Quantifying identifiability in independent component analysis

DOI10.1214/14-EJS932MaRDI QIDQ405369zbMATH OpenWikidataFDO

Authors Alexander Sokol, Benjamin Falkeborg, Marloes H. Maathuis

Publication date 5 September 2014

Published in Electronic Journal of Statistics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1401.7899, https://projecteuclid.org/euclid.ejs/1408540293

zbMATH Keywords

empirical processidentifiabilityindependent component analysisasymptotic statisticscontaminated distributionKolmogorov normlinear structural equation modellingam

Mathematics Subject Classification ID

Asymptotic properties of parametric estimators (62F12) Robustness and adaptive procedures (parametric inference) (62F35)

Abstract: We are interested in consistent estimation of the mixing matrix in the ICA model, when the error distribution is close to (but different from) Gaussian. In particular, we consider $n$ independent samples from the ICA model $X=Aepsilon$, where we assume that the coordinates of $epsilon$ are independent and identically distributed according to a contaminated Gaussian distribution, and the amount of contamination is allowed to depend on $n$. We then investigate how the ability to consistently estimate the mixing matrix depends on the amount of contamination. Our results suggest that in an asymptotic sense, if the amount of contamination decreases at rate $1/sqrtn$ or faster, then the mixing matrix is only identifiable up to transpose products. These results also have implications for causal inference from linear structural equation models with near-Gaussian additive noise.

Recommendations

• scientific article; zbMATH DE number 1928763
• Independent component analysis via nonparametric maximum likelihood estimation
• Independent component analysis: principles and practice
• 10.1162/jmlr.2003.4.7-8.1177
• scientific article; zbMATH DE number 1233949
• Nonparametric independent component analysis
• A quasi expectation-maximization algorithm for independent component analysis

Cites work

• scientific article; zbMATH DE number 1022658 (Why is no real title available?)
• scientific article; zbMATH DE number 782652 (Why is no real title available?)
• A linear non-Gaussian acyclic model for causal discovery
• A note on strong approximations of multivariate empirical processes
• Compact CramÉr–Rao Bound Expression for Independent Component Analysis
• DirectLiNGAM: a direct method for learning a linear non-Gaussian structural equation model
• Efficient independent component analysis
• Identifiability of Gaussian structural equation models with equal error variances
• Independent component analysis via nonparametric maximum likelihood estimation
• Independent component analysis, a new concept?
• Independent component analysis: recent advances
• Multiparameter Processes
• On the identifiability of additive index models
• Semiparametrically efficient inference based on signed ranks in symmetric independent component models
• Weak convergence and empirical processes. With applications to statistics
• Weak convergence of probabilities on nonseparable metric spaces and empirical measures on Euclidean spaces

Cited in

(4)

• Independent Vector Analysis: Identification Conditions and Performance Bounds
• Uniqueness of complex and multidimensional independent component analysis
• Maximum independent component analysis with application to EEG data
• 10.1162/jmlr.2003.4.7-8.1177

Describes a project that uses

Uses Software

• FastICA
• DirectLiNGAM

This page was built for publication: Quantifying identifiability in independent component analysis

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q405369)