Causal Discovery with Continuous Additive Noise Models

About

We consider the problem of learning causal directed acyclic graphs from an observational joint distribution. One can use these graphs to predict the outcome of interventional experiments, from which data are often not available. We show that if the observational distribution follows a structural equation model with an additive noise structure, the directed acyclic graph becomes identifiable from the distribution under mild conditions. This constitutes an interesting alternative to traditional methods that assume faithfulness and identify only the Markov equivalence class of the graph, thus leaving some edges undirected. We provide practical algorithms for finitely many samples, RESIT (Regression with Subsequent Independence Test) and two methods based on an independence score. We prove that RESIT is correct in the population setting and provide an empirical evaluation.

Jonas Peters, Joris Mooij, Dominik Janzing, Bernhard Schölkopf · 2013

Related benchmarks

Task	Dataset	Result
Bivariate Causal Discovery	SIM-c	Accuracy82	33
Bivariate Causal Discovery	Qd-V	Accuracy80	33
Bivariate Causal Discovery	SIM	Accuracy78	33
Bivariate Causal Discovery	D4 s1	Accuracy58	33
Bivariate Causal Discovery	AN	Accuracy99	33
Bivariate Causal Discovery	PER	Accuracy70	33
Bivariate Causal Discovery	LS	Accuracy72	33
Bivariate Causal Discovery	NN-V	Accuracy61	33
Bivariate Causal Discovery	Net	Accuracy76	33
Bivariate Causal Discovery	Tue	Accuracy63	33