Huawei HiAI - Enabling AI for Everyone
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 | 90 | |
| Author | ||
| License | CC Attribution 3.0 Unported: 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 | 10.5446/47613 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
|
2
10
13
14
26
35
37
45
48
54
55
56
63
64
68
73
74
77
78
80
82
85
89
00:00
Hill differential equationLocal GroupComputing platformOpen setComputerProcess (computing)Search engine (computing)Decision theoryNatural languageExtension (kinesiology)WeightCommutatorNumberCellular automatonTelecommunicationMetropolitan area networkComputerDifferent (Kate Ryan album)Endliche ModelltheorieMachine visionComputer animation
01:13
Mobile appMobile WebComputing platformOpen setNP-hardData modelSoftware developerFrequencyEndliche ModelltheorieMobile WebFitness functionComputing platformLatent heatFocus (optics)Software developerPlastikkarteOperator (mathematics)Cartesian coordinate systemProcess (computing)ExistenceComplex (psychology)SkewnessComputer animation
02:53
NP-hardData modelSoftware developerVapor barrierOpen setComputing platformLatent class modelAlgorithmLimit (category theory)Constraint (mathematics)Cycle (graph theory)Computer hardwareVideo game2 (number)Point cloudFood energyComputerFocus (optics)Roundness (object)Insertion lossWave packetSoftwareTerm (mathematics)AreaMultiplication signEndliche ModelltheorieRight angleSoftware testingStandard deviationOperating systemSoftware developerMobile WebNormal (geometry)IterationTelecommunicationCollaborationismAndroid (robot)Physical systemProcess (computing)Musical ensemblePower (physics)Capability Maturity ModelRevision controlElement (mathematics)NP-hardPRINCE2Network topologyPosition operatorDomain nameComputer animation
08:38
Endliche ModelltheorieComputing platformOpen setScheduling (computing)Service (economics)Mobile appHill differential equationMobile appLevel (video gaming)TensorComputer architectureProcess (computing)Latent heatScheduling (computing)Greatest elementEndliche ModelltheorieIntegrated development environmentSoftware developerINTEGRALService (economics)MereologyConnectivity (graph theory)Type theoryDifferent (Kate Ryan album)Arithmetic meanElectric generatorComputer animation
10:42
Asynchronous Transfer ModeSoftware frameworkOperator (mathematics)Covering spaceHill differential equationComputing platformTensorDataflowSoftware development kitOpen setResource allocationService (economics)Engineering physicsOpen sourceDifferent (Kate Ryan album)Computing platformComputerComplex (psychology)AlgorithmCore dumpRun-time systemExpert systemOperator (mathematics)Open setBasis <Mathematik>Integrated development environmentVariety (linguistics)Software developerRun time (program lifecycle phase)Multiplication signProcess (computing)Asynchronous Transfer ModeAndroid (robot)TensorProduct (business)PredictabilityEndliche ModelltheorieGreatest elementType theorySource codeSpeech synthesisCategory of beingMereologyComputerMessage passingSocial class19 (number)10 (number)Archaeological field surveyPower (physics)Library (computing)Right angleCircleElectric generatorCoprocessorProper mapComputer animation
14:44
Operator (mathematics)Data modelData managementCoprocessorConvolutionCompilation albumComputer hardwareCalculationMobile WebSoftware developerComputing platformOpen setCore dumpSoftware developerComputer hardwareComputing platformEndliche ModelltheorieCellular automatonReal-time operating systemCoprocessorFigurate numberComputer animation
15:29
Operator (mathematics)Data modelCoprocessorConvolutionCompilation albumComputer hardwareCalculationMobile WebSoftware developerComputing platformOpen setIntegrated development environmentEndliche ModelltheorieGoodness of fitComputerInterface (computing)Operator (mathematics)Structural loadData managementFunctional (mathematics)Variety (linguistics)Condition numberFilm editingRange (statistics)2 (number)Computer animation
16:16
Local GroupCategory of beingComputing platformOpen setLevel (video gaming)Digital photographyNormal (geometry)Parameter (computer programming)Functional (mathematics)Computer architectureOnline helpCASE <Informatik>Variety (linguistics)MetreIntegrated development environmentContext awarenessComputer animation
17:03
Computing platformOpen setHill differential equationArtistic renderingArtistic renderingCoprocessorCategory of beingOnline helpGroup actionAreaPower (physics)Frame problemDifferent (Kate Ryan album)Single-precision floating-point format19 (number)2 (number)Social classPoint (geometry)System callMultiplication signSound effectView (database)DemosceneComputer animation
