing in the community in a lot of different ways
21:20
we're putting a role that we call a research engagement manager we've got one in the UK and Imperial we've got one in Australia in Melbourne one in Hong Kong today we've got a different setup in the US because we're working with Indiana University to roll out a program on our behalf there but these
21:42
people are not Microsoft people they're actually researchers or their research associates or their RSEs so again what we're trying to do is to build bridges into the community and to say tell us what we need to do tell us how we need to work tell us what you need for you to be successful because
22:02
fundamentally if you are successful we will be successful yeah it's not hard it's sometimes hard for me to get my management to understand that but it's not hard it's just the way we should do things and next year you're going to see us do a lot more in terms of I use this terrible training word here
22:24
training for the research community actually what that what that really means is something that we did two weeks ago at Imperial in fact we did it a mix at Imperial and our reactor center in in London where we got a group
22:43
of interested people together some of whom are RSEs and said tell us the stuff we need to fix tell us the problems you run into we ran through a bunch of scenarios with them that they had brought to us not that we had started with and then we're looking at okay what were the roadblocks that
23:01
people here are there's this issue in AAD provisioning that doesn't set things up correctly that if you try and then use it as an RSC if you if you're using AAD hmm you don't get to provision the services that you want without making some basic changes we need to change that great our engineering people love that it's like instant instant this is what you need
23:25
to fix to allow people to use your service more and then we're publishing all that information out onto github and out into our technical documentation on docs.com so that it's there for anyone to take advantage of again it's
23:41
not our stuff it's your stuff we're just providing a place to someone to be able to host it and people to be able to go and find it and link to it and work with it in the future yeah hopefully that's what we call training yeah because if I don't use that term internally people really
24:01
don't get I mean that most of the people I work with have no concept of what you guys do anyway yeah so just let me use the training word on here don't don't take it to heart too much and then the other thing that we're doing is we're trying to integrate our our all of our technical
24:22
documentation and all of the github repositories that we and you create together inside the main azure portal and you'll see this roll out at the end of this year as well what this means is it's really easy to go to one place start project understand where you need to go in order to get the
24:46
documentation to help you with that project and if you're working on something new how easy it is to be able to push back into the community so that everyone else can use whatever it is that you're working on if we can make that kind of cycle easier then hopefully we'll help you
25:05
guys in your work in trying to help the researchers get their work done our ultimate goal is just to try to help researchers get their research done faster get their publications published quicker and make your lives a bit
25:21
easier in doing that and to try and help you guys stay up to speed because as people said earlier you know this stuff is advancing faster and faster and faster and the biggest problem is actually staying up-to-date making sure you're taking or able to take advantage of the latest developments
25:41
and the latest piece of things you know software is really eating the world every organization really is a software company these days which is kinda nice certainly should be good if you're in the software business every
26:01
organization is depending on people like you more and more and more it's just that not all of them realize that yet yeah I like to think that we do and that we're going to try and invest in this community and in building these communities more and more and more in the coming years but at least you understand what we're doing and why we're doing and the acceleration
26:25
that's going on in that I'm going to hand over to my colleague Sven he's going to talk you through our AI platform just before I do that and just to check you're awake any questions you can take I can take questions at the end
26:40
any questions at this point yes yeah fundamentally the question was what's the business case how do I get funded yeah as yours what the company's
27:05
betting on I mean you're you're still here people talk about Windows and Office and stuff like that but frankly it's Azure services or bust
27:21
yeah I mean it's it's we're taking the hardware off your hands if you want to yeah I mean it depends how far you want to go if you want to do functional computing you can do that today if you want to do VMS you can do that today we I still see too many people doing VMS in the cloud and
27:43
thinking it's cutting-edge and I'm like seriously you know hopefully the anyone else before I hand okay oh oh sorry go on then one more yes which gap
28:10
in particular yes yeah we're kind of going where the opportunity is so we were
28:25
the first to stick data centers in South America with the first to stick data sense in Africa there's more data centers going into Africa there's more data centers going into South America you have a long lead times on some of these things so it's not that it actually have to wait for some of the
28:43
cables to be laid as well that's part of the problem yeah it's not just concrete drying out sometimes I've sent Central America has some real issues around connectivity so does Central Africa North Coast of Africa is a lot easier you'll see some movement shortly from us there we're going
29:03
everywhere it's just a case of you know sometimes you've got to wait occasionally all right there's no it's not political thing we'll stick data centers wherever we've got them in China I mean trust me but you need
29:22
like approval from the China government to go there actually yeah so what I'm doing is I'm talking about AI so my name is Sven Sven Vilderman I'm a technical solutions professional at Microsoft and normally I had do not have researchers I even if I'm having all the universities as my customers normally I
29:43
do not speak to the researchers I speak to the people wanting the data centers in the organization as well and often I'm talking about AI and was what does that mean and Brad already explained that's mainly machine learning nowadays we often speak about weak AI that's what we are dealing with there's
30:02
no strong AI yet hopefully it will come we will see so and we made some big things like the last years and the improvements that have been made probably you made them or your colleagues made them white that's this is what research and we reached a lot of human parity with a lot of services
30:22
so there's some speech recognition tests where human parity has been achieved as well as in machine translation and conversational questions like this language there's text understanding stuff so that's that's cool and we do have a lot of own research departments all over the
30:44
world the cool thing here is they are not like separated from the rest of the organization they sit in the engineering and improvements researched us goes directly into our products so that's good to know when Microsoft twice to
31:01
think about AI we always divide things in three stuff that's always good for slides so that's why we do it and the first thing is AI apps and agents and now you might think what does that it mean is Alexa something that is very smart and an AI app and agent or kontana you may have heard of
31:22
it maybe not then you could say yeah probably and there are Azure cognitive services and there's an Azure bot service that is the stuff behind it so what does it mean Azure cognitive services is mainly it's just a bunch of API's you can use and it does something that you are able to do as
31:44
well detect something on a picture translate speech to text to do translation stuff like that and there's the bot service what is the bot service it mainly does one thing for you if you want to create a bot that works with Facebook and maybe slack and teams and whatsoever you don't need to program
32:04
these channels again and again and again you want one place to host it put your logic inside and don't want to need to handle all these different channels how how do I get to telegram channel and make that happen yeah we made it easy for you to do it and combining both things you can create
32:23
quite good absent agents helping you and your customers and your colleagues to ask questions so get help hand over like in a in a situation of in a call center stuff like that so these are all the column services we are
32:40
talking about there are five categories you can read it on your own look it up you can almost every one of those can be tested like totally for free online then you if you do a lot of calls it costs some money but if you just use them like once in a while they're very cheap but you don't even
33:01
cost anything at all so this taking part is a knowledge mining so what does it mean I also do you have a lot of other public or government customers out there and all the instruments as well and they have a lot of different data's data sets and those are not like not everything is stored in
33:22
database and easy accessible sometimes it's even paper that needs to be scanned or pictures or PowerPoint presentations or whatsoever and we need to access this data to make them available for one imagine like insurance you're calling them and you wanted want them to have all the information
33:44
about you maybe you send a picture of you if you damage and it should be easy to access those data so there's also some services we are combining we are talking about as a search that's like one of the cognitive services combined with the OCR detection language detection and stuff like that you can
34:02
make really good yeah productive services out of that we did one thing to deal with that we used the John F Kennedy files JFK files they are not confidential anymore so anyone can access them now we scanned all this seventy thousands documents and try to access them so there's a public
34:22
website also out there you can just access it it's JFK demo as you're just Google it or Bing it or whatever and and then you can go there and just type any anything you like and you can see for instance Oswald who's probably the guy he called turn off Kennedy we don't know and then you see the
34:45
pictures and all the records where his name is and where the meter text inside and go there the good solutions out there so that's what we mean when we talk about knowledge mining what is probably most important for you and I know I just have a couple of minutes left but that is machine learning and
35:03
so machine learning comes in all the different flavors and we have a lot of different tools and sets and stuff we are doing out there right so we have some pre-trained models like the cognitive services the bad thing is you cannot look behind the models we are using their whites that's just that's a
35:22
closed shop you can use them or not use them that's cool for commercial customers that want an easy solution to work like something that works now and but they don't need to explain in the details how it works so probably not the best solution for researchers but we also host all the different
35:42
data science tools and make them accessible in our services for instance Databricks and I don't know who of you knows spark as an engine spark spark spark some people like yeah that tool mostly everyone nowadays uses to access big data and do big data queries and Databricks is something you
36:06
would call spark premium there's some guys that also worked at the Apache spark and stuff and they made it even better and now there's a first party service in Azure it's called Azure Databricks and what is cool about that one normally when you're a data scientist working at the institute or
36:22
