Information Transfer in Linear Multivariate Processes Assessed through Penalized Regression Techniques: Validation...

Area of Science:

• Complex Systems Analysis
• Information Dynamics
• Network Science

Background:

• Information dynamics dissects network information into computation elements: generation, storage, transfer, and modification.
• Existing methods for information transfer/modification rely on vector autoregressive models and state-space representations.
• Standard methods face estimation challenges with limited data and high-dimensional time series.

Purpose of the Study:

• To improve the computation of information transfer and modification measures.
• To address estimation problems in vector autoregressive (VAR) models with limited data.
• To introduce a penalized regression approach for more robust network analysis.

Main Methods:

• Replaced Ordinary Least Squares (OLS) with Least Absolute Shrinkage and Selection Operator (LASSO) regression.
• Applied the VAR-SS-LASSO approach to simulated Gaussian systems and human physiological data.
• Quantified information transfer and modification in networks of interacting dynamical systems.

Main Results:

• LASSO regression accurately reconstructs network topology and information transfer patterns, even with data paucity.
• The VAR-SS-LASSO method successfully identified physiologically plausible interactions in human brain and peripheral data.
• Demonstrated the effectiveness of the new approach in complex network analysis.

Conclusions:

• The proposed VAR-SS-LASSO method offers a robust solution for analyzing information dynamics in complex networks.
• This approach overcomes limitations of traditional methods, particularly in data-scarce scenarios.
• The findings pave the way for novel applications in Network Physiology and related fields.