17:52
Software developerComputing platformOpen setCodeLine (geometry)Java appletMobile appCASE <Informatik>Endliche ModelltheorieCodeLine (geometry)Software developerMedical imagingState of matterComplex (psychology)Wave packetLetterpress printing
18:49
Computing platformProcess (computing)Broadcasting (networking)Digital photographyOpen setSoftware developerCodeLine (geometry)Java appletOptical character recognitionRippingControl flowElectronic data interchangeComputer-generated imageryEstimationVideoconferencingRing (mathematics)Integrated development environmentCASE <Informatik>Complex (psychology)Social classShared memorySpeech synthesisDifferent (Kate Ryan album)Digital photographyArithmetic meanCategory of beingCollaborationismPattern recognitionSoftware testingTap (transformer)Translation (relic)Asynchronous Transfer ModeComputerClique-widthComputer animation
20:26
Open setComputing platformParameter (computer programming)Interface (computing)Menu (computing)Hill differential equationServer (computing)Task (computing)Arithmetic meanComputing platformOffice suiteInformationTerm (mathematics)Tablet computerProduct (business)Web pageComputer animation
21:08
Computing platformService (economics)Software developerOpen setCategory of beingAverageSoftware developerCategory of beingCovering spaceMobile appService (economics)Goodness of fitMultiplicationPredictabilityAngleScaling (geometry)Translation (relic)AreaSocial classAverageECosArithmetic meanResultantComputer animation
22:52
Perfect groupDisintegrationAndroid (robot)Integrated development environmentComputing platformSoftware developerEndliche ModelltheorieMobile appOpen setIntegrated development environmentINTEGRALStack (abstract data type)AuthorizationService (economics)Functional (mathematics)Data managementAndroid (robot)2 (number)Multiplication signSystem callMereologyPower (physics)SurfaceEntire functionSystem identificationSoftware developerComputer animation
23:45
DisintegrationData modelProcess (computing)Computing platformOpen setSoftware developerEndliche ModelltheorieSoftware developerIntegrated development environmentLevel (video gaming)Core dumpProcess (computing)INTEGRALFunctional (mathematics)Endliche ModelltheorieMobile appMereologyTunisRight angleMultiplication signType theoryService (economics)Software testingOperator (mathematics)Cycle (graph theory)Video gameComputer hardwareWave packetGreatest elementComputer animation
25:36
Software developerInternet service providerService (economics)Android (robot)Computing platformHill differential equationPerfect groupIntrusion detection systemDisintegrationOpen setSoftware developerProcess (computing)Product (business)Integrated development environmentArchaeological field surveyParadoxCategory of beingMobile appHome pageMereologyAddition2 (number)Computer animation
26:47
Computing platformOpen setHill differential equationSoftware developerData storage devicePhysicalismHand fanBoolean algebraPlanningMereologyVideoconferencingDigital photographySmartphoneLevel (video gaming)Sound effectComputer virusComputer animation
28:19
Computing platformOpen setArtificial neural networkFilter <Stochastik>Multiplication signSmartphoneGoodness of fitPoint cloudComputer animation
29:05
Data modelProcess (computing)Computing platformOpen setSet (mathematics)Computer hardwareSmartphoneEndliche ModelltheorieAndroid (robot)ResultantVery-high-bit-rate digital subscriber lineMultiplication signHeat transferProcess (computing)SoftwareProjective planeDigital photographyBefehlsprozessorCartesian coordinate systemArtificial neural networkComputer animation
31:01
Mobile appMobile WebOffice suiteComputing platformOpen setLocal GroupProduct (business)Web pageInformation technology consultingGame theoryOffice suiteComputer animation
31:50
Suite (music)Office suiteComputing platformOpen setOffice suiteComputerTablet computerFreewareProbability density functionPoint cloudMultiplication signTerm (mathematics)Point (geometry)MereologyComputer animation
32:31
Mobile WebTotal S.A.Computing platformOpen setPowerPointComputer-generated imageryRotationPresentation of a groupInstallation artCloud computingPoint (geometry)Digital photographyKeyboard shortcutOnline helpMobile appPresentation of a groupOffice suiteMotion captureContent (media)Mobile WebPRINCE2Lattice (order)Computer animation
34:01
