Multiplex network disintegration strategy inference based on deep network representation learning.

Area of Science:

• Network Science
• Computer Science
• Artificial Intelligence

Background:

• Multiplex networks, with coupled nodes across layers, present unique challenges for disintegration due to complex inter-layer interactions.
• Existing disintegration strategies are often approximate or heuristic, lacking effectiveness and timeliness for complex network analysis.

Purpose of the Study:

• To develop a novel deep learning framework, MINER (Multiplex network disintegration strategy Inference based on deep NEtwork Representation learning), for effective and timely multiplex network disintegration.
• To address the limitations of traditional approximate and heuristic methods in analyzing the complex dynamics of multiplex networks.

Main Methods:

• MINER employs a deep network representation learning approach, transforming disintegration strategy inference into an encoding-decoding process.
• The encoding process utilizes an attention mechanism to capture inter-layer node coupling, while reinforcement learning evaluates disintegration actions in the decoding phase.

Main Results:

• The trained MINER model demonstrates direct transferability and applicability to multiplex networks of varying scales.
• MINER achieves excellent performance even when extended to scenarios with node attack cost constraints.

Conclusions:

• MINER offers a novel and effective deep learning-based approach for multiplex network disintegration.
• This framework provides a new paradigm for understanding and utilizing multiplex networks in complex systems analysis.