[music] [applause]>> hi, welcome back. i hope you had a fantastic day sofar. now this morning andthrough the general sessions, you've heard a lot about technology that'sin market today, or coming soon. well, now, we shift to talk aboutthe next horizon of innovation and transformation. now, it's these areas that willcollectively shape our future in this industry, so as you hearfrom satya, first, just enjoy it.
you're gonna hear about andsee some amazing things. but i also encourage you to thinkabout how you're gonna incorporate these technologies as idea and possibilities into yourit strategies today. so you'll have a map for how you can take advantageof them in the future. so without further ado,please welcome, satya nadella. >> [applause]>> good afternoon, it's fantastic to beback here this afternoon,
a big night keynote we're gonnahave some fun this afternoon. we're gonna see some realmagical technologies. and i'm gonna talk about,in particular, how we have set ourselvesa goal of democratizing ai. the central thesis andgoal we have with ai starts with our missionto empower every person and every organization onthe planet to achieve more. we are not pursuing aito beat humans at games. we pursing ai so that we canempower every person and every
institution that people build withtools of ai so that they can go on to solve the most pressing problemsof our society and our economy. that's the pursuit. and to build perspective on this, first let's go back to whatwe've been talking about. mobile first, cloud first. in fact, this morning scotttalked about how we're living in this world, how customers areachieving digital transformation. where it's about the mobility ofthe human experience across all of
the computing in our lives. that's what the cloud enables. even the cloud is nota single destination but a distributed fabric. that's what's driving all of theseambitions that we have in terms of technology. but what is at the intersectionof our three ambitions is ai. that ability to reason overlarge amounts of data and convert that into intelligence.
that intelligence shows up ashandwriting recognition on windows 10. or the windows hello feature,the ability to face recognize you. or even magical new deviceslike the holographic computer, where you have the ability todigitally reconstruct and recognize everything that you see and then tosuper impose objects in that world. how we infuse every application,cortana, office 365, dynamics 365 with intelligence. and the buildingblocks that constitute
intelligence are available asdevelopment services in azure. that's what we're doing. that's the approach we're taking. but to truly understand and perhaps build even more ofa perspective, let's step way back. to, perhaps,what is the first machine that democratized access to information. the printing press. in 1450 or so,when the printing press came out
the gutenberg bible got published. the moveable type became prevalent. before that, we had somethinglike 30,000 books in the world. and 50 years after the printingpress, we had 12 million books. it changed how humans both createdinformation and used information. you can in fact trace backeverything in the modern era through our ability to create anddiffuse information, and learn. the next inflection point, perhaps,in this information explosion, was 1989 andthe birth of the world wide web.
and it's pretty stunningto see the amount of data, the amount of informationthat we are generating. i was just reading this weekend, a report from the idc which said,in 2015 we generated close toten zettabytes of data. now, what's fascinatingto me is what that report projects we will generate in 2025. we will generate somethinglike 180 zettabytes. i mean, we're getting to a pointwhere we don't even know what
to name things. we've gotten to a point, thatmarch from peta to exa to zetta, what comes next we don't even know. so, in all of this informationexplosion what remains scarce is something that i'vetalked about in the past, it's human attention and time. ability to make sense ofall of this information. so, that's really what we allneed to turn our attention to. we've used technologyvery successfully to
democratize both creationof information and the distribution andaccess of information. and now, we need to turn totechnology to democratize creation and access to intelligence. that's the approach that microsoftis taking with our ai efforts. we have four core pillarsto what we're gonna do. it's agents, applications, services, and infrastructure. when we talk about agents,
in our case cortana,i think of it as the third run-time. what do i mean bythe third run-time? just like the pc operating system,or the mobile phone operating system, or the web and the browser,it's the new organizing layer. it's what helps mediatethe human computer interaction, your ability to get toapplications and information. this new category of the personaldigital assistant is a run-time, a new interface. it can take text input.
it can take speech input. it knows you deeply. it knows your context,your family, your work. it knows the world. it is unbounded. in other words, it is about you,it's not about any one device. it goes wherever you go,it's available on any phone. ios, androids, windows,doesn't matter. it is available across allthe applications that you will
use in your life. and we are well on ourway here with cortana. in fact, we have 133 million activeusers each month using cortana. and they are across 116 countries. and they've all ready asked12 billion questions. and that is what is driving eventhe skill's ecosystem of cortana. the fact that we have these sdksthat allow developers to be able to infuse cortana with moreintelligence is what makes cortana even more relevant every day forour everyday use.
