Learning Why Things Change: The Difference-Based Causality Learner

Mark Voortman, Denver Dash, Marek Druzdzel
Proceedings of the 26th Conference on Uncertainty in Artificial Intelligence, PMLR R8:640-649, 2010.

Abstract

In this paper, we present the Difference- Based Causality Learner (DBCL), an algo- rithm for learning a class of discrete-time dy- namic models that represents all causation across time by means of difference equations driving change in a system. We motivate this representation with real-world mechan- ical systems and prove DBCL’s correctness for learning structure from time series data, an endeavour that is complicated by the ex- istence of latent derivatives that have to be detected. We also prove that, under common assumptions for causal discovery, DBCL will identify the presence or absence of feedback loops, making the model more useful for pre- dicting the effects of manipulating variables when the system is in equilibrium. We ar- gue analytically and show empirically the ad- vantages of DBCL over vector autoregression (VAR) and Granger causality models as well as modified forms of Bayesian and constraint- based structure discovery algorithms. Fi- nally, we show that our algorithm can dis- cover causal directions of alpha rhythms in human brains from EEG data.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR8-voortman10a, title = {Learning Why Things Change: The Difference-Based Causality Learner}, author = {Voortman, Mark and Dash, Denver and Druzdzel, Marek}, booktitle = {Proceedings of the 26th Conference on Uncertainty in Artificial Intelligence}, pages = {640--649}, year = {2010}, editor = {Grünwald, Peter and Spirtes, Peter}, volume = {R8}, series = {Proceedings of Machine Learning Research}, month = {08--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r8/main/assets/voortman10a/voortman10a.pdf}, url = {https://proceedings.mlr.press/r8/voortman10a.html}, abstract = {In this paper, we present the Difference- Based Causality Learner (DBCL), an algo- rithm for learning a class of discrete-time dy- namic models that represents all causation across time by means of difference equations driving change in a system. We motivate this representation with real-world mechan- ical systems and prove DBCL’s correctness for learning structure from time series data, an endeavour that is complicated by the ex- istence of latent derivatives that have to be detected. We also prove that, under common assumptions for causal discovery, DBCL will identify the presence or absence of feedback loops, making the model more useful for pre- dicting the effects of manipulating variables when the system is in equilibrium. We ar- gue analytically and show empirically the ad- vantages of DBCL over vector autoregression (VAR) and Granger causality models as well as modified forms of Bayesian and constraint- based structure discovery algorithms. Fi- nally, we show that our algorithm can dis- cover causal directions of alpha rhythms in human brains from EEG data.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Learning Why Things Change: The Difference-Based Causality Learner %A Mark Voortman %A Denver Dash %A Marek Druzdzel %B Proceedings of the 26th Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2010 %E Peter Grünwald %E Peter Spirtes %F pmlr-vR8-voortman10a %I PMLR %P 640--649 %U https://proceedings.mlr.press/r8/voortman10a.html %V R8 %X In this paper, we present the Difference- Based Causality Learner (DBCL), an algo- rithm for learning a class of discrete-time dy- namic models that represents all causation across time by means of difference equations driving change in a system. We motivate this representation with real-world mechan- ical systems and prove DBCL’s correctness for learning structure from time series data, an endeavour that is complicated by the ex- istence of latent derivatives that have to be detected. We also prove that, under common assumptions for causal discovery, DBCL will identify the presence or absence of feedback loops, making the model more useful for pre- dicting the effects of manipulating variables when the system is in equilibrium. We ar- gue analytically and show empirically the ad- vantages of DBCL over vector autoregression (VAR) and Granger causality models as well as modified forms of Bayesian and constraint- based structure discovery algorithms. Fi- nally, we show that our algorithm can dis- cover causal directions of alpha rhythms in human brains from EEG data. %Z Reissued by PMLR on 04 October 2026.
APA
Voortman, M., Dash, D. & Druzdzel, M.. (2010). Learning Why Things Change: The Difference-Based Causality Learner. Proceedings of the 26th Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R8:640-649 Available from https://proceedings.mlr.press/r8/voortman10a.html. Reissued by PMLR on 04 October 2026.

Related Material