Last night, the annual Microsoft Build Developer Conference kicked off. As usual, the conference’s stage has moved online from Seattle in the Us. Unsurprisingly, tonight’s conference will be opened with a keynote speech by Satya Nadella, Microsoft’s current CEO. In just over 20 minutes of speech, keywords such as “developer,” “Azure,” “Microsoft 365, ” and “Windows” are all over the place, especially “developer.”
Look at the point: Microsoft smashed $1 billion to support the super-calculation came, a text to see last night’s Build Conference ten points!
He says there are now more than 50 million developers on GitHub and more than 3.5 million on Power Platform. Microsoft will also provide developers with a range of development tools and solutions to increase productivity.
The core highlights of this Microsoft Build conference around developers are as follows:
1, announced the development of acooperative aI supercomputer with OpenAI, dedicated to training large-scale distributed AI model.
2. Launch Microsoft Cloud for Healthcare for the healthcare industry to improve collaboration, decision-making, and operational efficiency for healthcare teams.
3, the launch of the AI development platform for industrial systems Project Bonsai, as well as The Project Moab experimental platform.
4, the launch of Windows developer platform Project Reunion, unified UWP and Win32’s existing APIs.
5, upgrade DeepSpeed library, mainly for the field of deep learning, and open source history of the largest language model – Microsoft Turing model.
6. Release Azure Synapse Link for real-time operational data analysis.
7, WSL 2 new features, including support for GPUs, Linux GUI applications, and simplified installation experience.
8, the launch of the Microsoft Teams platform feature updates, including a streamlined experience for developers.
9, open source and upgrade Fluid Framework, providing Fluid components and Fluid workspaces.
10. Provide responsible machine learning tools to reduce inequality.
In addition, Microsoft has launched the Windows Package Manager, a command-line interface that helps developers quickly search, view, and install a variety of management tools. At the same time, Microsoft Chromic kernel browser Edge has also added sidebar search, synchronization and other new features.
Notably, Microsoft also announced the acquisition of Softomotive, a maker of RoboticS Process Automation (RPA), to introduce RPA technology to Power Automate.
AI supercomputer unveiled for large-scale distributed AI model training
In July 2019, Microsoft announced a $1 billion investment in OpenAI, an artificial intelligence research lab, to jointly build a new Azure AI supercomputing platform that will primarily focus on training and running more advanced AI models, including Microsoft Azure AI overcomputing technology. The investment will also further help OpenAI develop AGI (General Artificial Intelligence) technology.
After a year, the supercomputer finally debuted tonight, mainly for large-scale distributed AI model training.
Microsoft claims that the supercomputer is in the top five in the world, with 285,000 CPU cores and 10,000 GPUs, each with a bandwidth of 400Gb/s.
Hosted in Azure, this supercomputer has been able to enable a range of modern cloud infrastructure capabilities, including rapid deployment, sustainable data centers, and access to all Azure services.
The supercomputer has been able to train large-scale AI models to gain insight into the nuances of language grammar, knowledge concepts, and contextual content. It can also summarize lengthy conversations, moderately engage in real-time games, parse complex legal files, and even generate code by searching GitHub.
Microsoft now uses the Turing model to improve language comprehension for Bing, Office, Dynamics, and other productivity products. For example, in Bing, the model generates text and answers questions faster by 125 percent.
Microsoft says it will further unleash large-scale AI models, training optimization tools and supercomputer resources through Azure AI services and GitHub, making it easy for developers, data scientists and business customers to harness the power of AI.
Microsoft Cloud for Healthcare: Improving Healthcare Efficiency
In Microsoft’s view, this year’s pandemic of the new coronavirus has affected almost every aspect of people’s lives, but also hindered the normal operation of the health care business, greatly reducing the efficiency of patient treatment and care.
In response, Microsoft Cloud for Healthcare, its first industry-specific cloud product, supports a new Bookings app in Microsoft Teams that enables caregivers to schedule, manage, and conduct virtual patient visits in Teams, and provide services that further enhance patient communication, care team collaboration, and management efficiency.
