Artificial Intelligence (AI) has become one of the most popular topics in recent years. Almost all enterprises have evolved to, or intend to set foot in, the AI field, with the hope and expectation of improving efficiency and reducing costs by leveraging AI technologies. However, in most cases, evolution to AI applications is slow due to existing production and management processes and the restrictions imposed by Service Level Agreements (SLAs).
Since establishing the Noah's Ark Laboratory for basic AI research in 2011, Huawei has applied AI technologies to much of its own business. Huawei is focused on the value that AI can bring to solving enterprise problems and is dedicated to automating production processes using AI. In this territory Huawei has achieved a broad range of success in many fields, including the smartphone and eCommerce industries.
In September 2017, HUAWEI CLOUD launched the Enterprise Intelligence (EI) platform to provide essential platform services, general-purpose services, and solutions for industry-specific scenarios. The HUAWEI CLOUD platform enables Huawei customers to build smarter enterprises.
Following are some of Huawei's successes in AI applications.
1.1 Cost Reduction for Supply and Procurement Systems
Huawei leverages AI technologies to optimize the internal supply chain process, streamlining the processes for forecasting, logistics, warehousing, customs declarations, transportation, and signing.
1.1.1 Scenario 1: AI-based transportation-path planning cost reduction
Currently, Logistics Service Providers (LSPs) pick up goods for shipment that are split to forwarders according to delivery documents. However, multi-point goods pickups impose high costs because extra pickup points lead to additional expenses. The impact of this factor is large, as the annual additional costs have reached as high as USD 1.59 million (CNY 10 million).
Figure 1: Goods transportation route using AI
Appropriately planning goods pickup helps Huawei reduce additional costs and improves delivery efficiency. Specifically, improved route planning includes allocating leased vehicles by day, optimizing pickup routes, and maximizes the number of full-loads in order to minimize the cost of additional pickups. AI is used to optimize the transportation route to further reduce additional costs. The result is a 30 percent savings due to control of route optimization.
1.1.2 Scenario 2: OCR improves the efficiency of recording customs documents
Traditionally, Huawei assigns specified staff to manually record information for millions of customs documents generated each year. Naturally, this method is not efficient and input errors may occur due to employee negligence or fatigue. Needless to say, a method for quickly and accurately processing numerous documents was required.
Figure 2: Recording customs documents using OCR
Optical Character Recognition (OCR) helps address such troubles. Automatic high-precision collection of document information significantly improves the efficiency of recording documents. OCR has a numeric value recognition accuracy of up to 97.37 percent. The use of Big Data analytics further reduces operating costs and improves service efficiency. Evaluation and analysis of customs documents help control risk exposure at the level of tens of millions of US dollars each year. The rate for business process automation has increased by 50 percent.
Huawei's application of AI in the supply and procurement process has been successfully applied to other enterprises. For example, path optimization for logistics and transportation has been successfully translated to the medical logistics industry. OCR has also improved service efficiency for enterprises such as Baoxiaobar and Jointown Pharmaceutical Group.
1.2 AI Improves the Efficiency of Global Production and Delivery
Huawei serves telecom carriers in more than 100 countries for whom order fulfillment involves the design, production, and delivery for millions of physical locations each year. Huawei leverages AI technologies to help their Global Technical Service (GTS) department efficiently deliver and manage the support for mobile base station installations worldwide.
1.2.1 Scenario 1: AI improves the efficiency of base station design
Huawei deploys and constructs up to millions of new and upgraded base stations every year. Base station design involves multiple phases and requires multiple data sources and complicated connections. In addition, senior personnel must be assigned to participate in the design process. Much repetitive work is required. Based on AI research and accumulated experience in base station design, Huawei has transformed the base station design process from being very labor-intensive to an intelligent, cloud-based model that greatly improves work efficiency.
Huawei has developed algorithms specific to the following six service scenarios:
Recommended solution for new main equipment deployment
Recommended solution for capacity expansion and reconstruction of main equipment
Design of new antenna deployment
Live network rule learning for main equipment
Probability graph and recommendation learning of the main equipment solution
Algorithm for calculating the evolution path of main equipment
AI machine-learning algorithms, such as classification, frequent item mining, recommendation sorting, graph calculation, and knowledge graphing are involved in these six service scenarios. When fully applied, the AI-based intelligent base station design process delivers up to 70 percent precision, increases design efficiency by 33 percent, and shortens the design period.