Interface (computing)Computing platformOpen setSoftware testingINTEGRALDigital photographyProcedural programmingArithmetic progressionInterpreter (computing)DataflowMultiplication signComputer animation
34:38
Computing platformOpen setAreaMultiplication signComputer fileMachine learningProfil (magazine)Virtual machineComputer animation
35:07
Computing platformOpen setIntegrated development environmentEmailEntire functionComputer animation
35:36
Computing platformOpen setMultiplication signScaling (geometry)Series (mathematics)Software developerDigital photographyOpen setClosed setComputer animation
36:08
Computing platformOpen setHill differential equationOpen setPosition operatorParameter (computer programming)Computing platformComputer animation
36:38
Computing platformData modelAlgorithmIntegrated development environmentOpen setSoftware developerEndliche ModelltheorieAlgorithmSystem callIntegrated development environmentComputing platformComputer animation
Transcript: English(auto-generated)
00:10
Okay, my name is Shandong Dui, it works? No, it works.
00:22
I'm now AI Chief Scientist of Huawei. Huawei is, I think, one of the biggest telecommunications companies in the world, with around 180,000 employers. And in my team, there are around, I can't disclose the number, but it's hundreds of best talents,
00:52
many of them are PhD, working on designing, implementing and realizing cutting-edge AI models for different purposes,
01:04
mainly focusing on computer vision and natural language processing. But also, to some extent, the decision engine, the recommendation engine and the search engine as well. It works now? Okay, great.
01:24
Well, so, as known to all of us, wait a minute, AI technology is processing very fast, we can observe a gap, however, we can observe a gap on the one hand,
01:41
on the one hand, there are a few smart mobile providers, such as Huawei, keep developing more and more AI capabilities onto the device. However, although we are a very big company, we are still not able to cover all the needs of hundreds of millions of users.
02:07
But on the other hand, there are many developers of the communities, and they have a very great idea about limited skills to leverage all the AI capabilities of the mobile phone.
02:22
So we should make our ecosystem partners focus on their own specific AI applications, fitting the targeted market, and where we, the platform provider,
02:41
we develop platform with excellent AI models in the heart, but making it easy to use, hiding all the technical complexity, in particular, related to the designing of algorithms.
03:02
Well, so summarizing all the problems we have been faced with related to AI in recent years, and it's a common problem, AI talents are scarce resources, and as I said at the beginning, Huawei has hundreds in my department,
03:27
and thousands of overall best AI talents in the world, but still, we also believe we belong to the first league of the company, but we are facing a very particular domain,
03:46
it's about bringing all the AI intelligence onto the device, so as to the constraints on the device, there are still very few research work done in this area,
04:01
so this is exactly our focus of research and also strengths and advancement. The first constraints we have been faced with is insufficient resources. As you know, a device has very limited resources in computational power,
04:23
energy consumption, data modeling, data management, etc. We don't have the luxury of the resources on the cloud. The second constraint is low efficiencies in development, and developing any AI capability on mobile devices,
04:41
I think the whole life cycle is not as smooth as on the cloud, so they are lacking a comprehensive, effective and efficient IDE, so the development cycle of any AI model for devices is, until today,
05:01
still very painstaking. So, yes, starting from collecting the training data, and first designing and training the model on the cloud, and then, because normally the model after being trained has hundreds of megabytes, if not over gigabytes,
05:23
so we need to minimize the model size, without great loss of the accuracy, then we download the model to the device and testing, and of course, in the first rounds of iterations, the performance and accuracies are not normally sufficient,
05:43
so we have to do several rounds of iteration, normally it will take much effort. So the process as of today is broken, and still very time consuming and inefficient.