At the same time, users can use Microsoft Cloud for Healthcare to extend the value of Microsoft Dynamics 365 Marketing, Dynamics 365 Customer Service, and Azure IoT for patient experience, physician referral management, patient testing, and more.
Of these, microsoft For Healthcare Bot, based on the Microsoft Cloud for Healthcare expansion, has had more than 1,600 instances of COVID-19 robots in service since March this year, serving 31 million people in 23 countries, further reducing the emergency hotline pressure on the cdc.
Industrial Systems AI Development Platform Project Bonsai
Simply put, Project Bonsai is an AI platform for building autonomous industrial control systems and a “machine teaching” service that combines machine learning, calibration and optimization capabilities to enable autonomous core control systems for machinery in industries such as manufacturing, chemicals, construction, energy and mining to better manage all types of industrial equipment.
Project Bonsai’s universal intensive learning platform coordinates the development of AI models, provides access to algorithms and infrastructure for the deployment and training of AI models, allows models to be deployed on-premises, devices, or the cloud, and supports simulators such as MATLAB Simulink, Transys, Gazebo, and AnyLogic.
In addition, users can view all work and training statuses on Project Bonsai’s dashboard, debug, inspect, and improve the model. Project Bonsai also supports multi-user collaboration to build and deploy new models.
At the same time, Microsoft has developed a hardware device called Project Moab for engineers and developers who want to try Project Bonsai.
It is a robot with three arms and a joystick controller that balances a small ball on the transparent plate at the top. The tool provides users with an analog environment that allows them to experiment with a simulator.
Project Reunion: Unifying The Windows App Eco Platform
Microsoft’s main goal is to unify Windows’ application eco-platform, enabling developers to move away from the limitations of different operating systems and easily achieve cross-platform development by integrating the existing APIs of Universal Windows Platform (UWP) and Win32.
Microsoft will add more common APIs and interoperable code between the two. This will provide a common platform for applications to help users use the latest feature updates, as well as modernize existing applications, whether in C, . NET (including WPF, Windows Forms, UWP), or React Native.
In addition, Microsoft has introduced Windows Terminal 1.0 for enterprise users, which enables users to run any command-line executable on multiple tabs and panes, including the WSL release and Azure Cloud Shell.
DeepSpeed Library Upgrade, Microsoft Turing Model Open Source
Back in February, Microsoft released an open source version of DeepSpeed and launched ZeRO (Zero Redundancy Optimizer).
DeepSpeed is primarily designed for distributed model training across multiple servers, and ZeRO is a technique that fits large models into memory to achieve results by reducing memory redundancy in parallel with data.
ZeRO-1 contains model state memory optimization, while ZeRO-2 provides optimization for activated memory and fragmented memory. At the same time, ZeRO-2 has also been improved for the training model on a single GPU. The ZeRO-2 training model is reported to be 30% faster than Google BERT.
Microsoft says ZeRO’s memory optimization technology can train machine learning models with 17 billion parameters, such as the 170-parameter Microsoft Turing model, which is currently the world’s largest language AI model, mainly used for natural language generation (NLG).
It’s worth noting that Microsoft announced that it will open up Microsoft Turing models in the short term and train them in Azure Machine Learning. The upgrade of the DeepSpeed library will also allow developers to use ZeRO-2 to train large neural networks.
Azure Synapse Link: Real-time operational data can be analyzed
Azure Synapse Link enables users to get direct analysis results from real-time operational data in Azure Synapse Analytics, without the need for extraction, conversion, or loading steps. It also combines real-time data with existing analytics repositories to get an overall view of the business.
In addition, Azure Synapse Link can query data at “pb-level” speeds under the guidance of SQL lines, with intelligent workload management and concurrency capabilities that enable real-time query performance.
In terms of security, Azure Synapse Link features automatic threat detection and always-on data encryption, dynamic data masking, fine-grained access control, and column/row-level security.
