Goals

Our mission is to investigate the theory and techniques for causality analysis, develop causal AI algorithms, and build AI-based/aided forecasting systems. Specifically, we want to develop the methodologies for causal learning and prediction based solely on observations, laying a foundation for the new-generation AI-based/aided ocean-atmosphere forecasting systems, in order to push beyond the predictability limit embedded within the traditional numerical models.

Ressearch  Progress

We have found that causality is actually a real physical notion that can be rigorously derived from first principles. This results in the Liang-Kleeman information flow-based causality analysis. In contrast to those in the traditional statistical or axiomatic formalisms, this causality analysis is quantitative; the resulting causality can be normalized, and is invariant upon arbitrary nonlinear coordinate transformation, indicating that it should be an intrinsic physical property. So far this method has been successfully applied to many real problems in the diverse disciplines such as atmosphere-ocean dynamics, climate attribution, neuroscience, quantum information, financial economics, to name a few. Based on it, our team has since developed a preliminary causal AI algorithm, with which a rather successful forecast has been made in an experiment for the prediction of the summertime precipitation over China.