06:02
And the third constraint we have been faced with in terms of AI intelligence on the device is fragmented capability. In the area of mobile intelligence, there are still many tiny and siloed models, and fragmented also in the user experience.
06:22
Many models are similar, mostly redundant, but they lack a structured and systematic approach in terms of unified and coherent hardware and software stake, architecture, and also well organized training and testing data,
06:40
and management, and also standard testing method and criteria, and all of that has been more or less matured on the cloud, but not in the world of mobile intelligence. To that end, the right collaboration of model development lifecycle
07:00
among the devices, among the telecommunication channel, and also between cloud and devices need to be addressed. So the last constraint we have been faced with is a difficult improvement and upgrade problem.
07:26
So far, as mentioned before, the AI model is improving very fast. A model needs hundreds of iterations to become mature. So ideally, it should be decoupled from the normal upgrade cycle,
07:42
such as for the operating system. And the upgrade cycle for operating systems in Android world normally takes much longer time than the upgrade cycle needed to be required by AI model. So we need to decouple both cycles.
08:03
So the solution of Huawei to the challenge mentioned above is first, we design and develop our own AI chip. I think we belong to the first one, coming out with, in my belief, the best AI chips,
08:20
optimized for the intelligence computation on the device. So apart from our own device, our own AI chip, based on this chip, we also develop our AI engine as well as AI service layer and opening up to the ecosystem.
08:52
In a nutshell, our architecture of high AI can be divided into four parts.
09:00
So they are from bottom-up, Huawei high AI foundation, then up to the mid-layer, Huawei high AI engine, Huawei high AI services, and we have also developed an end-to-end IDE, Integrated Development Environment.
09:22
So talking about high AI foundation, it's corresponding to the chipset on the bottom. To the chipset, converting and migrating existing models to obtain an optimal performance using heterogeneous scheduling and MPU acceleration.
09:45
MPU is our dragon. MPU is a specific process architecture specially designed for the process of tensor. And in the mid-layer, we have Huawei high AI engine
10:01
and corresponding to the device level with emphasis on simplicity. So it's about opening up our AI app capability able to integrate and compose different types of AI component into more comprehensive intelligent apps.
10:25
So the upper layer, Huawei high AI service, is about putting emphasis on achieving a win situation together with our partner. So it's about proactively pushing the service onto our users based on our users' needs.
10:42
So preparing and enabling services beforehand and based on our prediction model on the device. So looking more detail of our high AI architecture, so again, starting from the bottom,
11:02
Huawei high AI foundation, comprised of CPU, GPU, DSP, and MPU. MPU is, as I said, our product is also the core of our processor, which specially designed for the processing of tensor computation.
11:24
And normally with 50 times speed and also from end-to-end scenario, normally 25 times. And moving upwards, so we have DDK, it's our environment, it's our library,
11:42
and also runtime environment for offline mode. And also, of course, we support Android NN for online mode and also for its runtime engine. And with this setup, we are able to support a large variety of our frameworks. It's colored in red, such as the most common one,
12:02
TensorFlow, TensorFlow Lite, Caffe, et cetera. We're also actively participating in different open source communities, such as ONIX and TVM. So we are not only serving and supporting the most commonly used framework,
12:22
we're also actively involved and also influenced in the open source community. So in so doing, to make sure that our high-air platform will be sustainable and also open, have a more broad basis for openness and generalization.
12:44
So moving up to high-air engine, engine is just any physical engine of the car. The design of the engine, per se, is normally very technology-intensive, right? We need thousands of base engineers for the design of an engine.
