2019/04/19 by Ricardo Pio Monti, Kun Zhang, Monti, Ricardo Pio +3 · 4 citations
Chemistry · Computer Science · #Blind Source Separation Techniques #Electrochemical Analysis and Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Spectroscopy and Chemometric Analyses
paper · pdf · doi:10.48550/arxiv.1904.09096
openalex publication_date 2019/04/19 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
We consider the problem of inferring causal relationships between two or more\npassively observed variables. While the problem of such causal discovery has\nbeen extensively studied especially in the bivariate setting, the majority of\ncurrent methods assume a linear causal relationship, and the few methods which\nconsider non-linear dependencies usually make the assumption of additive noise.\nHere, we propose a framework through which we can perform causal discovery in\nthe presence of general non-linear relationships. The proposed method is based\non recent progress in non-linear independent component analysis and exploits\nthe non-stationarity of observations in order to recover the underlying sources\nor latent disturbances. We show rigorously that in the case of bivariate causal\ndiscovery, such non-linear ICA can be used to infer the causal direction via a\nseries of independence tests. We further propose an alternative measure of\ncausal direction based on asymptotic approximations to the likelihood ratio, as\nwell as an extension to multivariate causal discovery. We demonstrate the\ncapabilities of the proposed method via a series of simulation studies and\nconclude with an application to neuroimaging data.\n