Reconstructing the evolution history of networked complex systems.

Area of Science:

• Complex Systems Science
• Network Science
• Computational Biology
• Machine Learning

Background:

• Complex systems' evolution is encoded in their functional properties.
• Understanding network formation is crucial across various scientific domains.
• Previous theories struggle to collectively explain network evolution features.

Purpose of the Study:

• To extract the historical formation processes of networked complex systems.
• To demonstrate the scientific value of recovered evolution processes.
• To investigate the feasibility of network evolution restoration.

Main Methods:

• Application of machine learning algorithms.
• Analysis of diverse networked systems (e.g., protein-protein interaction, ecological, social networks).
• Evaluation of model performance against random link ordering.

Main Results:

• Successfully extracted evolution processes for multiple network types.
• Revealed key co-evolution features: preferential attachment, community structure, local clustering, and degree-degree correlation.
• Demonstrated that high-fidelity restoration is achievable with ML models slightly outperforming random guessing.

Conclusions:

• The historical evolution of complex networks is largely recoverable using machine learning.
• Recovered evolution processes offer significant scientific insights and applications.
• Network evolution restoration is a generally feasible approach for empirical networks.