and so,the first demo we wanna show you is where we are with cortana andthe cortana skills so that you can get more out ofevery moment of your life. to do that, welcome laura on stagefrom the cortana team, laura? >> thanks satya. hi guys. >> [applause]>> we're building cortana to bean indispensable personal assistant. cortana learns about me,my organization and
the world around me to betterassist me throughout my day. i can start interacting with cortanafrom the moment i get in front of my pc with my voice above lock. hey, cortana,what's my next meeting? >> coming up next, you have show off my skillz atignite from 5 o'clock to 6 pm. right, so as my calendar said, today i'm gonna be showingoff some of cortana's skills. many of these are alreadyavailable today in windows 10.
and some show our vision for thefuture of artificial intelligence. cortana's more thanjust a voice assistant. by learning about me, she can help me keep on top ofthe things that matter most to me. cortana works with office 365 to getto my organization and can tell me things like my next meeting orwhen i have a conflicting meeting. i don't know which one ishould choose from there. and cortana can also let mesee who i'm meeting with. so here's will.
if i want to know more about will,cortana brings in information from places like linkedin ormy upcoming meetings with him. i can even see ourcommunication history, in this way i'm always ready forwhat's next on my schedule. now one skill that we're bringingto cortana that i'm really excited about is soon cortana'sgonna be able to help me make sure i keep on top ofthe things that i promise to do. using machine learning, we'repicking out commitments that have been made in email, like this onethat i sent to talon telling him i
was gonna send a recapof this event. cortana's proactively remindingme about that commitment. so this way, i don't have to forgetabout missing the deadline and i can rest assured thatcortana always has my back. cortana can also keep on top of thethings i'm passionate about, like, university of florida football, it'skind of a rough game on saturday but it's still great tobe a florida gator. some claps, thank you. [laugh] so cortana can also help mekeep on top of my fitness routine.
with a new skill that we'rebringing called health insights. cortana's taking informationfrom the health cloud, things like my physical activity andsleep patterns and combining that with whatshe already knows about me. like my schedule andmy daily routine, and she's bringing me these proactiveinsights using machine learning. things i wouldn't haveunnecessarily known or maybe don't wanna know like i eattoo much fast food when i travel. and she can even tell me that i'mgonna miss a workout at the gym this
week because i'mtraveling to new york. not only do i get that alert buti get a recommendation for what time i canreschedule my workout. so this way i stay on topof my fitness routine. now cortana also works acrossdevices and in this way it can bring me information towhatever device i'm on. so here's an urgent text messagethat my boss sent to me. of course,it was originally set to my phone. but here i am on my pc andi'm getting that text message here.
i could go ahead and reply online. but because he's asking me to dosomething urgent that i don't want to forget after this demo,i'm gonna add it to my to-do list. and you know what,cortana can help me there too. cortana works with many apps andservices to augment her skills, one of those is wunderlist. so now i can easily addsomething to my list. add monthly report to my work list. >> okay,
i added that to your work list>> so you can see it's showing upright in the cortana canvas for when i need it, and i've also gotit here on my wunderlist app. now let me talk about thisnotification that i just got. we're thinking about ways that wecan use proactive information so that you can keep on topof your business metrics. so being on the cortana team, one ofthe metrics that's really important for me to track ismonthly active users. what if cortana, using informationfrom power bi, could alert me when
i've hit an important milestonethat i care about, or when things are trending up anddown in those monthly active users. i can take that notification andget more information. and here cortana's bringing in adata visualization from power bi of those monthly active users. now this is monthly active users butyou could think about how this could apply to any business metricthat really matters to you. it's all about getting theinformation you need just when you need it soyou can make business decisions.
now the last thing i want toshow you is how cortana works with sticky notes. so here's a sticky notethat i wrote earlier today. because this sticky note hasintelligence behind it, i can click on this phone number and makea phone call, straight from skype. or i can click here, andadd a reminder straight to cortana. it's gonna set that reminder. and since cortana's acrossall my devices, that text, that ink text is now turnedinto text on my phone.
and here on my android device,that same reminder is here. so it goes off whatever devicethat i have closest to me. sticky notes also work withbing to bring in information. so if i were to write an address,i could get a map, or i'm gonna write a flight number. there we go. and it's bringing detailsof that flight directly into this sticky note. and those are real time detailsabout the status of the flight.
