Causal Discovery and Inference: Concepts and Recent Methodological Advances

By Peter Spirtes, Kun Zhang. Applied Informatics, 2016

403 citations recorded by the source corpus.

This paper aims to give a broad coverage of central concepts and principles involved in automated causal inference and emerging approaches to causal discovery from i.i.d data and from time series. After reviewing concepts including manipulations, causal models, sample predictive modeling, causal predictive modeling, and structural equation models, we present the constraint-based approach to causal discovery, which relies on the conditional independence relationships in the data, and discuss the assumptions underlying its validity. We then focus on causal discovery based on structural equations models, in which a key issue is the identifiability of the causal structure implied by appropriately defined structural equation models: in the two-variable case, under what conditions (and why) is the causal direction between the two variables identifiable? We show that the independence between the error term and causes, together with appropriate structural constraints on the structural equation, makes it possible. Next, we report some recent advances in causal discovery from time series. Assuming that the causal relations are linear with nonGaussian noise, we mention two problems which are traditionally difficult to solve, namely causal discovery from subsampled data and that in the presence of confounding time series. Finally, we list a number of open questions in the field of causal discovery and inference.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/488f675419b6692e388e57e1324db87a82daa895

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Signal Processing

Citations in the corpus

Generalized Score Functions for Causal Discovery (2018, methodology): “For surveys of some of the recently proposed methods, one may refer to [29] and [36].”
Causal inference in partially linear structural equation models (2016, background): “A widespread specification of PLSEMs are linear Gaussian SEMs, which have the same identifiability properties as the general SEMs: without additional assumptions they are unidentifiable, whereas under faithfulness, their distribution equivalence class equals the Markov equivalence class; see, for ex”
Automated Machine Learning: Methods, Systems, Challenges (2019, neutral): “Wouldn’t is be nice to have a universal correlation coefficient that allows us to test (simply) whether two variables are dependent (not talking about causality, just dependence)? Does this exist? Here is a brief discussion of the subject [11, 24, 63]:”
A Survey of Learning Causality with Data: Problems and Methods (2018, background): “Confounding bias can lead to wrong causal conclusions [131].”
Causal network reconstruction from time series: From theoretical assumptions to practical estimation. (2018, background): “Subsampled time series are an active area of research (Smirnov, 2013; Barnett and Seth, 2015; Spirtes and Zhang, 2016), to some extent sub-sampled time series graphs can be identified as addressed in Gong et al. (2015) and Hyttinen et al. (2016).”
Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasets (2020, background): “…[Chickering, 2002], classical Granger causality (GC) [Granger, 1969], and the more recent restricted structural causal models (SCM) framework [Peters et al., 2017, Spirtes and Zhang, 2016], conditional independence (CI) based network learning algorithms [Spirtes et al., 2000] form a main pillar.”
Structural Agnostic Modeling: Adversarial Learning of Causal Graphs (2018, methodology): “Such methods, taking into account the full information from the observational data (Spirtes and Zhang, 2016) such as data asymmetries induced by the causal directions, have been proposed and primarily applied to the bivariate DAG case4, referred to as cause-effect pair problem (Hoyer et al., 2009;…”
Causal Confusion in Imitation Learning (2019, background): “Causal inference is the general problem of deducing cause-effect relationships among variables [52, 38, 40, 50, 10, 51].”
Explainable and Interpretable Models in Computer Vision and Machine Learning (2017, neutral): “The abovementioned score-based and constraint-based methods do not take into account the full information from the observational data (Spirtes and Zhang, 2016), such as data asymmetries induced by the causal directions.”
A Survey of Learning Causality with Data (2018, methodology): “Spirtes and Zhang [146] summarize methods for learning causal relations on both i.i.d. and time series data with a focus on several semi-parametric score based methods.”
A Survey on Causal Inference (2020, background): “Different from causal effect estimation, causal structure learning address a different class of problems, which is out of our survey’s scope; see [148] for more information.”
Detecting and quantifying causal associations in large nonlinear time series datasets (2017, positive): “Recently, a growing body of literature addresses the inference of causality without relying on time lags (59, 60), which could help to determine causal directionality for contemporaneous links.”
Detecting crosstalk errors in quantum information processors (2019, background): “tional dependency relationships between the operational variables [55], although we have not explored these.”
D’ya Like DAGs? A Survey on Structure Learning and Causal Discovery (2021, background): “, [69, 92, 230]), however, these reviews cover combinatoric approaches to causal discovery, hence why we primarily focus on the recent flurry of developments in continuous optimization approaches.”
Inferring causation from time series in Earth system sciences (2019, background): “Nevertheless, as illustrated in Box 1 GC is limited to lagged causal dependencies and, furthermore, has known de ﬁ ciencies in the presence of sub-sampled time series and other issues 38 .”
Causal Discovery with Attention-Based Convolutional Neural Networks (2019, background): “Structural Equation Model approaches assume that a causal system can be represented by a Structural Equation Model (SEM) that describes a variable Xj as a function of other variables X−j, and an error term eX to account for additive noise such that X := f (X−j, eX) [29].”
Causal Discovery with General Non-Linear Relationships using Non-Linear ICA (2019, background): “While randomized con-trol studies are the gold standard, such an approach is unfeasible or unethical in many scenarios (Spirtes and Zhang, 2016).”