Learning of networked spreading models from noisy and incomplete data.

Area of Science:

• Network science
• Computational epidemiology
• Machine learning

Background:

• Learning parameters of spreading dynamics is crucial but faces challenges like unknown network structures, noisy data, and missing observations.
• Accurate learning often requires minimizing sample size through efficient incorporation of prior information.

Purpose of the Study:

• To introduce a universal learning method for spreading dynamics that addresses common challenges in real-world data.
• To reconstruct both the network structure and the parameters of a spreading model using available prior knowledge.

Main Methods:

• A scalable dynamic message-passing technique is employed as the core of the universal learning method.
• The algorithm integrates prior knowledge about the model and data to overcome learning obstacles.

Main Results:

• The proposed method effectively addresses challenges such as unknown network structures, noisy data, and missing observations.
• It successfully reconstructs both network topology and spreading model parameters.
• The method exhibits linear computational complexity with respect to key model parameters, ensuring scalability.

Conclusions:

• The developed universal learning method offers a scalable and efficient solution for parameter learning in spreading dynamics.
• It demonstrates the ability to handle complex real-world data scenarios, including incomplete information and unknown network structures.
• The technique's scalability makes it suitable for analyzing large-scale network instances.