Using machine learning to assess short term causal dependence and infer network links.

Area of Science:

• Complex systems
• Dynamical systems theory
• Machine learning

Background:

• Inferring causal relationships from time-series data is crucial for understanding complex systems.
• Traditional methods often struggle with the inherent noise and complexity of real-world data.

Purpose of the Study:

• To introduce and validate a novel machine learning technique for inferring short-term causal dependencies in unknown dynamical systems.
• To leverage short-term prediction capabilities of machine learning for causal inference.

Main Methods:

• Utilizing reservoir computing, a type of machine learning, to estimate the Jacobian matrix elements of the dynamical flow.
• Applying the technique to time-series measurements of state variables from an unknown dynamical system.
• Testing the method on a network of interacting dynamical nodes for link inference.

Main Results:

• The machine learning technique successfully infers causal dependencies in simulated dynamical systems.
• Dynamical noise was found to significantly enhance the effectiveness of the causal inference technique.
• Observational noise was observed to degrade the performance of the causal inference method.

Conclusions:

• The proposed machine learning approach offers a promising method for causal inference in dynamical systems.
• The interplay between dynamical and observational noise is critical for the success of causal inference in practical applications.
• This technique has potential applications in various fields requiring the analysis of complex time-series data.