Recently, the third Data Science Online Summit hosted by DataFun was held grandly. Focusing on six major data science topics, including machine learning and data mining, AB experiments, causal inference, data middle-end and digital transformation, user growth and operations, and data science best practices, dozens of domestic and foreign front-line data scientists focused on cutting-edge data science technologies. In-depth sharing and exchange of results and practical application experience. Jiuzhang Yunji DataCanvas Company participated in the summit in depth and shared the latest research progress of cutting-edge data science and technology.
At the summit, He Gang, the AI architect of Jiuzhang Yunji DataCanvas Company, delivered a keynote speech on "Two-way Empowerment of AIGC and Causal Inference", exploring the possibility of integration and innovation of the recent hot artificial intelligence technology AIGC and the classic technology causal inference.
He Gang said that AIGC has performed amazingly in the generation of unstructured content, and that structured data analysis is still in a blank state. Causal inference is one of the most popular key technologies in the field of structured data analysis; with Agent-Based Modeling (ABM multi-agent modeling) is a bridge that can build a link between AIGC and causal inference, and realize the two-way empowerment of AIGC and causal inference.
ABM multi-agent modeling
ABM multi-agent modeling is a computing model used to simulate the actions and interactions of autonomously aware agents. It has the advantages of high simulation, emergent properties, and interpretability. The operation mode of ABM is to perform simulation operations under different parameter combinations. During the operation, data can be output and the data is stored to form a data set. It is very complete and has high quality features such as counterfactual availability, feature integrity, and controllability. Characteristic data assets. It has been verified that the high-quality data obtained from the ABM system has good applicability in the fields of causal effect estimation, causal discovery, and evaluation indicators in causal inference, breaking through the limitation of no counterfactual samples in the field of causal inference research.
ABM provides high-quality data basis for causal inference
It can be seen that ABM multi-agent modeling is very suitable as an AI carrier to realize the expansion of AIGC from unstructured data to structured data generation and make up for AIGC's shortcomings in the field of structured data.
At the same time, causal inference will accelerate the ABM multi-agent modeling process, especially in the two important process links of calibration data and emergence analysis. In the calibration data link, by applying causal effect estimation to parameter analysis, we can get closer to the calibration target, accelerate the calibration process of simulation, and accelerate business deduction and auxiliary decision-making; in the emergent explanation link, we use a combination of causal discovery algorithms to generate causal diagrams to The combination of causal discovery, machine learning, and sensitivity analysis techniques can provide a richer explanation of emergent behavior and further enhance the ability of emergent explanation.
Combining the theoretical innovation achievements of Jiuzhang Yunji DataCanvas Company in the field of causal inference in recent years and the research and development experience of YLearn causal learning software, an important open source tool, He Gang proposed to extend AIGC's ability to generate structured data and use AIGC to generate causal learning model reports. , and applying causal inference to expand the causal explanation ability of AIGC can become three research directions for the deep integration of causal inference and AIGC technology in the future.
YLearn causal learning software, released by Jiuzhang Yunji DataCanvas as an open source, is the only open source tool in the world that can end-to-end solve the five major causal learning tasks of "causal discovery, causal quantity identification, causal effect estimation, counterfactual inference and strategy learning" software tools. YLearn breaks through the limitations of machine learning based on correlation modeling, explores stable causal relationships in data, and realizes inference under counterfactual conditions. It can fully empower ABM multi-agent modeling and help tap the emergent capabilities of AIGC.
In the future, as an artificial intelligence basic software supplier based on "hard technology", Jiuzhang Yunji DataCanvas will continue to focus on the research and development of basic AI capabilities, further promote the integration and innovation of cutting-edge technologies such as AIGC and causal inference, and accelerate the development of cutting-edge technologies. Industrial application contributes independent innovation energy to the AI boom.