Abstract
Recent progress in the field of machine learning suggests that the joint distribution of two variables X, Y sometimes contains information about the underlying causal structure, e.g., whether X is the case of Y or Y the cause of X, given that exactly one of the alternatives is true. To provide an idea about these statistical asymmetries I show some intuitive examples, both hypothetical toy scenarios as well as scatter plots from real world data. I sketch some recent approaches to infer the causal direction based on these asymmetries and give some pointers to physics literature that relate them to the thermodynamic arrow of time.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
If X and Y refer to quantum measurements at two particles coming from a common source, there may not be such a variable Z that screens off the dependences in the sense of (1) due to quantum entanglement and the violation of Bell’s inequality. Nevertheless, one may visualize the scenario as causal relation of type (b) and replace Z with some joint quantum state | ψ〉 to indicate that the common cause is no longer a random variable.
- 2.
This is different, however, in quantum theory: a joint operator describing two systems that are correlated by a common cause is positive, while an operator describing correlations of cause and effect has positive partial transpose [4]. Therefore, one can sometimes tell apart cases (a) and (c) versus case (b) in Fig. 1, see [5].
- 3.
Here we have implicitly assumed that the joint probability distribution has a joint density to simplify explanations.
- 4.
Note that causal faithfulness has also been discussed in the context of Bell’s inequality: Wood and Spekkens [8] argue that classical explanations of the EPR-scenario (with superluminal communication, for instance) require similar fine-tuning of parameters.
- 5.
Note that vanishing correlation is not enough because linear least square regression automatically yields uncorrelated residuals. Instead, one needs a statistical dependence test that is able to detect higher order dependences.
References
H. Reichenbach, The Direction of Time (University of California Press, Berkeley, 1956)
Database with Cause-Effect Pairs, https://webdav.tuebingen.mpg.de/cause-effect/. Copyright Information for each Cause-Effect Pair is contained in the respective description file
J.M. Mooij, J. Peters, D. Janzing, J. Zscheischler, B. Schölkopf, Distinguishing cause from effect using observational data: methods and benchmarks. J. Mach. Learn. Res. 17(32), 1–102 (2016)
M. Leifer, R. Spekkens, Towards a formulation of quantum theory as a causally neutral theory of Bayesian inference. Phys. Rev. A(88), 052130 (2013)
K. Ried, M. Agnew, L. Vermeyden, D. Janzing, R. Spekkens, K. Resch, A quantum advantage for inferring causal structure. Nat. Phys. 11(5), 414–420 (2015)
P. Spirtes, C. Glymour, R. Scheines, Causation, Prediction, and Search. Lecture Notes in Statistics (Springer, New York, 1993)
J. Pearl, Causality (Cambridge University Press, Cambridge, 2000)
C. Wood, R. Spekkens, The lesson of causal discovery algorithms for quantum correlations: causal explanations of bell-inequality violations require fine-tuning. New J. Phys. 17(3), 033002 (2015)
D. Janzing, X. Sun, B. Schölkopf, Distinguishing cause and effect via second order exponential models. http://arxiv.org/abs/0910.5561 (2009)
D. Janzing, B. Schölkopf, Causal inference using the algorithmic Markov condition. IEEE Trans. Inform. Theory 56(10), 5168–5194 (2010)
J. Lemeire, D. Janzing, Replacing causal faithfulness with algorithmic independence of conditionals. Minds Mach. 23(2), 227–249 (2012)
M. Li, P. Vitányi. An Introduction to Kolmogorov Complexity and Its Applications (Springer, New York, 1997) (3rd edition: 2008).
D. Janzing, R. Chaves, B. Schölkopf, Algorithmic independence of initial condition and dynamical law in thermodynamics and causal inference. New J. Phys. 18(093052), 1–13 (2016)
J. Mooij, D. Janzing, J. Peters, B. Schölkopf, Regression by dependence minimization and its applications to causal inference in additive noise models, in Proceedings of the 26th International Conference on Machine Learning, Montreal, ACM International Conference Proceeding Series, New York, pp. 745–752 (2009)
Y. Kano, S. Shimizu, Causal inference using nonnormality, in Proceedings of the International Symposium on Science of Modeling, the 30th Anniversary of the Information Criterion, Tokyo, pp. 261–270 (2003)
P. Hoyer, D Janzing, J. Mooij, J. Peters, B Schölkopf, Nonlinear causal discovery with additive noise models, in Proceedings of the Conference Neural Information Processing Systems (NIPS) 2008, ed. by D. Koller, D. Schuurmans, Y. Bengio, L. Bottou (MIT Press, Vancouver, 2009)
J.M. Mooij, J. Peters, D. Janzing, J. Zscheischler, B. Schölkopf, Distinguishing cause from effect using observational data: methods and benchmarks. J. Mach. Learn. Res. 17(32), 1–102 (2016)
D. Janzing, B. Steudel, Justifying additive-noise-based causal discovery via algorithmic information theory. Open Syst. Inform. Dyn. 17(2), 189–212 (2010)
J. Peters, D. Janzing, A. Gretton, B. Schölkopf, Detecting the direction of causal time series, in Proceedings of the 26th International Conference on Machine Learning, ACM International Conference Proceeding Series, Montreal, vol. 382 (ACM, New York, 2009), pp. 801–808
D. Janzing, On the entropy production of time series with unidirectional linearity. J. Stat. Phys. 138, 767–779 (2010)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this chapter
Cite this chapter
Janzing, D. (2017). Statistical Asymmetries Between Cause and Effect. In: Renner, R., Stupar, S. (eds) Time in Physics. Tutorials, Schools, and Workshops in the Mathematical Sciences . Birkhäuser, Cham. https://doi.org/10.1007/978-3-319-68655-4_8
Download citation
DOI: https://doi.org/10.1007/978-3-319-68655-4_8
Published:
Publisher Name: Birkhäuser, Cham
Print ISBN: 978-3-319-68654-7
Online ISBN: 978-3-319-68655-4
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)