Telling cause from effect in deterministic linear dynamical systems

Naji Shajarisales, Dominik Janzing, Bernhard Schoelkopf, Michel Besserve
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:285-294, 2015.

Abstract

Telling a cause from its effect using observed time series data is a major challenge in natural and social sciences. Assuming the effect is generated by the cause through a linear system, we propose a new approach based on the hypothesis that nature chooses the “cause” and the “mechanism generating the effect from the cause” independently of each other. Specifically we postulate that the power spectrum of the “cause” time series is uncorrelated with the square of the frequency response of the linear filter (system) generating the effect. While most causal discovery methods for time series mainly rely on the noise, our method relies on asymmetries of the power spectral density properties that exist even in deterministic systems. We describe mathematical assumptions in a deterministic model under which the causal direction is identifiable. In particular, we show a scenario where the method works but Granger causality fails. Experiments show encouraging results on synthetic as well as real-world data. Overall, this suggests that the postulate of Independence of Cause and Mechanism is a promising principle for causal inference on observed time series.

Cite this Paper


BibTeX
@InProceedings{pmlr-v37-shajarisales15, title = {Telling cause from effect in deterministic linear dynamical systems}, author = {Shajarisales, Naji and Janzing, Dominik and Schoelkopf, Bernhard and Besserve, Michel}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {285--294}, year = {2015}, editor = {Bach, Francis and Blei, David}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/shajarisales15.pdf}, url = { http://proceedings.mlr.press/v37/shajarisales15.html }, abstract = {Telling a cause from its effect using observed time series data is a major challenge in natural and social sciences. Assuming the effect is generated by the cause through a linear system, we propose a new approach based on the hypothesis that nature chooses the “cause” and the “mechanism generating the effect from the cause” independently of each other. Specifically we postulate that the power spectrum of the “cause” time series is uncorrelated with the square of the frequency response of the linear filter (system) generating the effect. While most causal discovery methods for time series mainly rely on the noise, our method relies on asymmetries of the power spectral density properties that exist even in deterministic systems. We describe mathematical assumptions in a deterministic model under which the causal direction is identifiable. In particular, we show a scenario where the method works but Granger causality fails. Experiments show encouraging results on synthetic as well as real-world data. Overall, this suggests that the postulate of Independence of Cause and Mechanism is a promising principle for causal inference on observed time series.} }
Endnote
%0 Conference Paper %T Telling cause from effect in deterministic linear dynamical systems %A Naji Shajarisales %A Dominik Janzing %A Bernhard Schoelkopf %A Michel Besserve %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-shajarisales15 %I PMLR %P 285--294 %U http://proceedings.mlr.press/v37/shajarisales15.html %V 37 %X Telling a cause from its effect using observed time series data is a major challenge in natural and social sciences. Assuming the effect is generated by the cause through a linear system, we propose a new approach based on the hypothesis that nature chooses the “cause” and the “mechanism generating the effect from the cause” independently of each other. Specifically we postulate that the power spectrum of the “cause” time series is uncorrelated with the square of the frequency response of the linear filter (system) generating the effect. While most causal discovery methods for time series mainly rely on the noise, our method relies on asymmetries of the power spectral density properties that exist even in deterministic systems. We describe mathematical assumptions in a deterministic model under which the causal direction is identifiable. In particular, we show a scenario where the method works but Granger causality fails. Experiments show encouraging results on synthetic as well as real-world data. Overall, this suggests that the postulate of Independence of Cause and Mechanism is a promising principle for causal inference on observed time series.
RIS
TY - CPAPER TI - Telling cause from effect in deterministic linear dynamical systems AU - Naji Shajarisales AU - Dominik Janzing AU - Bernhard Schoelkopf AU - Michel Besserve BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-shajarisales15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 285 EP - 294 L1 - http://proceedings.mlr.press/v37/shajarisales15.pdf UR - http://proceedings.mlr.press/v37/shajarisales15.html AB - Telling a cause from its effect using observed time series data is a major challenge in natural and social sciences. Assuming the effect is generated by the cause through a linear system, we propose a new approach based on the hypothesis that nature chooses the “cause” and the “mechanism generating the effect from the cause” independently of each other. Specifically we postulate that the power spectrum of the “cause” time series is uncorrelated with the square of the frequency response of the linear filter (system) generating the effect. While most causal discovery methods for time series mainly rely on the noise, our method relies on asymmetries of the power spectral density properties that exist even in deterministic systems. We describe mathematical assumptions in a deterministic model under which the causal direction is identifiable. In particular, we show a scenario where the method works but Granger causality fails. Experiments show encouraging results on synthetic as well as real-world data. Overall, this suggests that the postulate of Independence of Cause and Mechanism is a promising principle for causal inference on observed time series. ER -
APA
Shajarisales, N., Janzing, D., Schoelkopf, B. & Besserve, M.. (2015). Telling cause from effect in deterministic linear dynamical systems. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:285-294 Available from http://proceedings.mlr.press/v37/shajarisales15.html .

Related Material