基于级联加性噪声模型的因果结构学习算法

doi:10.19678/j.issn.1000-3428.0060176

摘要/Abstract

摘要： 面向观察数据的因果结构学习是很多学科的基本问题。在局部观测的现实应用场景中，隐变量的存在对现有基于约束的方法和基于因果函数的方法带来众多挑战。虽然现有级联非线性加性噪声模型可以解决隐藏中间变量场景的因果方向推断问题，然而该方法并不适用于因果网络学习。为此，我们提出了一种基于级联非线性加性噪声模型的因果结构学习算法。在该算法中，推广了级联非线性加性噪声模型在多个因果隐变量下的适用性，进而提出了两阶段因果结构学习方法。首先，进行因果骨架学习，然后基于级联非线性加性噪声模型推断因果方向。实验结果表明本文提出方法在虚拟结构和真实结构下具有更高的准确度。

Abstract: 】】Learning causal structure from observational data is a basic problem in many disciplines. In the real application scenarios, the existence of hidden variables brings many challenges to the existing constraint-based methods and functional-based causal methods. Although the existing cascaded nonlinear additive noise model can discover the causal direction in the presence of the hidden intermediate variables, this method is not suitable for causal structure learning. For this reason, we propose a mixed causal structure learning algorithm based on cascaded nonlinear additive noise model. In this algorithm, we extend the cascaded nonlinear additive noise model under multiple causal hidden variables. Then a two-stage causal structure learning framework is proposed. The first stage is to learn the causal skeleton, and the second stage is to infer the direction of the causal pair based on the cascaded nonlinear additive noise model. The experimental results show that the proposed method has higher accuracy on both synthetic and real-world structures.

乔杰 , 蔡瑞初 , 郝志峰 . 基于级联加性噪声模型的因果结构学习算法[J]. 计算机工程, doi: 10.19678/j.issn.1000-3428.0060176.

QIAO Jie , CAI Ruichu , HAO Zhifeng . Causal Structure Learning Algorithm with Cascade Nonlinear Additive Noise Model[J]. Computer Engineering, doi: 10.19678/j.issn.1000-3428.0060176.

参考文献

[1] SPIRTES P, GLYMOUR C N, SCHEINES R. Causation, prediction, and search[M]. USA: MIT press, 2000.
[2] PEARL J. Causality: models, reasoning and inference[M]. England: Cambridge university press, 2009.
[3] CAI Ruichu, CHEN Wei, ZHANG Kun, et al. A Survey on Non-Temporal Series Observational Data based Causal Discovery[J]. Chinese Journal of Computers, 2017, 40(06): 1470-1490. 蔡瑞初, 陈薇, 张坤, 等. 基于非时序观察数据的因果关系发现综述[J]. 计算机学报, 2017, 40(06): 1470–1490.
[4] CAI R, ZHANG Z, HAO Z, et al. Understanding social causalities behind human action sequences[J]. IEEE Trans on neural networks and learning systems, IEEE, 2016, 28(8): 1801–1813.
[5] RUNGE J, BATHIANY S, BOLLT E, et al. Inferring causation from time series in Earth system sciences[J]. Nature communications, 2019, 10(1): 1–13.
[6] CAI R, ZHANG Z, HAO Z. Causal gene identification using combinatorial V-structure search[J]. Neural Networks, 2013, 43(1): 63–71.
[7] HERNÁN M A, ROBINS J M. Causal inference[M]. USA: CRC Boca Raton, FL;, 2010.
[8] MOOIJ J M, PETERS J, JANZING D, et al. Distinguishing cause from effect using observational data: methods and benchmarks[J]. The Journal of Machine Learning Research, 2016, 17(1): 1103–1204.
[9] COLOMBO D, MAATHUIS M H, KALISCH M, et al. Learning high-dimensional directed acyclic graphs with latent and selection variables[J]. The Annals of Statistics, JSTOR, 2012, 40(1): 294–321.
[10] OGARRIO J M, SPIRTES P, RAMSEY J. A Hybrid Causal Search Algorithm for Latent Variable Models[C]//Probabilistic Graphical Models - Eighth International Conference, PGM 2016, Lugano, Switzerland, September 6-9, 2016. Proceedings. 2016, 52: 368–379.
[11] CAI R, QIAO J, ZHANG K, et al. Causal discovery with cascade nonlinear additive noise models[C]// Proceedings of the 28th International Joint Conference on Artificial Intelligence. Palo Alto, CA: AAAI Press, 2019: 1609–1615.
[12] BRESSLER S L, SETH A K. Wiener–Granger causality: a well established methodology[J]. Neuroimage, Elsevier, 2011, 58(2): 323–329.
[13] ZHANG Yijie, LI Peifeng, ZHU Qiaoming. Joint Identification Method for Temporal and Causal Relations of Events[J]. Computer Engineering, 2020, 46(07): 65–71. 张义杰, 李培峰, 朱巧明. 面向事件时序与因果关系的联合识别方法[J]. 计算机工程, 2020, 46(07): 65–71.
[14] LAM W, BACCHUS F. Learning Bayesian belief networks: An approach based on the MDL principle[J]. Computational intelligence, Wiley Online Library, 1994, 10(3): 269–293.
[15] ANDERSSON S A, MADIGAN D, PERLMAN M D, et al. A characterization of Markov equivalence classes for acyclic digraphs[J]. The Annals of Statistics, Institute of Mathematical Statistics, 1997, 25(2): 505–541.
[16] SHIMIZU S, HOYER P O, HYVÄRINEN A, et al. A linear non-Gaussian acyclic model for causal discovery[J]. Journal of Machine Learning Research, 2006, 7(Oct): 2003–2030.
[17] JIANG Feng, ZHU Hui-sheng, WANG Wei. Estimation Algorithm for Non-Gaussian Acyclic Causal Model with Latent Variables[J]. Computer Engineering, 2010, 36(09): 178–180. 姜枫, 朱辉生, 汪卫. 含隐变量非高斯无环因果模型的估计算法[J]. 计算机工程, 2010, 36(09): 178–180.
[18] HOYER P, JANZING D, MOOIJ J M, et al. Nonlinear causal discovery with additive noise models[C]// Proceedings of the Twenty-Second Annual Conference on Neural Information Processing Systems. New York,NY: NIPS, 2008, 21: 689–696.
[19] ZHANG K, HYVÄRINEN A. On the Identifiability of the Post-Nonlinear Causal Model[C]// UAI 2009, Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. Corvallis, USA: AUAI Press, 2009: 647–655.
[20] HOYER P O, SHIMIZU S, KERMINEN A J, et al. Estimation of causal effects using linear non-Gaussian causal models with hidden variables[J]. International Journal of Approximate Reasoning, 2008, 49(2): 362–378.
[21] KINGMA D P, WELLING M. Auto-Encoding Variational Bayes[C]// the 2nd International Conference on Learning Representations, ICLR. 2014.
[22] HORNIK K, STINCHCOMBE M B, WHITE H. Multilayer feedforward networks are universal approximators[J]. Neural Networks, 1989, 2(5): 359–366.
[23] GÁMEZ J A, MATEO J L, PUERTA J M. Learning Bayesian networks by hill climbing: efficient methods based on progressive restriction of the neighborhood[J]. Data Mining and Knowledge Discovery, Springer, 2011, 22(1–2): 106–148.
[24] TSAMARDINOS I, BROWN L E, ALIFERIS C F. The max-min hill-climbing Bayesian network structure learning algorithm[J]. Machine learning, 2006, 65(1): 31–78.