so those are just some of the waysthat we're using intelligence to bring new skills to cortana and we look forward to showingyou more in the future. but for now i'm going tohand it back over to satya. thank you. >> [applause]>> thank you, laura. so now let's switch totalk about applications. and how the same approachof infusing intelligence into cortana as thatpersonal assistant,
we can take that approachto every application. in fact i want tostart with swiftkey. swiftkey is one of the most popularkeyboard apps on android and ios, a third party keyboard app. we have over 300 millionusers of it today. in fact it's already taken overa trillion plus key strokes. it's saved in fact people, something like 100,000 yearsof keyboard entry time. and the technique we've usedto do all of that was this
engram based approach, where wewere able to predict the next word based on the previous n words. just last week,we made a giant leap. we've switched todeploying neural networks. think about this, that meanseveryone of us will have a neural net that learns on how we type. so that means it goes beyondthe previous three words or four words that weentered to predict. it goes beyond that tothe semantic meaning
of what we are trying tocommunicate, across all the devices. so it's no longer that a keyboardis attached to a device. the keyboard is attached to you and has a neural network that'sconstantly learning and helping you get those magical typingskills that we always wanted. another application is mileiq. the mileiq is fascinating inthe sense that it's a context aware, location aware, intelligent app. it's actually in fact the number onefinance category app both on ios and
android today. there are over 60 million americanswho are road warriors who have to keep track of their mileage,especially for work so that they can get the taxrefund from the irs. and in fact this mileiqapplication has returned already $1.2 billion tothe users of mileiq. and the way it goes aboutthat is by really, again, taking all of the signal,in this case your driving and location information, but convertingit into intelligent action.
that approach is what wenow wanna take into our very mainstream products,office 365. in fact for me,when we talk about office 365, it's not just simply thatwe are moving to the cloud. and this is a new way to deliversome of the same technology as a service. in fact the most profoundshift is in the fact that the data underneaththe applications of office 365 is exposed in a graph structure.
and in a trusted,privacy preserving way, we can reason over that data andcreate intelligence. that's what's really the profoundshift in office 365. and you see this in many, many ways. you see it in this focused inbox. i mean, think about email triage,email triage on your phone or on your pc. if you wanna talkabout that scarcity we have of human attention and time.
the ability to deploya custom neural net model, that again understands your inbox. it's not a generic model. it understands the type of mails,the people that you're corresponding with, the semanticcontent of your inbox. and to be able to focus yourattention on things that matter the most. skype translate is somethingagain that's fairly magical. i mean there's been threedifferent strands of research
that came together tomake translation happen. there was speech recognition,there was speech synthesis, there was machine translation. and then skype data. so you take thosethree technologies, apply deep reinforcementlearning in neural nets. and the skype data andmagic happens. we already have eight languages. we see even these emergentphenomena, like transfer learning,
when you teach it onelanguage it learns the other. and really solve thathuman language barrier. even inside of word or outlookwhen you're writing a document, we now don't have simplethesaurus based spell correction. we have complete computationallinguistic understanding of what you're building,or what you're writing. and so, that means we can,of course, correct your spelling, in fact, i think that i wouldbe unemployable but for the red squiggly.
>> [laugh]>> and now, i even have the capability to be a better writerin terms of style and grammar and understanding of what it is that iwant to write about and communicate. some of the learning toolsthat we've built into word and onenote, to even solve for dyslexia, to help students withdyslexia improve their reading rate. and out of the new application of itthat we've just launched this week is, tap. that understands the contentyou're writing and, just imagine,
within that context to bring allthe other content that was created in your organizationjust a tap away. and lastly, myanalytics,because again, going back to thatnotion of scarce time, just like the fitness tracker givesme all this information about all that i need to do to keep my calorieintake and out take on balance. what if i have thatsame feedback loop informing me on termsof time i'm spending, whom am i spending with,on topics that i should spend it on?
that's what myanalytics is about. so office 365 to me, beyond what isthis traditional applications and workloads, it's about infusingthis next layer of intelligence. and we're not stopping there. in fact, we're expandingthat to dynamics 365. and take something like sales,in any business application you've alwaysexplicitly modeled the work. when it comes to sales,you have modeled sales people, their accounts, customers, leads,prospects, opportunity pipeline,
it's all modeled, there islots of data that's captured. but there's one real problem,which is, most of the sales activityhappens outside of a crm system. and so, the goal of intelligence is to be able to reason aboutyour sales data model, not inside just your crm system,but outside. so we're buildinga relationship assistant, that's going to ship in novemberas part of dynamic crm, to truly transform a crmapplication from the inside out.
