Let AI be omnipresent


Recently, the 16th Huawei Global Analyst Conference was held in Shenzhen, China. At the theme forum of “Let Cloud be everywhere, let AI be omnipresent”, Jia Yongli, general manager of Huawei cloud EI service product department, delivered a keynote speech “Let AI be omnipresent”, and shared the innovative solutions and related industry practices of Huawei whole stack AI in the process of enabling enterprises to intelligentize transformation.

Full Stack Scenario AI Solutions Make Intelligence Everywhere

At present, the application of AI is extending from single business scenario replication to more departments, from small-scale exploration of pioneers to the overall layout of industry enterprises. At the same time, the development and evolution speed of AI-related technology is also very fast, new AI algorithms are constantly refreshing the existing records, and computing frameworks are constantly emerging. Under the trend of rapid development, enterprises also encounter a lot of confusion in the process of exploring and using AI.

Aiming at the pain points of AI applications, Huawei formally released the full stack scenario AI solution at the 2018 Full Connection Conference, which provided the ability from the bottom AI chip, AI framework to application enablement for customers, and provided the full stack AI solution Huawei cloud EI for the government, enterprises and developers. In the future, based on Huawei Ascend AI chips and multi-core ARM chips “Kunpeng 920”, Huawei Cloud will be more energy efficient performance to help customers further improve the performance-price ratio of AI development, data processing and service experience.

Huawei Cloud Model Arts: Faster and Preferential AI Development Platform

In the future, with the increase of data model and the requirement of mass data preprocessing capability, the acceleration of life cycle management will become more and more important. Only with the optimization capability of end-to-end whole stack, can users quickly create and deploy models and manage life cycle AI workflow. Huawei Cloud Model Arts one-stop AI development platform came into being.

From the beginning of design, ModelArts focuses on AI developers at different levels. From data scientists, Algorithm Engineers and even AI beginners, they can manage AI model/workflow through a unified platform. ModelArts can help users carry out life cycle management including AI development and operation management, and provide more efficient solutions in data processing, model training, model management, model deployment, AI market and other aspects. Next, Huawei Cloud will continue to research AI frontier algorithms and theories, such as small sample training ability, semi-supervised learning ability, neural network automatic search ability, etc., in Huawei 2012 Laboratory, EI Product Department, and gradually become product-oriented and open to developers.

Specifically, first of all, data processing is very critical in AI development process. It sometimes accounts for more than the training time in the project. At the same time, it often consumes a lot of cost to label. ModelArts systematically designs this process, which provides many functions such as version management of data sets, semi-automatic labeling and so on. According to different projects, the human resource consumption in this stage is different. Can save 50% – 80%.

In the part of model training, ModelArts achieves training acceleration through collaborative optimization of hardware, software and algorithm. MoXing, a self-developed in-depth learning framework, can automatically transform the developer’s single-machine program into a large-scale distributed training program; provide automatic debugging and optimization of super-parameters; provide automatic search capability of neural network to help the developer realize model training automatically, greatly improve the efficiency and speed of algorithm development and training, and save training costs.

In the aspect of model management, the process metadata is unified managed and developed by graph engine, which can automatically visualize the relationship between workflow and version evolution, and then realize model traceability and precision tracking. In the aspect of model deployment, Model Arts can deploy AI model as online reasoning service or edge reasoning service by one key. Facing the edge reasoning scenario, modelarts can automatically optimize the adaptation of the model, such as neural network distillation, model compression, pruning and other processing, so that the AI model can better adapt to the edge deployment environment.

Aiming at the landing scenario of AI in three major industries, cooperation in various fields has achieved results.

Jia Yongli said that Huawei Cloud EI summarized the key path of AI technology landing in the industry through a large number of project explorations in more than ten industries. He said that the key to the AI industry landing is to find the right scenario. Huawei Cloud EI aims at three scenarios: massive repetitive scenarios, expert experience scenarios and multi-domain collaboration scenarios, aiming at achieving efficiency improvement, professional inheritance and breaking through the limits, and helping industries to upgrade their intelligence.

