A1214
Title: Extension of LiNGAM to functional data
Authors: Tianle Yang - Osaka University (Japan) [presenting]
Joe Suzuki - Osaka University (Japan)
Abstract: A causal order is considered, such as the cause and effect among variables. In the Linear Non-Gaussian Acyclic Model (LiNGAM), the order can only be identified if at least one of the variables is non-Gaussian. The notion of variables is extended to functions (Functional Linear Non-Gaussian Acyclic Model, Func-LiNGAM). First, it is proved that the order among random functions, if one of them is a non-Gaussian process, can be identified. In the actual procedure, the functions are approximated by random vectors. To improve the correctness and efficiency, optimising the coordinates of the vectors in such a way as functional principal component analysis is proposed. The experiments contain an order identification simulation among multiple functions for given samples. In particular, the Func-LiNGAM is applied to recognize the brain connectivity pattern with fMRI data. Improvements in accuracy and execution speed compared to existing methods can be seen.
