Estimating the decomposition of predictive information in multivariate systems.

Area of Science:

• Complex Systems Analysis
• Information Theory
• Time Series Analysis

Background:

• Understanding complex systems requires analyzing information storage and transfer within multivariate time series.
• Existing methods face challenges like the curse of dimensionality and bias in information-theoretic quantity estimation.

Purpose of the Study:

• To present a model-free framework for estimating information storage and transfer in multivariate dynamical processes.
• To address the curse of dimensionality and bias in information-theoretic calculations.

Main Methods:

• Employs a nonuniform embedding scheme to select relevant past components of a multivariate process.
• Utilizes a nearest-neighbor technique to compute information-theoretic quantities, compensating for dimensionality bias.
• Estimates prediction entropy, storage entropy, transfer entropy, and partial transfer entropy.

Main Results:

• The proposed framework demonstrates superior performance over traditional uniform embedding estimators in simulations.
• Successfully applied to physiological time series, providing interpretable information decompositions of cardiovascular, cardiorespiratory, and brain-heart dynamics.

Conclusions:

• The novel framework offers a robust and accurate method for analyzing information dynamics in complex systems.
• Provides valuable insights into physiological interactions, enhancing our understanding of system behavior.