13:01
But the operation and handling of the engine should be very simple, right? And also highly automated. So as a driver, you don't need to know how the engine was built. So, as I said, we have hundreds, thousands of AI experts
13:21
developing the core capability of cutting edge algorithms. But on the other hand, we encapsulate and hide all the complexity into our AI engine and making it easy to consume and also being robust. So at present, our engine is focused on CV, computer vision,
13:45
focused on ASR, which is for the process of speech, and NLP is for the process of text. And so far, we already have released 20-plus different types of AI API and 19-plus operators.
14:01
I think our opposite support is support most commonly used operators, such as convolution, deconvolution, LCM, RNN will be covered. So along the whole stack, from bottom to the top,
14:21
we also, on the right-hand side, we offer a very comprehensive toolkit, IDE, which can make the development work much more natural, intuitive, and easier. So all four parts, I mean the high-end foundation,
14:41
higher engine, higher service, and higher IDE build the core and the soul of our own device intelligence. Now, in more detail about high-end foundation,
15:02
in light of the benefit of our platform, I think we offer three features to our developer community. The first one is real-time. So our foundation is designed
15:24
for providing acceleration APIs, regardless of processor, hardware, so as to accelerate the models, computation, as well as operators' computation on the mobile devices. The second aspect, or the second feature, sorry,
15:40
the second feature is efficiency. I think even on the small device, we offer a very good AI model management API, ranging from model compilation, model loading, model running, and model unloading. And also, the third feature of our AI foundation is agile.
16:04
Agile is about providing basic operator computing API, including interfaces such as convolution, deconvolution, pooling, full-connection, RNN, LCM, and also a large variety of activation functions.
16:22
So just an example of our use case in the layer of our foundation, and with the help of our own chip architecture MPU, our camera becomes much more intelligent.
16:43
It enables our mobile phone to sense the environment, and so as to adjust the parameters of the camera automatically and also in a professional way. And so doing the normal users of Huawei mobile phone
17:01
suddenly becomes a professional photographer. So far, we support 500-plus scenarios in 19 different categories. And another example, also in the area of our high-ear foundation, just taking a Prisma scenario as an example
17:22
with the help of our processor MPU, we have achieved 50 times increase in performance. Effect ratio, 20 times increase in efficiency. And on the left-hand side is an example of the partner Prisma that achieved 2.5 seconds for a single frame rendering
17:43
and the performance of a style suggestion enabled by our Huawei AI Engine's API called scene detection has been improved by 300%. It's really amazing.
18:00
Now, further move on. Now, moving on to the middle layer, talking about high-ear engine. And just taking an example of easy-to-use capability, the case shown here is image aesthetic scoring.
18:21
An app developer just needs to insert two lines of code to utilize this model. So you can see on the right-hand side, only very simple two lines of code. And on average, on average, developing any app based on our AI Engine,
18:41
by the way, AI Engine, all the models are already been pre-trained. So it takes 10 to 120 minutes, end-to-end, of course, depending on the complexity of the use cases. Furthermore, so here's a nutshell of our AI Engine offering.
19:03
Clustered so far... Sorry. Okay. Okay, let's move on to this one. It's a nutshell of our AI Engine
19:22
clustered in six different categories, such as translation by taking photo, and voice-controlled photo taking, just a few to be mentioned. And of course, this offering is expanding in both width and depth dramatically, and we are working on that at full speed.
19:44
And mainly, again, as I said, mainly focused in computer vision, ASR for speech recognition, NLP for text recognition, and also for semantic understanding of the text. So supporting both online mode,
20:01
online mode such as an API, and also offline mode, our own offline mode. And offline mode is designed for situation without telecommunication, without any Wi-Fi, so I think we achieved both.
20:22
Online, meaning proactively collaboration with the intelligence of the cloud, and offline, meaning solely rely on the intelligence of the device. So here's another excellent example of our partner, Kinsoft.
20:41
It's about integrating the office product, WPS capabilities with our AI platform. It took only half day to complete this task. More information will be delivered later.
21:02
So here's a webpage of our AI platform. Any user can download SDK and other kinds of tools from this webpage, so we are not just talking. We work the talk. Here's an overview about the size and scale of our community.
21:24
So far we have 350,000 registered developers worldwide, and users, we have 340 million users. So this is a very large community,
21:43
and we can benefit from that. Huawei's phone shipment so far in 2017 was 153 million, and the revenue sharing, I mean we share the revenue collected with our developer
22:04
was 45 million US dollars. So you can see it's a win-win community in multiple angles. And the offering to our community is providing one-step services for most apps.
22:24
In so doing, the handling of the apps becomes much faster, easier, intuitive, and intelligent due to using more prediction capability of our high-air engine. So you can see we have very good coverage,
22:40
cover 50 plus categories of services, and also cover 80% most commonly used scenarios, and we can save on average three steps, right? Three steps on average, and translating in times, we can save 5 to 10 seconds,
23:00
which is very significant for each individual mobile users. Talking about IDE, this is a perfect integration of a Huawei high-air IDE and Android Studio. So it's compensating each other and comprising developing the typical lifecycle needed by developer.
23:25
Develop, debug, test, help, identification, and authorization management functionality. And also covering end-to-end the entire software stack. I mean the three layers I mentioned above,
23:41
Huawei foundation, Huawei high-air engine, and Huawei high-air services. So this full part built the soul and the core of our high-air capability on the device. So looking into more details, integration efforts for Huawei high-air engine
24:02
has been reduced to our level. And of course, our development can also directly use the functionality of foundation, and the integration effort with foundation can be reduced to the day level. So in the middle layer to our level,
24:24
in the bottom layer to the day level. So on the left-hand side is our IDE. So you can see our IDE is designed for two types of developers. On the left-hand side is an IDE process
24:42
for developers without self-owned models, just using our AI API as an apps layer based on our services requirement. As I said, all the models have already been pre-trained and also ready to use, simply use our API.
25:01
And there is no effort required for model development, training, testing, and deployment. So this is for one type of user. Of course, our IDE is also designed for another type of user. On the right-hand side, it's about a developer with self-owned models.
25:21
So a comprehensive IDE process covering the whole DevOps cycle, end-to-end. So there are so many items in the middle. I think you can see eight items. I don't need to go through every item. This process covers a typical lifecycle of DevOps.
25:47
In summary, four real and tangible products offer to our developer community. Foundation, engine, service, and IDE.
26:03
They are well-designed in coherence and also supporting each other. And one more thing. We have more to offer to our developer in this community. And we have around all four parts I mentioned before.
26:27
We also offer apps gallery with a homepage recommendation in app gallery and AI category. And the second offering, additional offering, is recommendations through Huawei official channels,
26:43
such as a spatial column recommendation through various Huawei official social network accounts and sticky on Huawei fan club. And the third offering, additional offering, is a Huawei store recommendation.
27:02
It's about promotion in Huawei's physical store. So far, we have 10,000 physical stores. Last but not least, joint marketing. Our partner can join PR, planning, and promotion. So next, I'd like to welcome Paul to the stage presenting Prisma.
27:37
Hello. Oh, okay.
27:41
So hello, my name is Paul. And nice to see you all here. Thank you, Mr. Zhu, for presenting our company. And we are very happy to be here at Droidcon in Berlin. And in Prisma Labs, we're working hard to create AI-powered effects
28:00
for photo and video and enable new opportunities to users to create very amazing pictures with their smartphones. So let me start with a question. How many of you use your smartphone for taking pictures? Raise your hand, please. Yeah, almost all of you.
28:21
Good. So people take more and more pictures with their smartphone. And of course, I think, oh, it's next. And I guess everyone wants to, their pictures looks great. But our goal is to improve user experience with a smartphone camera
28:47
using our AI capabilities and our experiences on running neural networks on a device, on a smartphone without any cloud-based solution. To allow people to avoid spending time choosing good filters
29:06
or appropriate settings to their pictures to look great. But just take pictures and it automatically looks great. So this is our goal. And as our partnership project with Huawei,
29:27
we increase speed of our flagship technology style transfer that's used in our application called Prisma. Because speed of processing is extremely important
29:40
because people don't want to wait a lot of time while picture is processing. And in Prisma, we use different approaches to make our neural networks fast and accurate at the same time. But if hardware is also optimized for such computations,
30:00
it could give you really, really good results. And we are very happy that we can increase our AI models using NPU acceleration from Huawei. And we accelerate our style transfer technology using this NPU.
