Continuous causal structure learning from incremental instances and feature spaces

Dianlong You, Hongtao Wu, Jiale Liu, Huigui Yan, Chuan Ma, Zhen Chen, Xindong Wu

Abstract: Highlights•Initially study CSL from incremental instances and feature spaces instead of fixed ones.•Propose a nonlinear CSL model with hidden variables to improve performance and avoid faithfulness constraints.•Leverage gradient-based continuous optimization to adapt changes of feature spaces.•Relax the constraint bound using ζ<math><mi is="true">ζ</mi></math> instead of zero to enforce acyclicity in dynamic scenes.