[Beijing, China, October 31, 2019] On October 29, Huawei released the first autonomous driving network (ADN) solution. Today, Han Yufa, President of Huawei Network AI Product Dept, delivered a keynote speech titled "AI Enables ADN" at the AI summit of the 2019 ICT China High-Level Forum. Executives and experts from the Ministry of Industry and Information Technology (MIIT), China Academy of Information and Communications Technology (CAICT), three major operators from within China, research institutes, and industry partners witnessed Huawei's innovative achievements in AI-based ADN.
Han Yufa, President of Huawei Network AI Product Dept, introduces Huawei Network Artificial Intelligence Engine (NAIE)
5G deployment is gaining momentum in over 50 cities and 30 provinces in China. The first 5G base stations are expected to number 800,000. 5G is incubating innovative services across a diverse range of industries, including education, healthcare, agriculture, and tourism. Examples are the 4K live broadcast of China's seventieth anniversary celebration parade, smart manufacturing, and unmanned driving. Diversified 5G application scenarios have different requirements on 5G networks. For example, live broadcast requires higher bandwidth, unmanned driving requires lower latency and higher reliability, and smart manufacturing requires massive connections. An automatic, intelligent network is the solution to these demanding challenges that are being faced in terms of service provisioning and network O&M.
Industry players have begun to explore how to deploy an automatic, intelligent network. In November 2017, the ITU set up a 5G network machine learning team with that focuses on the usage of AI on networks. Since May of this year, TMF and 3GPP have released white papers on autonomous networks and autonomous driving on mobile networks respectively. China Telecom released a research report on the different levels of intelligent networks. Evolution towards an automatic, intelligent telecom network has become the new trend.
Since putting forward the ADN concept last year, Huawei also released the overall ADN architecture and cases at the PT Expo China 2019. This architecture includes four units: network, network management and control, cross-domain intelligent O&M, and network AI. With a focus on simplified O&M and a simplified network, AI is being introduced to the network element (NE) layer, network layer, and cloud full stack, facilitating the deployment of an automatic, intelligent network.
Han Yufa also introduced Huawei NAIE, which enables developers to rapidly develop AI applications. The NAIE provides data lake, model training, model generation, and communication model services on the public cloud. These services solve the most complex tasks during AI application development, including data preparation, data feature exploration, and model optimization, enabling developers to rapidly acquire NAIE capabilities.
The data lake service provides data processing tools for data collection, integration, modeling, analysis, labeling, and other purposes, and it also provides data governance templates as cloud services for developers to improve data governance efficiency. The data collection interface is standardized to support over 30 NEs and automatic interconnection with over 100 types of devices. More than 10 types of built-in templates in different telecom service scenarios, efficient telecom data labeling tools, and 480 million online training data samples are offered to shorten the data preparation time from three months to one week, which is a 90% reduction.
The model training service provides an integrated development environment (IDE) for data processing, feature extraction, model training and verification, and other functions. This service supports mainstream algorithm frameworks, such as TensorFlow, Caffe2, and SParkML, and integrates the 30 years of knowledge and experience that Huawei has accumulated in the network domain. Over 30 preset telecom network feature exploration tools, more than 50 telecom domain assets, and SDKs designed for algorithm commissioning and feature services and processing shorten the model design and exploration period from three to one week, which is a 70% reduction.
The model generation service further simplifies the model development process based on the model training service. Developers can simply offer training data in a preset typical scenario model to rapidly train, verify, and generate a model. This simplifies telecom AI model development and shortens the development period. Take DC PUE optimization as an example. Traditionally, heating, ventilation, and air conditioning (HVAC) experts, data engineers, algorithm experts, application developers, and many other experts need to spend at least half a year to develop a model. Huawei model generation service enables a single HVAC expert to train a model within only two weeks, reducing the labor and time required for model development by over 95% and significantly lowering model generation costs.
The communication model service relies on a cloud-based inference framework. Users can rapidly perform inference by merely invoking APIs and providing inference data. The inference result can be used for service application development. This service is applicable to general models, such as disk fault detection. Since most disk fault symptoms and characteristics are the same, users can provide SMART data to acquire disk health check results online. Similar scenarios include KPI anomaly detection and ECA anomaly detection. This service can be deployed on a cloud to facilitate service integration.
Huawei collaborates with China Telecom to deploy the NAIE on China Telecom's e-Surfing Cloud, enabling China Telecom to acquire AI development capabilities and facilitating the usage of AI technologies in different telecom services. China Telecom has deployed AI applications to improve O&M efficiency and energy efficiency in many provinces, such as Guangdong and Guangxi. Intelligent core network KPI anomaly detection enables China Telecom to detect faults 5.5 hours in advance and take preventive measures, reducing service losses caused by accidents.
Han Yufa, President of Huawei NAIE Product Dept, said, "Huawei NAIE is designed to facilitate the usage of AI, improve AI development efficiency for network services, and simplify network AI development for operators, enterprise partners, universities, and other developers. Huawei is dedicated to jointly building a secure AI industry environment with industry partners, ensuring orderly and controlled data flow among data providers, model developers, as well as users, and improving the new AI ecosystem."