so when you log in to a crm system, what you're gonnasee are these cards. these cards that allowyou to take action inside the system based on activitythat is happening outside. so, for example, it'll know becauseof its ability to crawl the web, about changes that are happeningwith your customers. changes that are happeningon linkedin, on one of yourprospect's job titles. and so, now you can gochange the information
in context of things thatare happening on the outside. so, the wed graph informsyour crm actions. similarly, the microsoft graph,say you get an email from one of your customers, instead ofyou having to triage, copy, paste, reenter into a crm system,what if when you log into a crm system you got to triage,in fact, your email actions. so, a new rfq request comes in,a new lead comes in, as new opportunity, somebody, in fact one of your colleaguesin the sales accounting,
sends you a mail and that getsflagged as an opportunity risk item. and lastly, of course, we'll applythe intelligence to the crm data model itself, so that you canget alerts around when is it that an opportunity is gonna close,or monitor account activity. so this is a complete revamp of how one even goes about thinkingabout a crm sales module. a similar approach is what we'retaking with customer service. again, traditionally whatwe've done, is to build, again, a model of whata customer service agent does.
how do they open a case? how do they escalate a case? how do they keep track of allthe work flows that happened within the customerservice department? really, customer service,starts with the customer contact. so at microsoft, we today haveat support.microsoft.com, a virtual assistant. this is live in us english today, and we are gonna expandthis to all countries.
so customers come in andinteract with the agent, they ask it questions, this virtualagent answers those questions. but, of course,it also runs out of steam and needs to escalate to a realcustomer service from time to time. and that's whenthe real magic starts. if you go behind the scenes,this is the interface that our customer servicereps are using today. what you have on the left hand sideis the conversational canvas where the customer service rep isinteracting with the customers,
solving their problem. but the bot, or the assistant,is on the right hand side. it is, in fact, helping the customerservice rep get better. so this virtual assistant,through a mechanism called reinforcement learning,is not only helping the customer service rep get better,it itself is getting better. so this phenomena of applyingai to customer service will get your customer serviceoutcomes to be more efficient, your customersatisfaction to improve.
so these are two simple but profoundexamples of how ai in sales, in support,are going to transform dynamics 365. so i want to then move to services. the capability that yousee underneath office 365, dynamics 365, swift key,all of that is what we wanna expose as services,building blocks so that you can build the same kindof intelligent applications. the cortana intelligence suitetoday, already is transformative. equal labs is using it for watermanagement, schneider electric is
using it for energy distribution innigeria, l v prasad eye institute is using it to bringaffordable eye care in india. rolls royce is using it forfuel efficiency. this notion of using machinelearning on large amount of data is already having that transformativeeffect across every industry, across every country. and now we're adding new capabilities tothe cortana intelligence suite. the first is the bot framework.
bots are like applications, just like how you build a website ora mobile app. every business for every businessprocess is going to build a bot interface, because it'sa convenient way for users to interact with your information,your data, your process. but in order to build a bot, youneed to have these building block services that haveconversational understanding, know how to parse natural language,how to have a dialog. so that's what we have nowencapsulated in this bot framework
so that you can build a botthat is available on skype, it's available online,it's available on facebook. so again, we're takingan approach where any bot you build is not captive to anyone conversational canvass. it is available everywhere. and since build,which is when we launched it for the first time, we've had 45,000developers, building these bots. hipmunk and trower, star trek,stubhub, getty images. so many,
many developers taking advantageof the bot framework already. so we now are partnering,which is a fun bot with nfl. it turns out nfl as you can imagine,has lots of data. and one of the applications thatthey create is for fantasy football. and it's the most data drivenapplication at least i've came across. and so we started experimenting and this is something that we are inthe early stages of building. and our hope is for the next season,we'll have this bot.
which will allow each one of us toreally engage in a very different way with what all of us,at least in the united states, are obsessed with. which is fantasy football. and so what i thought is to reallyshowcase this, i'll invite up on stage someone who knows a thing ortwo about football, deion sanders. please help me welcomedeion on from the stage. >> [applause]>> hi, how are you, deion? >> i made it>> [applause]
>> forget the superbowls and world series, i made it. >> [laugh] jim,we're gonna go to this station. i know that you're the only personwho was in the superbowl and a world series and have you ever tried a real gameof cricket though [laugh]? >> cricket, you meanthe kind that you fish with? >> i'll teach you that, that's theone sport you gotta learn [laugh]. >> i gotta learn that.
>> so we're logged in hereto your skype account and i look at your friends list. wow, one of these days,i'll have those kinds of friends. >> no,your friends are doing pretty good. >> [laugh] so this is a bot that webuilt or we are in the process of building with nfl,that obviously you know a lot about. and the idea is how can we changethat fantasy football interface and make it more fun andmore data driven? so what it allows you to do is toreally compare player profiles,
change your roster, orimprove your roster. so maybe what we should dois look at your roster. should we do that? >> yes. >> let's go, let's click that. >> i like when they talkto you like they know you. welcome back, deion. >> [laugh] that'sconversational understanding. >> love that.
>> and so here is your roster. i mean it's a pretty-. >> pretty good. >> pretty good, i mean i don't know what 100in-flight points means, is it good? >> it's okay.>> [laugh] so how about we ask for some recommendations? let's see what the bot says. wow.
>> matt ryan with drew brees. they're playing tonight. i like matt ryan, but i love drew brees becauseatlanta's secondary isn't good. but when you think about drew brees,if they're losing, he's gonna get more opportunities to throw, sohe may receive more fantasy points. >> so you think the bot's wrong? >> it's all right. >> [laugh] sohow about let's compare the players,
let's see what the botknows that we may not. >> now that's good. >> so it's sort of saying that bingpredicts that saints are gonna win. and drew brees will scoremore fantasy points. he's playing at home andthe weather will not be a factor, so you should bench matt ryan. >> you know i played foratlanta and it's kinda hard for me to pick drew breesover matt ryan. and i play two sports right here.
but i like what they're saying. that's a pretty good summary. >> all right, lets do it. all right, so here is your dreamteam for this week, i guess. so that is really the beginningof hopefully what can be transformative of even how a deionsanders can interact with a bot and be the manager ofa fantasy football league. >> they had more knowledge than idid because i didn't even consider that in a dome the weatherwould not be a factor.
>> there you go. [laugh] thank you so much,deion, for being here. >> thank you.>> it's such a pleasure. >> i appreciate it. thank you, thank you. >> [applause] >> so bot frameworks are beside, behind these simpleconversational interfaces. it's some of the mostsophisticated ai capabilities
to understand human language. and that means we will democratize application usage foreveryone and everything. that's what's excitingabout this new interface. i wanna now to talk about anotheraspect of cortana intelligence. it's gonna be availableto developers and is available to developers. that is the cognitive apis. again, we launched theseat belt and, since then,
we've had more than a billion apicalls to these cognitive apis. and just to kind of putthis in perspective, the capability that is on tap. that is one api call away. are some of the world class,world leading technologies. microsoft has, today,the world's speech record. the way you manage, you sort of keeptrack of it is the word error rate. just two weeks ago, we publisheda new world record at 6.3% word error rate in what iscalled the switchboard test.
so that's the speech api that isnow part of the cognitive apis. microsoft also has the world recordwhen it comes to image recognition. and then again in a competitionon the image net, we deployed 152-layereddig deep neural net. using a new techniquecalled residual learning. which again,has been pretty transformative in terms of its objectrecognition capability. that is available as a serviceas part of cognitive services. so imagine what you as developerscan start doing with this tech.
and that's what's leadingto develop this everywhere. the first developer wantto talk about is uber. just this week,they launch a new app. they now takes selfies oftheir drivers, recognize, and identify their driversbased on image track. and the reason for that is simple. it's driver safety andpassenger safety. let's roll the video. >> we come to work everyday to pilot, test, and
launch new technology solutions. real-time id check isthe latest technology example. where we at uberare constantly developing and testing new solutions to predict,prevent, and reduce security risks in waysthat weren't possible before. it's through this partnership withmicrosoft that we've been able to develop this technology quickly andensure that every rider and every driver hasan excellent experience. real-time id check is a promptthat appears in the driver's app,
asking them to take a self-photo. we can do a check in real time tomake sure that that identify of the person who took the picturematches the account holder who's been approved to drive. doing that servesa couple of purposes. drivers know thattheir identities and their accounts are being protected. and riders know that the driver whothey're with has been screened. >> jen?>> evan?
>> hi, yeah. >> real-time id isa smart technology. what that means is it factors in andaddresses the edge cases. the situation where the driver iswearing glasses, or a hat, and they weren't in the identificationthat we have on file. the beautiful thing is they canrecognize these changes and ask the driver to remove theirsunglasses or retake the photo. the partnership with microsoftcognitive services allowed us to go from idea to execution toimplementation across the country in
a matter of months. already, we've been able to makethousands of right safer and very soon we're gonna be makingmillions of right saver through this >> [applause]>> sticking with the automotive theme, i wanna talk a little bit aboutthe work we're doing with volvo. i mean volvo stands for safety. that's their brand. and one of the big issues,long before fully
autonomous cars,is distracted driving. in fact,the devices that we now take into cars are probably going to bea source of a lot of the accidents. and so volvo has been workingto sort of understand how they can recognize distracted driving,and then give driver feedback. beause they want to be ableto design cars that can help divers not be distracted. so to do that theybuilt a simulator, using again cognitive services.
