CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1460
Title: Constrained LiNGAM approach for tensor data Authors:  Ippei Takasawa - Doshisha University (Japan) [presenting]
Kensuke Tanioka - Wakayama Medical University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: Independent Component Analysis (ICA) for three-way data has been proposed to analyze fMRI data in the domain of cognitive science. However, ICA cannot estimate causal relations. LiNGAM is one of the methods that can be applied to estimate causal relations among variables. Some extensions of LiNGAM and LiNGAM for tensor data have been proposed. Circumstances in which causal relations change based on the group of individuals and depending on the period during which data is observed when causal relations are estimated for three-way data do exist. In such circumstances, it is advantageous to reveal each groups distinct causal relations; however, current LiNGAM for tensor data cannot make these estimations. To overcome this problem, we propose a constrained LiNGAM approach for tensor data as an extension of LiNGAM for tensor data that enables both common causal relations and each groups distinct causal relations to be revealed. Using this method, common causal relations and the groups distinct relations can be estimated and interpreted. In particular this method is based on the concept of the three-way modeling and estimation of causal relations using LiNGAM. Furthermore, the proposed method is applied to real-world data and the result is evaluated.