On the afternoon of September 7, at the "Integrated Machine Learning and Operations Optimization" forum of the Shanghai Bund Conference, Ant Group officially opened the open source graph learning system Ant Graph Learning (AGL), which is the industry's first general-purpose industrial graph learning system.
Picture description: On the forum, Ant Group officially open sourced the graph learning system AGL.
AGL currently realizes information collaboration and structure perception on trillion-scale graph data, built digital graph intelligence solutions for multiple industries, and accumulated many excellent algorithm practices. Based on AGL, Ant Group has published more than 60 CCF-A/B international journal conference papers, authorized more than 40 invention patents, ranked first in five international lists and competitions, and is also a core participant in the national standard for graph neural networks.
This open source AGL v0.1 version also provides the industrial-grade graph learning system that Ant has polished for many years and a series of out-of-the-box graph learning algorithms that have been verified by business practices. The code repository has been put on GitHub that day. Through open source, AGL provides full-link solutions for industrial-level large-scale graph learning tasks. It hopes to provide developers with a powerful tool and platform to better apply graph learning technology to solve practical business problems. At the same time, through community co-construction, Absorb excellent system and algorithm practices from industry and research, continue to lower the application threshold of graph learning, promote the exchange and innovation of graph learning technology, and promote the widespread application of graph learning in all walks of life.
Since 2017, graph learning has been widely used in various fields and has become one of the current research hotspots in the field of artificial intelligence. This technology is the new strategic commanding heights of global technology competition. The "14th Five-Year Plan" Software and Information Technology Service Industry Development Plan issued by the Ministry of Industry and Information Technology mentioned that it is necessary to break through the key basic technologies of large-scale parallel graph data processing and support heterogeneous data management. Innovate key technologies and accelerate the development of new machine learning and other technologies.
AGL has also been widely used in Ant's diversified businesses and has achieved excellent business results. For example, when applied to the digital supply chain "Dayan" system of MYbank, the accuracy of supply chain identification is increased by 50%, and the loan availability rate is increased from 30% to 80%; when applied to the Alipay digital open platform, it helps ecological merchants improve their rights and supplies. Distribution efficiency exceeds 50%.
Data-driven intelligent decision-making has become an important tool for more and more enterprises in the decision-making process. At this forum, Zhu Jun, professor of computer science at Tsinghua University and director of the Fundamental Theory Research Center of the Artificial Intelligence Research Institute, shared an introduction to offline reinforcement learning Based on the diffusion model method, Zhang Guochuan, professor of Zhejiang University and vice chairman of the Chinese Operations Research Society, talked about the ideas, methods and technologies of operations research optimization under the digital economy. Chen Liang, associate professor of the School of Computer Science at Sun Yat-sen University, believed that industry characteristics need to be fully considered to improve intelligent financial decision-making. Robustness, collaborative efforts in multiple dimensions such as data, models, training, and computing power.