30:24
And we improve speed and achieve high speed on NPU compared to usual CPU on Android. And also we decrease the model size by up to 25 times. So we really value our partnership with Huawei.
30:42
And we are sure that with mutual efforts, we can achieve very big results in a smartphone camera user experience and enable people to just shoot and to be a professional photographer. Thank you and back to Mr. Joo.
31:06
Thank you, Paul. Next, let's welcome Lina to the stage. Hi, everyone. I'm Lina from Kingsoft in Beijing.
31:22
It's a great pleasure to meet you here. First, please allow me to introduce our company and our product. Is there anyone who uses WPS office? Okay, thank you. Let me introduce it for you. Kingsoft was founded in 1988 and listed in Hong Kong in 2007.
31:45
It has four subsidiary companies, Kingsoft Office, Chetan Mobile, Kingsoft Cloud and CSUN Games. Our offices are located in China and the United States.
32:01
WPS office is one of the most popular all-in-one office suite. With WPS, you can create, read, edit all kinds of files, include Word, Excel, PowerPoint and PDF. You can share and access them anywhere and anytime.
32:22
You can use WPS free on all your devices, your Android phone, your iPhone, your tablet and your computer. All your documents could be saved and synchronized across all your devices with our cloud service. There are over 1.25 billion installs and more than 250 million monthly active users.
32:51
WPS dominates China market and we are expanding our global footprint and increasing international brand awareness.
33:03
As what we are talking about here today is how AI is used in our WPS office and how AI is redefining mobile applications. Just like today, you come here during the keynote and you want to share the presentation with your team.
33:22
What will you do? Now you can finish it easily with the help of HAI AI and WPS office. HAI AI and WPS office can help you take notes such as meeting minutes, lecture notes without handwriting and keyboarding.
33:41
What you need to do is to take out your mobile, take photos, choose photos which you want to add in your presentation. HAI AI and WPS will recognize the main content in each of your photos, capture and clean the images and transform them into a PowerPoint presentation.
34:04
Then you can send it to anyone you want. Yes, AI makes it easy. You couldn't imagine that how fast the integration progress is. It took us only half a day to deal with the HAI AI APIs
34:22
to make the main procedure work. And we spent several days and took a lot of photos to test and optimize the working flow to improve the user experience. After we created this feature, users opened more files and spent more time in WPS.
34:48
We think that AI and machine learning will explore and unlock many interesting areas. We must keep learning and get ready for the upcoming challenges. That's why we cooperate with HAI AI.
35:01
During our cooperation, HAI AI helps us a lot with its professional AI skills. It builds a favorable ecological environment and provides all-round technical support. Now we keep working on creating more intelligent features with HAI AI.
35:24
We hope that you could join us. Let us work together to get AI to help everyone. Here is my email. For any questions and cooperation, please feel free to contact me. Thanks for your time.
35:42
Thank you, Lena. So, let me do the closing. So, in summary, we offer four aspects. HAI AI Foundation, HAI AI Engine, HAI AI Service, and HAI AI IDE
36:01
in a full scale of both width and depth, as well as a total openness to our developer community. Our ultimate goal is fast, simple, and win. So, we have more sessions to offer. Today, from 14.05 to 15.30, we have our colleagues Yang Jin and Yingwen Shui
36:34
talking about HAI AI Foundation open platform, as well as HAI AI Engine open platform
36:41
in more technical detail, and the venue will be Jelly Bean. And next, also tomorrow, from 14.05 to 15.30, we have another colleague, Dong Xin, talking about how to transplant the AI algorithm model to HAI AI platform,
37:03
as well as HAI AI platform IDE itself, so also in the venue of Jelly Bean. Well, that is pretty much what I'd like to share with you today. Thank you.