BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions

D. Rothenhäusler; C. Heinze; Jonas Martin Peters; N. Meinshausen

BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions

D. Rothenhäusler, C. Heinze, Jonas Martin Peters, N. Meinshausen

12 Citations (Scopus)

Abstract

We propose a simple method to learn linear causal cyclic models in the presence of latent variables. The method relies on equilibrium data of the model recorded under a specific kind of interventions ("shift interventions"). The location and strength of these interventions do not have to be known and can be estimated from the data. Our method, called BACKSHIFT, only uses second moments of the data and performs simple joint matrix diagonalization, applied to differences between covariance matrices. We give a sufficient and necessary condition for identifiability of the system, which is fulfilled almost surely under some quite general assumptions if and only if there are at least three distinct experimental settings, one of which can be pure observational data. We demonstrate the performance on some simulated data and applications in flow cytometry and financial time series.

Original language	English
Title of host publication	Advances in Neural Information Processing Systems 28 (NIPS 2015) : Proceedings from the conference, "Neural Information Processing Systems 2015." •
Editors	C. C. Cortes, N.D. Lawrence , D.D. Lee, M. Sugiyama, R. Garnett
Number of pages	9
Publisher	NIPS Proceedings
Publication date	2015
Pages	1513-1521
Publication status	Published - 2015
Externally published	Yes

Cite this

Rothenhäusler, D., Heinze, C., Peters, J. M., & Meinshausen, N. (2015). BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions. In C. C. Cortes, N. D. Lawrence , D. D. Lee, M. Sugiyama, & R. Garnett (Eds.), Advances in Neural Information Processing Systems 28 (NIPS 2015): Proceedings from the conference, "Neural Information Processing Systems 2015." • (pp. 1513-1521). NIPS Proceedings.

BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions. / Rothenhäusler, D.; Heinze, C.; Peters, Jonas Martin et al.

Advances in Neural Information Processing Systems 28 (NIPS 2015): Proceedings from the conference, "Neural Information Processing Systems 2015." •. ed. / C. C. Cortes; N.D. Lawrence ; D.D. Lee; M. Sugiyama; R. Garnett. NIPS Proceedings, 2015. p. 1513-1521.

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Rothenhäusler, D, Heinze, C, Peters, JM & Meinshausen, N 2015, BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions. in C C. Cortes, ND Lawrence , DD Lee, M Sugiyama & R Garnett (eds), Advances in Neural Information Processing Systems 28 (NIPS 2015): Proceedings from the conference, "Neural Information Processing Systems 2015." •. NIPS Proceedings, pp. 1513-1521.

Rothenhäusler D, Heinze C, Peters JM, Meinshausen N. BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions. In C. Cortes C, Lawrence ND, Lee DD, Sugiyama M, Garnett R, editors, Advances in Neural Information Processing Systems 28 (NIPS 2015): Proceedings from the conference, "Neural Information Processing Systems 2015." •. NIPS Proceedings. 2015. p. 1513-1521

Rothenhäusler, D. ; Heinze, C. ; Peters, Jonas Martin et al. / BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions. Advances in Neural Information Processing Systems 28 (NIPS 2015): Proceedings from the conference, "Neural Information Processing Systems 2015." •. editor / C. C. Cortes ; N.D. Lawrence ; D.D. Lee ; M. Sugiyama ; R. Garnett. NIPS Proceedings, 2015. pp. 1513-1521

@inproceedings{bdc9a5a2064a4de68a46d3ccae6c4096,

title = "BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions",

abstract = "We propose a simple method to learn linear causal cyclic models in the presence of latent variables. The method relies on equilibrium data of the model recorded under a specific kind of interventions ({"}shift interventions{"}). The location and strength of these interventions do not have to be known and can be estimated from the data. Our method, called BACKSHIFT, only uses second moments of the data and performs simple joint matrix diagonalization, applied to differences between covariance matrices. We give a sufficient and necessary condition for identifiability of the system, which is fulfilled almost surely under some quite general assumptions if and only if there are at least three distinct experimental settings, one of which can be pure observational data. We demonstrate the performance on some simulated data and applications in flow cytometry and financial time series.",

author = "D. Rothenh{\"a}usler and C. Heinze and Peters, {Jonas Martin} and N. Meinshausen",

year = "2015",

language = "English",

pages = "1513--1521",

editor = "{C. Cortes}, {C. } and {Lawrence }, {N.D. } and Lee, {D.D. } and Sugiyama, {M. } and Garnett, {R. }",

booktitle = "Advances in Neural Information Processing Systems 28 (NIPS 2015)",

publisher = "NIPS Proceedings",

}

TY - GEN

T1 - BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions

AU - Rothenhäusler, D.

AU - Heinze, C.

AU - Peters, Jonas Martin

AU - Meinshausen, N.

PY - 2015

Y1 - 2015

N2 - We propose a simple method to learn linear causal cyclic models in the presence of latent variables. The method relies on equilibrium data of the model recorded under a specific kind of interventions ("shift interventions"). The location and strength of these interventions do not have to be known and can be estimated from the data. Our method, called BACKSHIFT, only uses second moments of the data and performs simple joint matrix diagonalization, applied to differences between covariance matrices. We give a sufficient and necessary condition for identifiability of the system, which is fulfilled almost surely under some quite general assumptions if and only if there are at least three distinct experimental settings, one of which can be pure observational data. We demonstrate the performance on some simulated data and applications in flow cytometry and financial time series.

AB - We propose a simple method to learn linear causal cyclic models in the presence of latent variables. The method relies on equilibrium data of the model recorded under a specific kind of interventions ("shift interventions"). The location and strength of these interventions do not have to be known and can be estimated from the data. Our method, called BACKSHIFT, only uses second moments of the data and performs simple joint matrix diagonalization, applied to differences between covariance matrices. We give a sufficient and necessary condition for identifiability of the system, which is fulfilled almost surely under some quite general assumptions if and only if there are at least three distinct experimental settings, one of which can be pure observational data. We demonstrate the performance on some simulated data and applications in flow cytometry and financial time series.

M3 - Article in proceedings

SP - 1513

EP - 1521

BT - Advances in Neural Information Processing Systems 28 (NIPS 2015)

A2 - C. Cortes, C.

A2 - Lawrence , N.D.

A2 - Lee, D.D.

A2 - Sugiyama, M.

A2 - Garnett, R.

PB - NIPS Proceedings

ER -

BACKSHIFT: Learning causal cyclic graphs from unknown shift interventions

Abstract

Fingerprint

Cite this