It is reported that Azure Synapse Link will initially be released in Azure Cosmos DB, but will soon be available to all operating systems, further helping developers reduce costs and time.
WSL 2 Adds Support for GPU, Linux GUI Applications
WSL 2 adds several new features, including support for GPUs, Linux GUI applications, and simplified installation experience.
On the one hand, WSL 2 supports GPU computing workflows, enabling Linux tools to leverage GPUs to hardware acceleration for a range of development scenarios, such as parallel computing, training AI, and machine learning models. This feature will be officially updated in the second half of this year.
WSL 2, on the other hand, will support Linux GUI applications, allowing users to run Linux GUI applications directly when opening a WSL instance without the need for a third-party server.
Later this year, WSL 2 will support a simplified installation experience. At that time, the developer will be able to simply run the “wsl.exe -install” command and restart it.
Microsoft Teams Platform Features
During the outbreak, there was a significant increase in usage of the Microsoft Team Smart Conference Platform, a core component of Microsoft 365.
In April, Satya Nadella said That Microsoft Team had more than 200 million single-day attendees, resulting in more than 4.1 billion minutes of meeting records. In addition, Teams currently has more than 75 million active users every day, two-thirds of whom share, collaborate, or interact with teams.
As a result, Microsoft has made a series of feature updates to Microsoft Teams for online conferencing, including extending Visual Studio and Visual Studio Code to allow developers to develop applications using tools they are familiar with, and Providing Virtual PowerTeams Bots to eliminate the use of Power Power Agentbots in Teams The user’s need to log on repeatedly, etc.
Open source and upgrade Fluid Framework
In 2019, Microsoft launched the collaboration platform Fluid Framework, which helps users collaborate better and enhance sped-sharing experiences.
At tonight’s conference, Microsoft announced that it will open-source Fluid Framework, along with a series of upgrades to its collaboration format, mainly in Outlook and Office.com.
On the one hand, users can insert charts, task lists, etc. in the web version of Outlook, so that users’ sales data, project tasks and research reports and other information to keep up-to-date;
On the other hand, users can create and manage Fluid workspaces, such as document activity feeds and recommendlists, or search in office.com. Because the Fluid Framework build is lightweight, users can edit it instantly.
Provide responsible machine learning tools
Eric Boyd, vice president of Microsoft’s Artificial Intelligence platform, points out that more and more developers are now being asked to build an AI system that is easy to explain and meets non-discrimination and privacy regulations.
With this in mind, Microsoft decided to publish responsible machine learning tools in the Azure Machine Learning and OSS toolkitto, reducing inequities and ensuring data privacy and confidentiality by improving the interpretation of the model, and further helping developers deploy AI models more responsibly.
On the one hand, by combining Fairlearn with Azure Machine Learning, developers and data scientists can use specialized algorithms to ensure that everyone has fairer results.
On the other hand, by combining the new WhiteNoise Difference Privacy Toolkit with Azure Machine Learning, it enables data science teams to build machine learning solutions that protect privacy while preventing the re-identification of personal data.
In addition, Azure Machine Learning provides data and network protection for secure model training and deployment. These include support for the Azure virtual network, dedicated links to machine learning workspaces, and customer management keys.
Conclusion: A Technology Feast for Developers
Judging by the launch of Microsoft Build conference tonight, Microsoft is paying more and more attention to the developer experience and emphasizing its responsibilities and obligations as it provides developers with more and more convenient and rich development tools.
Among them, the introduction of The AI supercomputer developed by Microsoft in cooperation with OpenAI, the first Azure service for professional applications, the unification of Project Bonsai, UWP and Win32, the AI development platform for industrial systems, and the launch and upgrading of heavy products such as Microsoft Turing Model, undoubtedly demonstrate the dependence and empowerment between Microsoft and developers.
With the global digital transformation boom in all walks of life, how will Microsoft further explore its cooperation and future with developers, AI technology, and industry in the process? Time will tell us the answer.