Recently, the academic paper “Learning Semi-parametric Tree Models from Mixed Data”, co-authored by ZHOU Can and LI Nan, lecturers at the School of Statistics and Data Science, Nanjing Audit University, in collaboration with WANG Shuai, lecturer at Heilongjiang University of Science and Technology, WANG Xiaofei, professor at Northeast Normal University, and GUO Jianhua, professor at Beijing Technology and Business University, has been published online in Artificial Intelligence (AIJ), a leading journal in the field of artificial intelligence.
This study focuses on the problem of semi-parametric tree model learning in mixed-data scenarios. In practical applications, data typically contain both continuous and ordered variables, such as physiological indicators and disease severity classifications in medical diagnostics, or continuous scores and categorical labels in financial risk control. However, most existing structural learning methods are designed primarily for purely continuous or purely discrete data, making it difficult to effectively capture the hierarchical structure within mixed data and identify latent variables. To address these challenges, the research team proposes a novel semi-parametric tree model framework. Based on Gaussian copulas and a thresholding mechanism, this model enables the unified modelling of continuous and ordered variables. Regarding structure learning, the team further designs a bottom-up algorithm based on additive information distance for the recursive recovery of tree structures. Theoretical analysis demonstrates that, under the ideal scenario where the true information distance is known, the proposed algorithm can accurately recover the true tree structure with a computational complexity of . Furthermore, the paper establishes the probabilistic approximate correctness of the algorithm and provides the finite sample bounds required for achieving exact structural recovery.
Paper Information： Can Zhou, Nan Li, Shuai Wang, Xiaofei Wang & Jianhua Guo (2026). Learning semi-parametric tree models from mixed data. Artificial Intelligence, 353, 1-25. https://doi.org/10.1016/j.artint.2026.104499.
