 Thank you, Arbit. Good morning, my friends. I'm Lillian from Vital Cloud. I'm very happy to meet you at the O&E at Chrome today. Well, unfortunately, we can only communicate online. For the past three years, we have been separated by the pandemic, unable to see our colleagues, friends, and even families confronting such difficulties. Digital technology is particularly important. In fact, our world is being surrounded by ubiquitous computing. There are countless data generated every moment, and this continuous data needs to be processed in timely and proper manner. This data not only requires abundant computer power to possess, but also need a variety of computing methods to meet the requirements of the scene. Today, we are familiar with the cloud, which brought fundamentally changes to the information technology in both economics and agility. But beyond the large data center, there are more data that need to be processed, such as a variable video conferences, urban management, and a smart hardware in-home, which requires lower latency, cheaper bandwidth, and higher privacy protection. To solve this, we need edge computing. Since 2017, Baidu has been investing the development of edge computing technology, continuously enriching our enterprise service matrix and providing developers and customers with better products. We believe that edge computing and cloud computing are highly complementary technologies, and the two should cooperate with each other to meet today's increasingly complex digital needs. So we proposed the concept of edge cloud fusion. The concept means to develop a test and manage the application on the cloud and perform actual data processing and device control at the edge. Applications and data are transported between them, thus forming a live system, a positive cycle of efficiency. Following the concept of edge cloud fusion, Baidu has established a commercial edge computing product called Baidu Intelli Edge, short as BIE. And in BIE, users can collect the data from different kinds of terminal devices, your local network, process data in edge devices by several different kinds of computing, including small chipset like a Raspberry Pi or dedicated PC servers, and control all of them in the cloud. There are four key rules we share in the Baidu Intelli Edge system. The fusion to combine the cloud and edge are unified to unify all kinds of different hardware, software and network. To keep it open, open to all users, protocols and standards and practical, we provide a lot of different practical solutions. Following the... So we believe that it is impossible to promote the development of edge cloud computing by just one company, but by the power of more individuals and organizations, by the power of the community. So we refine the core technology in Baidu Intelli Edge into an independent open source product and join the newly established LF Edge in 2019, which is the Beetle product. Beetle, it's committed to provide a unified environment, management and a security model for all edge computing to become the link between the edge and the cloud. Next, I'd like to share the technologies and innovations of Baidu and Beetle. The first version of Beetle 1.0 adopted a containerized design. It leveraged the Docker and a series of function containers. You can find that the Beetle master program runs in the host environment and it received the remote configurations, save it persistently and command the Docker system. And then a series of function containers were launched by the Docker team and provide services. So by using this version, the initial version of Beetle terminal devices could communicate with each other in the lotto network by sending and receiving messages. And data could be processed in real time and synchronized between cloud and the edge. In the first version, we got a lot of features to Beetle and users can write Python scripts to process the data and even use TensorFlow to do AI inferences. And in this version, we support four kinds of mainstream hardware and Linux, Windows and Mac OS, the mainstream operating systems. In June 2020, a year after joining LFM, Beetle 2.0 released with Kubernetes support. By this graph, you can find that all your configuration, including the applications, device definitions and your data and the models is collected in the cloud and packed together into a data pack and send it to the remote ad instances. In the app side, the Beetle will unpack the data and convert it to the Kubernetes command and make the local Kubernetes cluster to apply your services. And even more, by this version, before this, all those developers could deploy multiple ad instances in different locations. But each instance was limited to only one computing node. In 2.0, an instance can be composed of multiple computing nodes as a local ad cluster and with the local load balancing and the fillover and the share storage and network between them. In this way, ad computing can enter key screen hours like energy and electricity power. More importantly, Beetle 2.0 reflects our thinking on open economy and vendor neutrality. Also, there's a lot of different ad computing systems today to support Kubernetes, but a lot of them choose to alter the internal protocol of Kubernetes and even rewrite the underlying key services, such as Kubernetes, which results in that users being limited to a specified Kubernetes vendor or a specified Kubernetes version. And by this way, altering Kubernetes means you need to stimulate ad instances as Kubernetes worker nodes. That is, it can difficult to support ad pairs and the limited capabilities of Kubernetes in the app. But Beetle choose a different way. We believe that users should be allowed to freely choose Kubernetes vendors and deployment models according their actual conditions. Then they use the lightweight Kubernetes version like a Rancher K3S to save the power or use a full-power multi-node system from the standard Kubernetes community or from the commercial services. Therefore, we make cloud and edge as equal. In this picture, you can find that every edge instance is an independent cluster. You may choose the single node cluster or multiple cluster or even cluster in some other infrastructures. And all this connected together by a cluster to cluster architecture. In theory, this mode can bring arbitrary Kubernetes functions to the app, even including third party extensions like the CRDs. That is impossible today and that will be true in the next major version of Beetle. One important feature of by-do-in-tally edge is AI integration. And it's common to do AI inferences on the edge but there is always be a real issue of edge AI is that the models are often trained by frameworks and chips at different than the edge for inferences. They're just incompetent. So we introduced an automatic model conversion matching the corresponding model for the AI hardware reported from Beetle. In one cases, a model was trained on the cloud using the Palo Palo framework on the by-do-in-tally edge chipset which was then converted to the open-vino framework and applied to an edge instances with an interwovenous chipset. By the way, this technology is part of the computer vision solution in BIE. In this solution, the edge devices can quickly become an intelligent vision system. The user just needs to connect the camera no matter the other network or USB and plug in the AI chipset like Movidus and push the power button. It would automatically download the software, capture videos and sampling the images to AI inferences and send the results out. This solution has been applied by dozens of customers in China and is used in important occasions such as safety construction, fire detection and environmental protection. These innovations have brought infinite possibilities to edge computing. Next, I'd like to show more collaborations and application cases. The first thing is the collaboration between the Beetle community and AdX Foundry community. AdX is committed to provide an open-source IoT solution on the edge, but it requires all the configurations to be applied on the edge side. By using Beetle, all these configurations can be switched from the edge side to the cloud side and do automatically deployment. That will significantly reduce the complexity of operation and maintenance. In the last Beetle co-host, the AdX challenge China hacked them with Beetle special award. The UN saw a lot of creative projects combining Beetle, AdX and AI technologies and I'm really looking forward to them continuing to grow and major. The next case is a smart grid. China has the world's largest power network and probably the most complicated one. It not only requires to rely efficiently transmission and stable supply of electricity in the land of millions of square kilometers, but also support unstable energy sources such as power and solar and reduce the overall carbon emission. This means that we need to achieve a fast and refined power dispatching in a very large range. Baidu has given a plan to combine artificial intelligence and edge computing. In the cloud, we provide a full functional AI training system allowed to train and test the various power distribution models. In the meantime, edge computing is deployed in every area, receiving the issued models and control strategies connecting to the local sensors, drones and power equipment to collect data and achieve precise control. That is a smart grid. Another case is industrial inspection. Nowadays, using a computer system for automatically quality inspection is become a common method, but it still faces a series of problems. It is slow to upgrade models and hard to upgrade models with false and missing data. In response to these problems, Baidu has proposed an edge cloud fusion solution. We deploy a computer vision system by Baidu IntelliEd next to the production line. The system can automatically de-sensitize the error data and upload it to the cloud. The iterated model can then be sent back to the edge remotely and do a more precise and accurate detection. This makes a reducing of model deployment time from one day to one minute. And that means you got a smart factory. The last case is autonomous driving. To achieve an L4-level autonomous driving, we need to not only intelligent vehicles, but also intelligent roads. If we could detect the problems on the road early, we can have the vehicles to make a more accurate judgment. We cooperate with the Baidu Autonomous Driving Team to launch a V2X solution. In this solution, we start the cameras and series of sensors on the roadside and pass the information to the nearest NEC computing node through the 5G network. The node is set up with V2 and the road recognition algorithm. The recognition is without them be sent to the nearby vehicles and transmitted to the regional computing center for global optimization. This solution can reduce the cost of autonomous driving by 60%. Finally, I'd like to show the future of V2X there. In the next major version upgrade, we will integrate more deeply with Kubernetes, deliver all the Kubernetes capabilities to the edge and manage both the cloud workloads and edge workloads in the same control plane. Some features may appear in our June version upgrade, so keep turning. And if you are interested in V2X, welcome to join our community. You can get all the source code on GitHub and using the main list, betel-at-list.lfh.org and visit iot.bydo.com to get more practical pieces. And you can also scan the QR code on screen to join our Chinese VCHAT group. So that is all I'd like to share with you today. Thank you. Excellent. Thank you very much. It has definitely been great progress, right? Especially seeing the grid use case, the V2X use case. Manufacturing was already one of the early use cases, but I think we are really excited to see the progress you have made both in the open source community as betel and as bydo. So congratulations. If there is any questions from the audience, feel free to ask on the Q&A. I do have one simple question and we may have a couple of minutes here, which is... Yeah. So as you see, as you see the community kind of maturing and adapting to these use cases, how are you going to expand the number of use cases that you are supporting as services because edge computing has like almost 20 to 100 use cases, right? All over the place. And we are focusing on a few segments like manufacturing or automobiles or things like that. Where do you see us 2022 focus from a bydo perspective? Well, in 2022, bydo will try to focus on the energy systems. China is committed to the greater carbon emission peak and make our growing world. So we do a lot of focus on the energy system to collect data from our factories, our generations and more systems to more precisely, it precisely got the real energy usage and the real carbon emission. And that will help the industries, the producers, industries and the government to monitor the steps. In the future. And by having this collected data, we may develop an algorithm to reduce the carbon emission. That is what we are going to do in these years. And also there's a lot of different situations that will continue to support the combination of the community power and the commercial power to push more development in the actually. Oh, this is brilliant. I'm personally so excited that edge computing can play a big role in energy and all. So again, thank you for the leadership. I think we're out of time. So let me thank you again for the keynote and really appreciate it. Thank you, Yopin. Bye-bye.