In the typical mass repetition scenario – Logistics industry, there are countless cargo damages every year, most of which are caused by improper manual operation. In view of the logistics cargo transportation and security inspection scenario, Huawei cloud EI uses visual technology to help Debon Express achieve a comprehensive automated detection, save a lot of time and labor costs, and effectively reduce cargo losses. Prior to this, Debon Express used 13,000 video channels and manual viewing. Each person watched eight cameras at a maximum of four times the speed of a day. The estimated human cost was more than 400 people.

In view of multi-domain collaboration scenario, Huawei Yun helps an airport to solve such problems as flight bridge-to-bridge ratio, passenger bridge-to-bridge ratio, walking distance, ground clothing consumption and taxiing conflict rate. The core index bridge-to-bridge ratio has been increased by 5% artificially, and the conflict ratio has been reduced by 10% artificially, thus improving the efficiency of commanders and airport operation in an all-round way. The number of passengers who need to take the ferry has been reduced by 2.5 million due to parking space allocation and other reasons.

In any industry, experts are very valuable and scarce resources, especially in the medical field, which also provides the most suitable basis for expert experience scenarios. On the scene of the conference, Li Yinghua, CIO of Jinyu Medicine, shared the innovative achievements of Jinyu Medicine and Huawei YunEI in the medical field. He said that Jinyu Medicine and Huaweiyun have cooperated on large data, AI and other aspects. For example, AI technology such as in-depth learning can be used to learn and train a large number of pathological sections and expert labeled data. The resulting deep neural network model will help pathologists diagnose patients’pathological sections more efficiently, timely and accurately. In the current cooperation between the two sides, based on the professional labeling of senior cell pathologists in Jinyun Medical College and the efficient and high-quality network model construction and training of Huawei cloud EI vision team, the project has made a breakthrough in stages. The sensitivity of AI-assisted screening for cervical cancer cytology is over 99%, which is comparable to the level of experts.

Accelerate the release of AI technology dividends and build a more complete ecosystem

For any industry, developers are a key factor in the AI landing process. For partners, Huawei Cloud EI provides a variety of support plans, such as marketing planning, solution enabling, joint innovation Lab, to enable partners to flourish AI development capabilities.

For developers, universities and research institutes, Huawei released the Fertile Land AI Developer Enabling Program at the 2018 Full Link Conference. Through the Fertile Land AI Developer Enabling Program, we will build a channel for technology exchange, talent cultivation and opportunity creation.

In addition, developers can use ModelArts to quickly launch innovative activities. Taking Shanghai Jiaotong University as an example, with the help of ModelArts, students have transformed the traditional racing car into an AI-capable self-driving car in a short time. The modified racing car can recognize traffic lights, obstacles, lanes and other environments, and can follow specific targets.

At the same time, through the end cloud collaboration AI application development platform HiLens, users can more easily manage a large number of terminal devices, and seamlessly communicate with ModelArts. For example, Shenzhen Seabird Technology, through the AI skills developed by HiLens, sends the AI model to the home security cameras it has developed to make the ordinary cameras more intelligent.

Not only will the AI development process become more efficient and efficient, but Huawei Cloud also hopes to help more enterprises and developers better utilize and share their AI capabilities with better AI technology and ecology, so as to build a more complete ecosystem. Developers can share and publish AI models and data sets through the newly released ModelArts AI market in March 2019. They can also obtain basic data sets and models from the market for further development. In addition, Atlas 200 developer suite, HiLens wit in-depth learning camera and other developer suites will be officially commercialized in May.

At present, Huawei Cloud is continuing to apply AI scene to the ground. It continues to explore and practice “Cloud + AI” to help bring the digital world into everyone, every family and every organization, so as to build an intelligent world with interconnection of all things.

To learn more about Huawei Cloud EI free courses, visit Huawei Cloud College (https://edu.huaweicloud.com/)