Abstract:Causal discovery from observational data is a fundamental problem in science. Though the linear non-Gaussian acyclic model (LiNGAM) has shown promising results in various applications, it still faces the following challenges in the data with multiple latent confounders: 1) how to detect the latent confounders and 2) how to uncover the causal relations among observed and latent variables. To address these two challenges, we propose a hybrid causal discovery method for the LiNGAM with multiple latent confounders (MLCLiNGAM). First, we utilize the constraint-based method to learn the causal skeleton. Second, we identify the causal directions, by conducting regression and independence tests on the adjacent pairs in the causal skeleton. Third, we detect the latent confounders with the help of the maximal clique patterns raised by the latent confounders and reconstruct the causal structure with latent variables. Theoretical results show the correctness and efficiency of the algorithms. We conduct extensive experiments on synthetic and real data, which illustrates the efficiency and effectiveness of the proposed algorithms.

What problem does this paper attempt to address?

The paper primarily aims to address the problem of discovering causal structures from observational data in the presence of multiple latent confounders. Specifically, the paper tackles two major challenges faced by the Linear Non-Gaussian Acyclic Model (LiNGAM): 1. **How to detect latent confounders**: Identifying these factors in datasets with multiple latent confounders. 2. **How to reveal causal relationships between observed and latent variables**: Determining the causal relationships between these factors and observed variables after detecting the latent confounders. To solve the above problems, the authors propose a hybrid causal discovery method (MLCLiNGAM), which mainly includes the following three steps: - **Step 1**: Use constraint-based methods (such as the PC-stable algorithm) to construct a causal skeleton, removing edges between independent variables and retaining actual undirected causal relationships. - **Step 2**: Identify partial causal order through regression and independence tests, and determine some causal directions based on the found causal order. - **Step 3**: Detect the positions of latent confounders using the maximum clique pattern and determine the causal relationships between observed variables that are not affected by the same latent confounders. The main contributions of the paper include: 1. Proposing an effective algorithm to estimate the causal structure of observed variables and latent confounders in the presence of multiple latent confounders. 2. Using the maximum clique pattern to detect the positions of latent confounders and proving the effectiveness of this method under certain assumptions. 3. Providing theoretical analysis of the algorithm's correctness and complexity. 4. Validating the effectiveness and efficiency of the proposed method through extensive experiments. In summary, this research aims to solve the problem of causal discovery in datasets with multiple latent confounders through a novel method, providing a powerful tool for understanding the mechanisms behind complex systems.

Causal Discovery in Linear Non-Gaussian Acyclic Model With Multiple Latent Confounders

A Linear Non-Gaussian Acyclic Model for Causal Discovery

A linear non-Gaussian acyclic model for causal discovery.

Causal Discovery with Multi-Domain LiNGAM for Latent Factors

Functional Linear Non-Gaussian Acyclic Model for Causal Discovery

Causal Effect Identification in LiNGAM Models with Latent Confounders

Causal Discovery of Linear Non-Gaussian Causal Models with Unobserved Confounding

TSLiNGAM: DirectLiNGAM Under Heavy Tails

Causal Discovery with Generalized Linear Models through Peeling Algorithms

Causal Discovery under Latent Class Confounding

Learning linear non-Gaussian networks: A new view from matrix identification.

Learning linear acyclic causal model including Gaussian noise using ancestral relationships

Local Causal Discovery with Linear non-Gaussian Cyclic Models

Generalized Independent Noise Condition for Estimating Causal Structure with Latent Variables

Causal Discovery with Mixed Linear and Nonlinear Additive Noise Models: A Scalable Approach

Redefining the Shortest Path Problem Formulation of the Linear Non-Gaussian Acyclic Model: Pairwise Likelihood Ratios, Prior Knowledge, and Path Enumeration

Combining Linear Non-Gaussian Acyclic Model with Logistic Regression Model for Estimating Causal Structure from Mixed Continuous and Discrete Data

Causal Discovery for Linear Mixed Data

Learning to Recover Causal Relationship from Indefinite Data in the Presence of Latent Confounders

ALCM: Autonomous LLM-Augmented Causal Discovery Framework

Causal Discovery in Linear Structural Causal Models with Deterministic Relations