Higher Order Cumulants-Based Method for Direct and Efficient Causal Discovery.

Area of Science:

• Causal inference and machine learning
• Statistical modeling and analysis

Background:

• Causal discovery is crucial for prediction and decision-making.
• Existing methods often rely on computationally intensive independence tests.
• A need exists for more efficient causal discovery techniques.

Purpose of the Study:

• To propose a direct and computationally efficient method for causal discovery.
• To determine causal relationships between two observed variables in the linear non-Gaussian case.
• To introduce practical methods for high-dimensional causal discovery.

Main Methods:

• Leveraging joint cumulants to infer variable (in)dependence.
• Introducing the 'cause difference criterion' based on cumulant products.
• Developing high-order cumulant (HC) and HC-LiNGAM methods for causal discovery.

Main Results:

• The cause difference criterion effectively infers causal relationships.
• Proposed HC and HC-LiNGAM methods are suitable for high-dimensional data.
• Theoretical analyses confirm the identifiability of the criteria and methods.
• Experimental results demonstrate superior performance compared to existing methods.

Conclusions:

• The cause difference criterion offers a direct and efficient approach to causal discovery.
• The proposed HC and HC-LiNGAM methods provide practical solutions for complex datasets.
• This work advances causal discovery by offering a computationally tractable alternative.