Kernel-based Conditional Independence Test and Application in Causal Discovery
By Kun Zhang, Jan Peters, Dominik Janzing, Bernhard Schölkopf. 2011
738 citations recorded by the source corpus.
Conditional independence testing is an important problem, especially in Bayesian network learning and causal discovery. Due to the curse of dimensionality, testing for conditional independence of continuous variables is particularly challenging. We propose a Kernel-based Conditional Independence test (KCI-test), by constructing an appropriate test statistic and deriving its asymptotic distribution under the null hypothesis of conditional independence. The proposed method is computationally efficient and easy to implement. Experimental results show that it outperforms other methods, especially when the conditioning set is large or the sample size is not very large, in which case other methods encounter difficulties.
Subfields: Artificial Intelligence, Computational Theory and Mathematics, Causal Inference & Discovery
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
- Bernhard Schölkopf
- Yoshua Bengio
- Sergey Levine
- David Lopez-Paz
- Eric P. Xing
- Dacheng Tao
- Quoc V. Le
- Mohit Bansal
In Computational Theory and Mathematics
- Bernhard Schölkopf
- Arthur Gretton
- Kun Zhang
- Dino Sejdinovic
- Kenji Fukumizu
- Bharath K. Sriperumbudur
- David Lopez-Paz
- Clark Glymour
In Causal Inference & Discovery
- Bernhard Schölkopf
- Jan Peters
- Kun Zhang
- Dominik Janzing
- David Lopez-Paz
- Yoshua Bengio
- Peter Buhlmann
- Stefan Bauer
Citations in the corpus
- Differentiable Causal Discovery from Interventional Data (2020, methodology): “For these methods, we always pick the most advantageous independence test: partial correlation test for Gaussian linear data and KCI-test [42] for nonlinear data.”
- Toward Causal Representation Learning (2021, background): “Recent progress in (conditional) independence testing heavily relies on kernel function classes to represent probability distributions in reproducing kernel Hilbert spaces [43], [61], [74], [91], [92], [193], [280].”
- Kernel Mean Embedding of Distributions: A Review and Beyonds (2016, methodology): “§3.5). Injectivity of P →µP makes it suitable for applications that require a unique characterization of distributions such as twosample homogeneity tests (Gretton et al. 2012a, Fukumizu et al. 2008, Zhang et al. 2011, Doran et al. 2014). Moreover, using the kernel mean representation, most learning”
- Towards Causal Representation Learning (2021, neutral): “progress in (conditional) independence testing heavily relies on kernel function classes to represent probability distributions in reproducing kernel Hilbert spaces [90, 91, 73, 278, 60, 191, 42].”
- Causality for Machine Learning (2019, methodology): “Recent progress in (conditional) independence testing heavily relies on kernel function classes to represent probability distributions in reproducing kernel Hilbert spaces (Gretton et al., 2005a,b; Fukumizu et al., 2008; Zhang et al., 2011; Pfister et al., 2018b; Chalupka et al., 2018).”
- Learning Neural Causal Models from Unknown Interventions (2019, methodology): “…based on observational data (Chickering, 2002; Tsamardinos et al., 2006; Goudet et al., 2017; Hauser & Bühlmann, 2012; Spirtes et al., 2000; Sun et al., 2007; Zhang et al., 2012; Shimizu et al., 2006; Hoyer et al., 2009; Daniusis et al., 2012; Budhathoki & Vreeken, 2017; Kalainathan et al., 2018).”
- Causal Discovery from Heterogeneous/Nonstationary Data (2019, methodology): “We used Gaussian kernels, and the hyperparameters, e.g. the kernel width, were set with empirical values; please refer to (Zhang et al., 2011) for details.”
- The hardness of conditional independence testing and the generalised covariance measure (2018, methodology): “In the following, we compare the GCM with three other conditional independence tests: KCI [51] with its implementation from CondIndTests [26], and the residual prediction test [26, 46].”
- Fast Conditional Independence Test for Vector Variables with Large Sample Sizes (2018, methodology): “Fukumizu et al. [2008] and Zhang et al. [2012] report Type I and Type II errors on a number of datasets.”
- Behind Distribution Shift: Mining Driving Forces of Changes and Causal Arrows (2017, positive): “We first learned causal skeletons by the procedure in Algorithm 1, with PC search [9] and kernelbased conditional independence (KCI) test [12].”
- Causal Discovery from Nonstationary/Heterogeneous Data: Skeleton Estimation and Orientation Determination (2017, methodology): “In this work, we use the kernel-based conditional independence test (KCI-test [Zhang et al. , 2011]) to capture the dependence on C in a nonparametric way.”
- Computing functions of random variables via reproducing kernel Hilbert space representations (2015, neutral): “Wecanmapconditional distributions andperform Bayesian updates (Fukumizu et al. 2008, 2013; Zhang et al. 2011), andwecan connect kernelmaps toFourier optics, leading to a physical realization as Fraunhofer diffraction (Harmeling et al.”
- A Wild Bootstrap for Degenerate Kernel Tests (2014, methodology): “…tests based on distribution embeddings into reproducing kernel Hilbert spaces have been applied in many contexts, including two sample testing [17, 14, 31], tests of independence [16, 32, 4], tests of conditional independence [13, 32], and tests for higher order (Lancaster) interactions [23].”
- Quasi-Monte Carlo Feature Maps for Shift-Invariant Kernels (2014, background): “, 2013), causal modeling (Zhang et al., 2011) and many more.”
- A low variance consistent test of relative dependency (2014, background): “The statistics for such tests are diverse, and include kernel measures of covariance (Gretton et al., 2008; Zhang et al., 2011) and correlation (Dauxois & Nkiet, 1998; Fukumizu et al.”
- A Kernel Test for Three-Variable Interactions (2013, methodology): “a two-variable based test; Test for X ⊥ Y |Z from [18]”
- Hypothesis testing using pairwise distances and associated kernels (2012, methodology): “Second, results from Gretton et al. (2009a); Zhang et al. (2011) may be applied to get consistent two-sample and independence tests for the energy distance, without using bootstrap, which perform much better than the upper bound proposed by Székely et al. (2007) as an alternative to the bootstrap.”