Machine learning dismantling and early-warning signals of disintegration in complex systems

Area of Science:

• Network science
• Complex systems analysis
• Machine learning applications

Background:

• Natural and artificial systems exhibit complex topologies affecting robustness to perturbations.
• Network dismantling, identifying critical units to disintegrate a network, is an NP-hard computational problem often addressed with heuristics.

Purpose of the Study:

• To develop a machine learning model for efficient network dismantling.
• To identify higher-order topological patterns in complex systems.
• To quantify systemic risk and detect early-warning signals of collapse.

Main Methods:

• Training a machine learning model on smaller systems to identify dismantling strategies.
• Applying the trained model to large-scale social, infrastructural, and technological networks.
• Developing a probabilistic assessment of attack impact on system disintegration.

Main Results:

• The machine learning model achieved more efficient network dismantling than human-based heuristics.
• The model identified higher-order topological patterns crucial for network structure.
• A quantitative method for assessing systemic risk and detecting collapse precursors was established.

Conclusions:

• Machine-assisted analysis offers a powerful tool for understanding and managing complex systems.
• This approach enhances policy and decision-making by quantifying system fragility and response to shocks.
• The findings demonstrate the potential of AI in predicting and mitigating large-scale system failures.