even at a company you want to start today and then you are dealing with your big data cluster installing spark getting Hadoop or Cloudera running on your cluster and then spark on top of that and trying to to deal with it and I did not have seen one person now that did it right at the first try
36:44
so it's very very hard to configure those stuff and Databricks makes it easy for you to just go on what you really want to do you want to analyze your data right and you can create your cluster in just 10 minutes you can just put some parameters in how many worker nodes do you want how big should
37:04
they be and you get a head note and if you work on notes and then you are done you can do your work so it's quite nice there's also an Azure machine learning a service that combines like all the training and testing I'm going to that in the next slide and so Microsoft is not the
37:20
Microsoft we have been like 10 years ago and we are not we do have our own our own framework but to be honest no one used that so we don't really stick to that one people out there they really use pytorch or tensorflow and so what we are doing more and more is focused on that that once and make it
37:43
easy to run on that in our cloud to make it easy for customers to do the data science through the machine learning because that's something all the Institute's and researchers and companies want to do nowadays and we try to stick what's out there and what's out in the market and not try to reinvent the wheel so to say and Brett already talked about and powerful
38:05
infrastructure we have behind the FPGAs GPUs and CPUs out there so there's one thing that's if you think about machine learning that's a big problem because the typical data scientists I met there normally they think they're done when
38:23
they found a good model with like 98% precision and we call so and then they think they're done but that's not a production right so it's easy it's not easy to go beyond that one and put that into production and deploy the
38:41
model and monitor the model as well and there are not many tools out there mlflow from open source project I guess some is another tool that does it but there are not many tools like considering the whole workflow and there's something out that's called Azure DevOps it's mainly made for developers three tools there's another one three minutes left okay okay and
39:06
Azure DevOps we bought a got an extension out there for Azure DevOps so that you can do all the machine learning operations out there with Azure DevOps as well so training your model validating your model deploying
39:20
model your model and monitoring it and you probably need to do it again once a while because your data normally changes your customer change and stuff like that and yeah there's it there's a tool for it now so try it out if you have any questions let me know and if you think that's that it's not good at all that you some bugs needs to be fixed it's good for us to get the
39:43
feedback because that's like something we want to do we want to be better and like do valuable stuff for you it's not just like competing against others and just having the same feature set that others have but also having like nice software and I guess Visual Studio code and it's one of the IDEs some of
40:02
you might know it once on Windows Linux and Mac and a lot of developers love it because we try to do what you will always want to have and not just sell another product with the Microsoft logo in it so one point and where also already points out there is just it's an open platform so if you do
40:25
not have the resources on your own and you want to do something and it might not work with your institute or you want to be fast and you need cloud infrastructure to do it and one of the options is Microsoft we are doing that and this is our main focus nowadays to work on stuff like that Windows is
40:44
still out there but that's like not the main bet we're doing nowadays it's it's really it's Azure and all the things working working there and AI and machine learning is really what we are concentrating on and I would be happy if one of you is just trying it out and giving it a try and let me know
41:03
even if it doesn't work just just let me know just let me know why it does not work for you I'm also having some customers doing genomical data and the genomic data and they want to prepare them and get fast on those data sets and often they need to plan like in the HPC cluster
41:24
when to do those stuff and that's like one slot in a month or so but they want it like need to be fast in the research paper so that's one of the things where we can help and some projects already showed that we are able to help and you don't need to change your code away from tensorflow or
41:43
to any Microsoft product that's not needed anymore if may you have thought so so thanks a lot for having me here if there are any questions just yeah get them now thank you thanks very much then and Brad and I'm sure you or your
42:13
colleagues will be around at the booth for any questions today and also your
42:23
I think I'll hand over to unless are there any questions just there's one in the back right right the question was are there any institutions in Germany
42:55
where we have which have established a policy for using these kinds of
43:00
services but I'm not sure if they have like I'm not we have some institute already working with cloud software on it I'm not sure if I am allowed to put out the names I need to check with my colleague and I'm not sure
43:21
if they have like a policy out of there but I'm sure there there are a lot of institutes or there are some institutes are already doing that stuff and we also even having in a kanken for the wrong the opera it's in our
43:46
on top of the level in a cloud that's actually had far by today's instance all this one didn't not wish for school and our middle of all it's all losing steps the for sure so long so to manage the nation were banned transport and by by the lager wrongs on a practice or not as a nation or in
44:02
transport and at Western or edge compute practice middle of island or specialists in Canada mending could for Anna and need CPU can hold as of