This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies.Read our private policy

Knowledge Computing for Enterprise AI

Enterprises have sufficient data, but lack the computing power to use it well with AI.

By Jia Yongli, President, AI Domain, Cloud BU, Huawei
Jia Yongli

In 2018, Huawei asserted that AI deployment in industries must be scenario-specific and proposed three typical scenarios: repetitive, high-volume work; tasks that require expertise; and work that requires multi-domain coordination. 

However, can clear business scenarios alone guarantee the successful implementation of AI in industries? Numerous projects have shown that the answer to this question is no. Industries have sufficient data, but they lack the required computing power for AI. In addition, the implementation of AI usually involves iterations, and the obstacles to this process aren’t due to technology, but existing organizational and talent structures. That's why in 2019, HUAWEI CLOUD defined the following four keys to successful AI implementation in industries: clear business scenarios, readily available computing power, continuously evolving AI services, and organization and talent.

AI & industry know-how

More enterprises are adopting AI. After analyzing its experience during more than 600 projects, HUAWEI CLOUD has found that over 30 percent of these cases used AI in their core production systems, which brought an average increase of more than 18 percent in profitability and efficiency gains.

However, during this process HUAWEI CLOUD also identified deeper problems regarding AI adoption in a number of areas.

Communicate 87

Collaboration between industry experts and AI specialists: These two types of experts have difficulty understanding each other and working towards common goals.

Integration between industry know-know and AI models: Industries have usually spent decades or longer gaining their own expertise, and developed a large number of proven mechanism models based on physical, chemical, biological, and other types of information. The problem is whether these models can be integrated with data-driven AI models. If they can, then how can this be done effectively?

Integration between industry applications and AI systems: What is the exact relationship between AI systems and the application and control systems that industries have developed over the years? How can these industry applications evolve smoothly to become intelligent?

The basic issue behind the three problems above is: How can we leverage AI to make the most of industry know-know? If this mechanism is defined, the problems with expert collaboration and system integration will be easily solved.

To solve these problems, HUAWEI CLOUD has come up with a whole new approach – knowledge computing. This uses a series of AI technologies to extract and express various types of knowledge. After this, computing is performed based on large amounts of data to generate more accurate models and empower machines and humans. Knowledge computing is essentially about efficient, knowledge-driven and data-driven integration.

Large amounts of knowledge has been generated during the development of each industry, such as mechanism models about production systems, technical literature, valuable expert experience, summaries of methods developed over the years, and test reports.

Industries don’t lack knowledge. Instead, they lack the methods to efficiently utilize this knowledge. To explain the details of a borehole in the oil and gas industry, for example, well logging experts need to rely on repeated analysis of different systems by specialists in multiple domains, a process that usually takes months.

However, if such knowledge is expressed through graphs, and related borehole data is turned into vectors with graph embedding technology, the large number of vectors that are generated can be used to create a powerful model, which can perform accurate and efficient inference. This approach allows the more effective use and transmission of expert knowledge while saving significant time for experts.

Full-lifecycle knowledge computing solution

At HUAWEI CONNECT 2020, HUAWEI CLOUD officially launched the industry's first full-lifecycle knowledge computing solution. Based on the ModelArts AI development platform, the solution consists of four modules: knowledge acquisition, knowledge modeling, knowledge management, and knowledge application.

Knowledge acquisition is the starting point for knowledge computing. In this module, multimodal data parsing and processing are performed. This is the first key step in the process to convert data into knowledge. After this, preliminary knowledge can be used to perform knowledge modeling based on business scenarios.

Communicate 87

Knowledge modeling is the most important module. There are many approaches for knowledge modeling such as the graph embedding mentioned in the borehole example. In industrial settings, different approaches can be used to meet the requirements of different scenarios, such as adopting parallel computing powered by mechanism models and AI, and embedding mechanism models into deep neural networks. In this module, the reliability and explainability of the overall algorithm must be considered, in addition to whether massive amounts of data can be used to improve knowledge computing efficiency.

Once knowledge modeling is complete, effective knowledge management capabilities are required, including automatic updates, conflict management, and quality control. In the knowledge application module, HUAWEI CLOUD provides both basic capabilities, for example, high-concurrency real-time queries, searches, and recommendations, and advanced capabilities, for example, knowledge inference and prediction.

Enterprises can build their own knowledge computing platforms based on the HUAWEI CLOUD knowledge computing solution, and use these platforms in core processes such as R&D, production, operations, sales, and after-sales services. This solution has already been adopted in the oil, automotive, healthcare, chemical fiber, coking coal, steel, and transportation industries.

In the automotive industry, China FAW Group has used the HUAWEI CLOUD knowledge computing solution to build its own platform. This platform presents knowledge by business scenario that’s convenient and digital and quickly drives employee upskilling.

China FAW Group's platform has benefited one of the company's Hongqi 4S stores in many ways. The store's one-time repair rate increased by 4 percent, customer wait times for repairs dropped by 23 percent, and both manufacturer support and involvement rates and the time required for training maintenance technicians saw a 30 decrease.

In the steel industry, building on the HUAWEI CLOUD knowledge computing solution, Yantai Walsin Stainless Steel, integrated the alloy batching industry mechanism with AI to build a new AI model. This helped alloy batching engineers make more informed decisions and determine the optimal proportioning for alloy batching, which balances the quality of steel with economic benefits. This AI-assisted model increased the accuracy of predicting alloy ingredients to over 95 percent, 10 percent more than mechanism models. This has saved the company 20 million yuan every year.

In the healthcare industry, HUAWEI CLOUD has partnered with the team led by Prof Han Dali from the Beijing Institute of Genomics of the Chinese Academy of Sciences. Prof Han's team leverages knowledge computing to integrate hydroxymethyl DNA data with gene knowledge graphs compiled over the course of numerous research experiments. This approach accurately identifies the markers of key organisms in the blood, increasing the accuracy of early cancer diagnosis by 9 percent. 

In the transportation industry, HUAWEI CLOUD has been working with traffic management authorities to develop a systematic control solution based on AI computing. This is achieved by pairing knowledge computing with a wide range of expertise, collecting real-time feedback on traffic information, and integrating expert experience. The solution is currently able to optimize traffic flow at urban intersections and across entire regions. It has been verified at more than 300 intersections in Shenzhen, where it has reduced the congestion index by 8%. Moving ahead, knowledge computing will be further applied to improve the multidimensional traffic management of roads, metro systems, and airports.

Knowledge computing will transform the way knowledge is used and unlock its true power, while preparing every industry for new advancements. Huawei is ready to work with its industry customers and partners to build industry-specific AI knowledge computing platforms, deliver pervasive intelligence for all scenarios, and create new value for all industries.