it's ability to not justrecognize people but to even recognize emotions,distractions. >> distraction has been a leadingfactor in accidents today. technologies can help us workaround this in our simulator we're actually taking one step furtherbeyond connected car scenarios. we're trying to get into moreapplying artificial intelligence and cognitive services inour scenarios for cars. the relationship with volvo isalmost a perfect match to what we intend to do with the simulator.
once the car knows the emotionalstate of the driver, it could help the driver to getinto a preferred emotional state. >> so when you step in the car,you will be looked at in how you drive the car froma mechanical standpoint. we also monitor you using ourcameras to improve our cognitive services in detecting your emotions,your alertness level, how much you are basicallyinteracting with the environments. so when we put allthis data together, both mechanical and emotion,
we can actually have a good way toassess your driving conditions. your performance andwhether it’s safe or not safe to be in a car with you. >> [applause]>> this next example perhapspersonifies what is possible now by bringingtogether the two most magical technologiesof these times. cognitive services andmixed reality. we're working with those inpartnership with pinterest to
completely re-imagine whatretailing could look like. a whole remodel which i just wentthrough is a fascinating process. you go to the store,you get samples, you go back home, you look at them,you go back to the store, and you kinda repeat that process what feelslike an infinite number of times. >> [laugh]>> but what if we could in fact use the combination of the socialsignal from your pinterest board. because really don't idea forthe real model starts long before. you may even wizard the story.
it's in your board on interest. what if he can takethat signal next set? with the ability to see the remodelbefore it's done at the store. that's what we're workingon with lowe's and to really show you this inaction let me invite up on stage jennifer stevens fromour hollow lens team. jennifer? >> thank you satiya. >> [applause]>> microsoft and lowe's have partnered to deliver aninnovative new approach to the home
remodel experience, andit's in pilot in stores today. remodels are big projects andbig investments, too. so lowe's wants to ensurethey deliver a highly personalized experience forall of their customers. but interestingly, most remodels start before acustomer ever walks in to the store. many of us begin by pinning pictures of our favorite designs onsocial apps like pinterest. these images can be super valuable
because they're a window intoour unique style preferences. but it didn't tell now. it's been hard for retailer likelowes to extract that data. let's see how we canchange that with and the work were doing with pinterest. as you can see here, in the lowe'sin store app there's a couple, peter andsonja who worked here in a and ensure that pinterestimages with us. now this looks like a simple app,but
on the back end there is a cortanaintelligence deep neural network that's been trained on millionsof pictures of kitchens. when i hit analyse, that dnn isactually extracting peter and senja's unique style preferences. it’s then matching to the best fitof what’s in the lowes product catalog. as you can see i got backa handful of recommendations. it looks like the strongest match iswith the cornerstone one of lowes feature design, an 83% confidence.
if i wanted to i could continue to customize i got some additionalrecommendations fits here. but i think this is a good start forwhat i know about peter and sonja. a great customer experience isn'tjust about better style-matching, swatches, printouts can'tshow how that new kitchen will really look in someone's home. but with hololens we interactwith product recommendations from cortana intelligence we can find animmersive life size experience, and help make that no regretsremodel possible.
let me invite peter and cindy to stage, to designa future kitchen using hololens. now at the end of the stage,you're going to see a camera, and that's to bring the hologramsto everyone in the audience. hi peter.hi sonja. before we get started is it okay if we record this session toimprove our customer experiences? >> yeah sure. >> great.>> i ran up pinterest images you
shared with me throughour style app, came back with a recommendation,it's the cornerstone, so i could put your hotlines on now andlet me know what you think. >> woah. >> cool. >> do you see this, thoseare the cabinets you really like. >> they are, look at the window,peter, it's so real. >> yeah. >> this is also soclose to the pictures we picked.
>> this is really close. check out the counter. i like the sample we looked atearlier but now that i see it full size, i think it might betoo dark for our kitchen. >> maybe. >> can we make changes? maybe try something lighter? >> sure. let me pull up somerecommendations right now.
you're going to see cardshovering throughout the kitchen. just tap the ones thatyou want to change. >> okay. i can try matching the counterwith the cabinets. but i think with thislight countertop. we need a different paint color. this one? >> that looks great. i love this color.
what about those tiles? can we change them andmake them a little higher? that's the back splash. let me pull up an option that's beenpopular with that same cabinet and counter top. yeah, i definitely like this better. what do you think sonja? >> i like it too. this is so much easier than justtrying to imagine how things that
look and fit together. >> totally. jennifer, we wanna addan island to our kitchen, can we try something like that too? >> sure,let me put one up right now. sweet. now go ahead and resize it to how you think itwould fit best in your kitchen. >> well, we like to entertain a lot.