Figure 3: Machine-learning process for intelligent MBB design
1.2.2 Scenario 2: Image processing improves base-station delivery quality check to achieving efficient acceptance
To ensure quality, the Huawei GTS requires intense manpower to check more than 24 million quality items and more than 45 million images or data items each year. Further delays are the result of inefficient manual checks due to repeated site visits. The result is that process of deployment operations is affected and project costs increase.
By using deep learning technologies such as image classification, and object detection and segmentation, machine-assisted reviews conduct standards compliance checks using job images. The review method is fast, available 24/7, and improves review efficiency and reducing delays.
Figure 4: Machine-assisted intelligent review of job images
For wireless hardware installation projects in Thailand and Indonesia, single-site quality check efficiency was improved by approximately 10 times, the report generation period was cut by two-thirds, and the acceptance cost was decreased by 40 percent
1.3 Adopting AI to Improve User Experience with Huawei Smartphones
Huawei provides terminal services in the form of mobile phones in addition to selling hardware, software, and services for carriers and enterprises. Huawei has become the third largest mobile phone vendor in the world with shipments of 153 million handsets in 2017. In addition, Huawei's terminal business involves mobile Internet and eCommerce. With a huge customer base covering such important fields, Huawei is leaning on AI technologies to enable more accurate recommendations and a better user experience.
1.3.1 Scenario 1: AI-based online recommendations enable accurate application recommendations
Nearly 300 million subscribers have registered with the Huawei smartphone application marketplace, where the average daily distribution is over 100 million downloads and the peak volumes are over 200 million downloads. Key technologies are required to provide accurate, effective, and stable personalized recommendation services. In the marketplace, personalized recommendation algorithms are needed to fulfill more than 10 scenarios, including apps, games, advertisements, and news.
The Huawei-developed real-time online recommendation technology implements advanced parallel, incremental, and real-time recommendation algorithms and real-time streaming systems. This technology supports 10-billion-dimensional features and delivers personalized recommendation services, recommendation engine scale-out, and template customization based on industrial practice. Further, this technology enables quick access to newly recommended services (shortened to three weeks from three months), minute-level model updates, and second-level feature updates. The recommendation technology has been involved in 40 percent of generated service traffic and has increased the app download conversion rate by approximately 20 percent.
In addition to recommendations on the cloud, Huawei's AI online recommendations apply to handset terminals. For example, the EMUI Smart Assistant offers advice on suitable applications and content services. It provides various entries involved in 10 service fields, such as payment, food, hotels, scenic spots, banks, movies, and travel, as well as personalized content services that include news, music, and video cards.
1.3.2 Scenario 2: AI-based risk control engine ensures secure transactions for eCommerce
The Huawei Vmall service faces tens of millions of attacks each day. When a new mobile phone is launched, the rate of attacks will increase by 100 to 200 percent. Moreover, these attacks are directed against accounts and transactions, and attack rules cannot be determined.
AI-based intelligent risk control enables user-defined rules by developers. Dynamic rule writing is supported, and all rules can be defined using standard SQL (like SQL99) stored procedures so that all developers can quickly master the development rules. Amazingly, the new rules take effect within one second of being invoked.
The detection of fraudulent transactions requires large amounts of computing power — specifically, running at least 1,000 rules and associating data in the time window of the past month. AI-based intelligent risk control makes it possible to complete this computation within 50 milliseconds (ms). In the financial field, a detection process with a response of 50 ms or less helps identify transaction risks in real time. If there is fraudulent activity, the transaction is stopped or a warning is generated, effectively preventing payment risks.
Huawei AI-based technologies help achieve other successes. For example, the intelligent Q&A system applied in the IT customer service system greatly reduces labor costs and increases the automatic closure rate up to 65 percent. Applications based on machine learning, like accurate marketing and subscriber retention, have been widely used by traditional telecom carriers.
1.4 Conclusion
Mr. Ren Zhengfei, president of Huawei, said that Huawei is focusing on AI because of its potential for daily use. At Huawei, AI research encompasses basic theory research, product development, and enabler development to reconstruct process management. The HUAWEI CLOUD EI leverages Huawei's accumulation of technologies and experience in AI and cloud computing to enable online services for government and enterprise customers with the aim to build a fully connected, intelligent world.
