Least-squares independence regression for non-linear causal inference under non-Gaussian noise

DOI10.1007/S10994-013-5423-YzbMATH Open1317.68198arXiv1103.5537OpenAlexW2026871688WikidataQ130551392 ScholiaQ130551392MaRDI QIDQ479477FDOQ479477

Authors: Makoto Yamada, Jun Sese, Masashi Sugiyama 

Publication date: 5 December 2014

Published in: Machine Learning (Search for Journal in Brave)

Abstract: The discovery of non-linear causal relationship under additive non-Gaussian noise models has attracted considerable attention recently because of their high flexibility. In this paper, we propose a novel causal inference algorithm called least-squares independence regression (LSIR). LSIR learns the additive noise model through the minimization of an estimator of the squared-loss mutual information between inputs and residuals. A notable advantage of LSIR over existing approaches is that tuning parameters such as the kernel width and the regularization parameter can be naturally optimized by cross-validation, allowing us to avoid overfitting in a data-dependent fashion. Through experiments with real-world datasets, we show that LSIR compares favorably with a state-of-the-art causal inference method.

Full work available at URL: https://arxiv.org/abs/1103.5537

Recommendations

• Non-linear causal inference using Gaussianity measures
• scientific article
• A linear non-Gaussian acyclic model for causal discovery
• Learning causal graphs of nonlinear structural vector autoregressive model using information theory criteria
• Justifying additive noise model-based causal discovery via algorithmic information theory

zbMATH Keywords

non-Gaussian noiseleast-squares independence regressionnonlinear causal inferencesquared-loss mutual information

Mathematics Subject Classification ID

General nonlinear regression (62J02) Learning and adaptive systems in artificial intelligence (68T05)

Cites Work

• Title not available (Why is that?)
• Elements of Information Theory
• Pattern recognition and machine learning.
• Theory of Reproducing Kernels
• On Information and Sufficiency
• On the limited memory BFGS method for large scale optimization
• Title not available (Why is that?)
• A Consistent Test for Bivariate Dependence
• Algorithmic Learning Theory
• Kernel dimension reduction in regression
• Title not available (Why is that?)
• On the influence of the kernel on the consistency of support vector machines
• Title not available (Why is that?)
• Nonlinear programming and variational inequality problems. A unified approach
• A linear non-Gaussian acyclic model for causal discovery
• Statistical analysis of kernel-based least-squares density-ratio estimation
• Justifying additive noise model-based causal discovery via algorithmic information theory
• Sufficient dimension reduction via squared-loss mutual information estimation

Cited In (3)

• Robust sparse Gaussian graphical modeling
• Non-linear causal inference using Gaussianity measures
• Causal regression for online estimation of highly nonlinear parametrically varying models

Uses Software

• L-BFGS
• bootstrap
• PRMLT

This page was built for publication: Least-squares independence regression for non-linear causal inference under non-Gaussian noise

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q479477)