Low-dimensional approximation searching strategy for transfer entropy from non-uniform embedding

Area of Science:

• Dynamical systems analysis
• Information theory
• Causality inference

Background:

• Transfer entropy is widely used for inferring causal relationships in dynamical subsystems.
• High-dimensional data presents challenges for accurate transfer entropy calculations.
• Existing methods can be computationally intensive and less accurate with increasing data complexity.

Purpose of the Study:

• To develop a more accurate and efficient method for transfer entropy-based causality inference.
• To address the limitations of high-dimensional data in non-uniform embedding.
• To improve the sensitivity and specificity of causality detection in complex systems.

Main Methods:

• Decomposition of high-dimensional conditional mutual information using low-dimensional approximations.
• Application in the non-uniform embedding search procedure for significant variables at different lags.
• Validation through a series of simulation experiments to assess performance.

Main Results:

• The proposed method demonstrates improved statistical accuracy in multivariate causality analysis.
• Low-dimensional approximations enhance the performance of transfer entropy compared to previous algorithms.
• The method exhibits increased efficiency, particularly with growing data length.

Conclusions:

• Low-dimensional conditional mutual information is effective for improving transfer entropy accuracy.
• The novel approach offers a more robust and efficient tool for causality inference in dynamical systems.
• This method provides a significant advancement for analyzing complex, high-dimensional data.