8th TUC Meeting – Yinglong Xia (Huawei), Big Graph Analytics Engine
Yinlong started his talk with an introduction of his new position at Huawei, what is the company doing and more specifically how is it involved with Big Data Research and graphs. He also explained that his research center is currently working on Big Data Analytics and Management from 4 sides: Natural Language Processing, Graph analyrics, Machine Learning and Deep Learning. His team at the same time, focuses on 4 market segments that include financial graph analytics, consumer data gathered from smartphones and other portable devices, telecommunications and cloud technology.
Mr. Xia also introduced the concept of how graphs can be used in traditional markets, as he calls them, such as telecommunications. He compares this market to nowadays’ cloud market, with its virtual machines and networks where companies are tending to build a unifying layer of software above these networks (Software-Defined Networks). Then he carried on this thinking to talk about building another layer on top of the SDN, an Open Network Operating System (ONOS) that could benefit from LDBC’s work.
Continuing with the network example, he pointed out that, if we understand the network as a graph, it is interesting to know how the topology of such graph impacts the propagation of information and how modifications to the topology could influence the propagation (like stopping the information flow right away or allowing it to reach the main vertices as soon as possible). After this, Yinglong mentioned the variety of applications that graph technology has in today’s markets ranging from Finance, with transaction analysis and fraud detection to Social Security.
After this argument, he explained the challenges his team is facing at the moment to build a powerful graph analytics platform:
- The huge scale graphs that need to be analysed (10B-1000B vertices), a few hundreds of properties that can be both static and dynamic.
- The irregularity of the data acces to the graph. Fata access patterns are different among diverse graph analysis algorithms.
- Near real-time requirement. That needs to be incremental as the graph updates.
- Efficiency and productivity balances
As always full talk and slides of Yinglong's presentation shown below: