How does graph intelligence move toward general artificial intelligence? This study at the Bund Conference has useful information

“Generative artificial intelligence is the most important technological revolution in the past 40 years.” This is the latest judgment of Microsoft co-founder Bill Gates. When intelligence emerges, how to promote scientific research and application innovation has become the focus of the industry.

On September 7, at the "New Generation Data Base - Exploring the Application and Development of Graph Intelligence" sub-forum at the 2023 Bund Conference, Ant Group brought a fusion research-"Large Graph Model" , referred to as LGM). This research combines graph computing with graph learning and large language models, using the generation capabilities of large language models and the correlation analysis capabilities of graph computing to provide more intuitive and comprehensive information presentation and more accurate insights, thereby better Solve massive and complex digital application problems. At present, Ant has completed the first phase of research work on "Generative Heterogeneous Graph Enhancement", and the relevant results papers have been included in the World Computer Conference (WWW 2023).

Graph computing is a powerful data processing technology that can solve correlation problems in complex relationship networks. It has applications in financial anti-fraud, weather forecasting, drug development, and even brain-inspired research. It is known as artificial intelligence "Ox nose". Large models are the most likely technology to move towards general artificial intelligence, achieving tasks equivalent to or even better than humans in some areas.

Why use cutting-edge technology to drive cutting-edge technology? Can’t large language models independently complete data analysis and mining tasks? Liu Yongchao, a senior technical expert at Ant Group, said that large language models can infer implicit relationships, but cannot draw relationship diagrams. Researching data relationships requires clear links, and using graph structure representation is easier to understand. "Combining large language models with graph computing means first logically reasoning from massive information, and then using supercomputing to calculate relationships. This is like connecting an external supercomputer to the human brain, with stronger capabilities," Liu Yongchao explained.

(Liu Yongchao of Ant Group shared the research results of the “Big Picture Model”)

In this study, Ant Group mainly carried out two tasks. First, use large language models to enrich graph data. Different from ordinary context-dependent models, large language models can generate new data points based on existing data. The work, titled “Generative Heterogeneous Graph Augmentation,” expands and enriches different types of graph data with large language models. Second, Prompt (an instruction or prompt) is used to guide the model to learn and discover specific data characteristics. For example, if you set a Prompt "Common characteristics of groups that have defaulted more than three times in a year", the model will generate data samples that meet specific conditions. This capability can accelerate the process of data analysis and feature discovery.

Ant Group is a leader in the field of graph computing. The graph computing platform TuGraph jointly developed with Tsinghua University has broken the LDBC SNB world record in the authoritative evaluation of graph databases three times. In 2021, it won the "Leading Scientific and Technological Achievements" award at the World Internet Conference in 2023. Selected into the "Leaders" quadrant of IDC MarketScape China's graph database market. In recent years, the industry has carried out various attempts to move graph intelligence toward general artificial intelligence. The "big graph model" is Ant's research starting in early 2022 in order to solve industry problems. With the outbreak of large models at the end of 2022, the feasibility of this research has been verified.

In this forum, experts and scholars in the field of graph intelligence at home and abroad also shared more hot developments. M. Tamer Özsu, a professor at the University of Waterloo in Canada, shared the challenging flow graph computing technology in the industry. Chen Huajun, a professor at the School of Computer Science at Zhejiang University, talked about the opportunities and challenges of knowledge processing in the era of large models. Chen Hongyang, deputy director of the Graph Computing Research Center of Zhijiang Laboratory, brought the latest research on scientific computing and biomedical research and development of Zhijiang Zhuque Graph. Li Yazhou, co-founder and deputy editor of Machine Heart, believes that research combining graph intelligence and large models is expected to bring significant improvements to data intelligence.

Regarding the possibilities of artificial intelligence and graph computing, the Bund Graph Intelligence Forum brings together insights from different perspectives and charts an important development path for the development of graph intelligence.

The above is the detailed content of How does graph intelligence move toward general artificial intelligence? This study at the Bund Conference has useful information. For more information, please follow other related articles on the PHP Chinese website!