why don't we try this at bar height. >> ooh,i don't know about this height. we mostly cook for ourself. let me try lower itone more time and see. leave it here? >> yeah, i think you're right. this'll give us the extraspace that we need. >> and i can see myselfmaking cookies here. >> yeah.>> how do you both feel about
the rest of the kitchen? >> jennifer, i love it. i think for me, this is it. and i can't believe how much wegot done in just one store visit. >> you know this was great. so what's next? >> i'll email you a recordingof the session and your design. you guys can take it home,review it, make any changes that you'd like,and we can finalize from there.
sound good. >> thanks.>> great, thank you. >> [applause]>> with hololens and the power of machine learningfrom cortana intelligence, retailers like lowe's cantransform their businesses. but there's more we can do. do you remember wheni asked peter and sonja if we couldrecord their session? when we aggregate that data, and
anonymize it across all the othercustomers who've been through similar experiences, we cancreate a powerful feedback loop. here's the cornerstone kitchenthat we were just visualizing. now, when i have the heat mapi'm layering in the hole in and let me did it. i'm giving you an instinctinstant into the costumer's spent the most time looking, you see the countertops, thecabinets and even the kitchen sink. now using our cabinets servicesip's i can add another layer,
this is customer's verbal sentiment, the actual reactions they had to geta sense of how well they liked or didn't like the partsthey saw in the kitchen. now for this particular kitchen,as i kind of glance through it, it looks like customersreally liked it. but there's one item, those cabinet knobs that didn'tseem to be resonating as well. i can investigate tolearn a bit more. when i click on the cabinet knobs.
the word cloud to the topright of the dashboard, is updated with the most frequentlyused words to describe those knobs. with cortana intelligence, findingcustomer insights like this is easy. so what you've seen is howwe're able to leverage the intelligence and power,of the microsoft cloud in seamlessly integrate it intoa mixed reality world. this has been transformative forretailers like lowe's and for their customers. >> [applause]>> it's awesome.
i mean, it's the most harmonious, amicable remodel thatyou can imagine. >> [laugh]>> and to think about technology that can changeindustries, like retail. the cognitive services and mixedreality and signals from the web and social, reallyare what's now possible. and that's really what ilook forward to seeing in the years to come. and i want to end with the lastpillar, which is infrastructure.
now, we're now talkingabout infrastructure that allows you tocreate intelligence. scott talked to us earlierthis morning about azure and how we now we have34 global regions. how it's the most trustedcloud with its compliance. and then the intelligencecapabilities. but the most factoring thing forme, is how we are able to support in azure. the cpu compute fabric upscale and
not just having the scaleof cpus but the ability for developers to use any framework forcreating the eye. tensorflow or caffe or torch orcntk which is what we're re-innovating and building one ofthe best in class frameworks for creating intelligence. but it's not just limited to cpus. we now have the best in class gpuvirtual machine support in azure. you already see developerslike jellyfish using it for computational rendering.
virginia tech using it forgenome sequencing. but we're not stopping there. we are now taking those neural nets, deep neural nets,convolution nets, and saying and asking ourselves,what if we can run them, not just on cpus orgpus, but on silicon, and that sort of led usto build out the fgpas. we now have fpga support acrossevery compute node of azure that means we have the abilitythrough the magic of the fabric that
we have built, to distributeyour machine learning tasks, your deep neural nets, to all ofthe silicon that is available, so that you can get thatperformance, that scale. and to show you this,what i believe is the first ai supercomputer in action. i wanted to invite up on stagedoug burger from microsoft research. doug? >> thank you, satya. [applause] i have to say, i'm really excited to sharethis with you, today.
as a company, we've been on a journey to developthe world's most intelligent cloud. now, we already haveindustry-leading capabilities, with our azure dq offering,which is fantastic for building trained ai models offline. okay, but to support live aiservices, with very low response times at large scale with greatefficiency, better than cpus. we've made a majorinvestment at fpgas. now fpgas are programmable hardware.
what that means is that you getthe efficiency of hardware, but you also get flexibilitybecause you can change their functionality on the fly. and this new architecture thatwe've built effectively embedded an fpga based ai supercomputerinto our global hyperscale cloud. we get awesome speed,scale, and efficiency. it will change what's possible forai. now over the past two years quietlywe deployed across our global hyper scale data centers in 15 countriesspanning five continents.
okay, so let's start with a visualdemo of what happens when you add this fpga technology toone of our cloud servers. we're using a special type of neuralnetwork called a convert illusional neural net to recognize the contentsof a collection of images. okay, on the left of the screen, what you see is how fast wecan classify a set of images using a powerful cloud basedserver running on cpus. on the right, you see whathappens when we add a single. 30 watt microsoft designedfpga board to the server.
this single board turbo chargesthe server allowing it to recognize the images significantly faster. it gives the server a huge boost forai tasks. okay, now let's trysomething a little harder using a more sophisticated neuralnetwork to translate languages. the deep neural network basedapproach we're using here is computationally much harder. it requires much more compute but it's achieving record settingaccuracy and language translation.
okay? so to test the system,let's see how quickly we can translate a book fromone language to another. now i picked a nice small book forthis demo. war and peace. it's about 1440 pages. and we'll go over tothe monitor here, and using 24 high-end cpu cores, we will start translating the bookfrom russian to english, okay?
now, we'll throw fourboards from our fpg-based super computer at the same problem,which uses a fifth less total power. as you can see. >> [applause]>> thank you. >> we're not done. as you can see our acceleratedcognitive services run blazingly fast. eight times fasterwhile using less power. these four boards can translatewar and peace in just two and
a half seconds. but even more importantly, we can now do accelerated ai ona global scale at hyperscale. now [inaudible] a big reader,with a big personal library. so we thought abouttranslating his library, but the thought of sneaking into myceo's house, scanning all of his books, and getting out without beingcaught, wasn't so attractive, so we thought about what's a differentstore of text that we can translate? we decided to go to wikipedia,
now english is the largestlanguage in wikipedia, it has 15, i'm sorry, 5 million articles,about 3 billion words. if printed on paper, it would bea stack a quarter of a mile high. so if we threw the same four notesrunning around 8 trillion operations per second at the problem, as youcan see, it would take nearly 4 hours to grind throughthat quarter mile stack. now of course, we have this fabric,which is global and hyperscale so we could easily throw 50 nodes,50 fpga boards at the problem. which would take us up to buta hundred trillion operations per
second and of course bringthe time down further. but to show you the rawpower of this hyperscale ai supercomputer we've embedded inour our cloud, in our global cloud, i'll show you what wouldhappen if we decide to throw most of our existingglobal deployment at it. okay let's go all the way overhere and see how fast we can go. less than tenth of a second. >> [applause]>> we could translate all five billions words into another languagein less than a tenth of a second.
now, some of you may have blinked insurprise when you saw that result. you may or may not know thatit takes a human being about two tenths of a second to blink. so, we can actually translatethose five billion words if we throw deployments at it in lesstime than it takes to blink once. that’s hyperscale acceleration. that crazy speed shows the raw power of what we've deployed inour intelligent cloud. okay, so now what does it mean,
you may have noticed on the screenthat we're running it over an exa app and if we through ourwhole deployment at it. what does it mean to be at exascale? that means we can run a billionbillion operations per second. it means we have ten times the a.i capability of the world'slargest existing super computer. it allow's us to do things on ascale that hasn't been done before. we can solve problems thathaven't been solved before. the things that weren't evenpossible few years ago.
we're already porting some ofour cognitive services to run on this fabric. now this fpga fit based fabricthat we've built is very flexible. so azure already announced today that they're using it to providetheir accelerated networking. the fastest cloud network. industry leading cloud network. running 25 gigabytes per secondat 10x reduction in latency. so this fabric that's been deployedin between our servers and
our network can be used to runthe world's most powerful ai, it can be used to run the world'sfastest cloud network, or both. so it's providing leadership forcloud networking and potentially for ai. when you do a bing search you'realready touching this fabric, bing ranking is running on it. while many companiesare experimenting with much smaller scale bolt on systems microsoft isthe first to have its global hyper scale cloud enhanced with post cputechnology, in this case fpgas.
it gives us the most powerful cloud,the most flexible cloud, and the most intelligent cloud. and we're committed to using it,to support and empower you, our partners and customers, as we move forward into the ageof machine intelligence together. [applause]>> thank you very much, doug. and hopefully thatgives you a feel for our ambition to democratize ai for what we're doing with cortana andits skills.
infusing office 365 anddynamics 365 with intelligence, the cognitive services,the back framework, the machine learning analytics,for every developer and then building out thisfirst ai supercomputer. but i wanna close where i started,it's never about our technology. it is really to meabout your passion, your imagination and what you can dowith technologies that we create. what societal problem,what industry will you reshape? that is really what we dream about,what we are passionate about.
we wanna pursue democratizing ai, just like we pursuedinformation at your fingertips. but this time around, we want tobring intelligence to everything, to everywhere, and for everyone. thank you all very very much. thank you.[applause]