亚信科技

On October 15, the “AI Innovation and Application Competition - Special Competition on Intelligent Network” co-sponsored by CAICT, AIIA, China Institute of Communications, and China Mobile ended in Beijing. The competition is divided into two events of “4G/5G Cell-level Multi-index Prediction” and “Network Cloud Equipment Exception and Root Cause Analysis”, attracting 220 enterprises and university teams across the country to participate. After fierce competition in the preliminary, semi-final and final, AsiaInfo finally won 1 championship, 1 runner-up, and won the qualification for promotion 2021-2022 “ITU AI/ML in 5G CHALLENGE”, and will represent China to compete with the top 5G AI teams around the world.

In this competition, ZTE, China Mobile Design Institute, China Mobile Jiangxi Company and other companies won the Second Prize, China Unicom, China Mobile Guangdong Company, China Mobile Sichuan Company, China Unicom E-Payment and other companies won the Third Prize, China Mobile Research Institute, Beijing University of Posts and Telecommunications, Root Cloud and other companies and universities won the Excellence Award.

Figure 1: The award-winning scene of the AI Innovation and Application Competition

Figure 2: The trophies of the AI Innovation and Application Competition

Building a network AI algorithm capability set

to solve the practical problems of network “planning, construction, maintenance, optimization, and operation”

The “4G/5G Cell-level Multi-index Prediction” contest required the participants to use historical data at a certain stage to predict the 4G/5G index value of “hour” granularity in each cell in the following 7 days. By adopting Dense-MLP-based 4G/5G cell-level multi-index prediction algorithm model and applying dynamic diminishing learning and other methods based features such as strong data periodicity, difficulty in processing and sensitive evaluation indexes, while taking into account the accuracy of business index prediction and the practicality of the model, AsiaInfo’s team won the championship of the contest.

Figure 3: Prediction algorithm model and prediction effect

The “Network Cloud Equipment Exception and Root Cause Analysis” contest required the participants to detect 1,124 split logs of hundreds of thousands of system operation logs and locate the row of root cause of exceptional log. The contest data comes from a distributed system which has complex and diverse exception causes and has not marked the root cause of exception, resulting in the great difficulty in detection and location. AsiaInfo’s team introduced the “network cloud equipment exception and root cause analysis algorithm model based on unsupervised integrated learning”. The model has high real-time performance of “minute-level training and millisecond-level prediction”, and supports multi-source logs and multi-dimensional data and other multi-scenario expansion, which can efficiently meet the service needs of communication operators.

Figure 4: Log exception detection and root cause analysis algorithm design

Creating “excellent products” with craftsmanship spirit

to help customers reduce costs and increase efficiency

AsiaInfo has been applying itself in the field of “network intelligence” and aims to become a “leader of 5G network intelligence and an innovator of communication AI”. It has achieved phased achievements in technological innovation and product research and development, and is among the best in 5G network intelligent products and intelligence operations products.

AISWare 5G Network Intelligence provides the intelligent operations and operating support capabilities required by 5G communication networks (compatible with 4G and other networks) for communications operators, aiming to help customers cope with the challenges such as rapid business launch, stable network operation, network planning and optimization, customer perception and experience management under the trend of network cloudification and software-oriented process. It can give full play to AsiaInfo’s mature artificial intelligence, machine learning and BSS/OSS business synergy capabilities to help operators improve operations efficiency, reduce costs, and improve customer satisfaction.

Based on its own AI platform algorithm model, AISWare AIOps provides AIOps capability engine for operators to empower the operations systems of each domain. The product focuses on the three directions of “quality assurance, cost management and efficiency improvement”. Oriented at the solutions of scenarios such as failure discovery, diagnosis, disposal, whole process of prediction as well as intelligent decision-making, capacity planning and resource optimization, the product will conduct componentized encapsulation to build reusable and evolvable operations model and stipulation and externally provide capabilities in the form of Open API, in order to support the rapid implementation of intelligent operations service, simplify the connection process, and reduce the cost of intelligence integration and infusion of the operations system.

AsiaInfo’s Telecom AI Labs and Institute for AI Industry Research, Tsinghua University - AsiaInfo-THU AIR Joint Lab of 5G Intelligence was established in 2018. By taking advantage of the multi-field research resources and industry expert resources of Institute for AI Industry Research, Tsinghua University, as well as AsiaInfo’s rich communications knowledge and industry application scenarios, a national-level research and innovation institution in the field of communication AI has been jointly built, establishing benchmarks in the field of communication AI, accumulating innovative 5G+AI algorithm system and industry solutions, and building a scientific research practice base for cultivating high-end talents and an incubation base for the commercialization of cutting-edge new technologies in the field. The lab has formed leading technical merit and complete commercial products and solution system in the fields of 5G network intelligence, edging intelligence, reinforcement learning, and federated learning. The lab has applied for more than 100 international and domestic patents; formulated over 20 international and domestic standards for 3GPP, GSMA, ITU, etc.; had more than 10 awards winning in international AI competitions; and published more than 50 monographs and academic papers on 5G+AI. All members of the lab come from well-known universities at home and abroad, with master or doctor degree or above. This has produced a convergence effect of mid-to-high-end 5G+AI talents and capabilities, and formed a closed-loop innovative operation mode of industry-university-research integration for research, experiment and commercialization of “5G+AI”.