Pseudolikelihood decimation algorithm improving the inference of the interaction network in a general class of Ising...

Area of Science:

• Statistical Physics
• Network Science
• Computational Physics

Background:

• Inferring interaction network topology is crucial for understanding complex systems.
• The pseudolikelihood method (PLM) is effective at high temperatures but less reliable at low temperatures.
• Low-temperature PLM analysis requires subjective parameter choices, limiting its applicability.

Purpose of the Study:

• To develop a novel, automated method for inferring interaction network topology in Ising models.
• To overcome the limitations of standard PLM at low temperatures.
• To provide a robust and user-independent approach for network inference.

Main Methods:

• A decimation procedure based on PLM is introduced.
• The method recursively sets insignificant couplings to zero.
• Stopping criterion based on pseudolikelihood variation prevents removal of relevant couplings.

Main Results:

• The new automated method demonstrates superior performance compared to standard PLM.
• Effective inference across diverse Ising models, including random graphs, lattices, ferromagnets, and spin glasses.
• Eliminates the need for subjective parameter tuning, enhancing reproducibility.

Conclusions:

• The proposed automated decimation procedure offers a more reliable and objective approach to inferring Ising model network topology.
• This advancement is particularly beneficial for low-temperature analyses where standard PLM struggles.
• The method's robustness across various model types suggests broad applicability in statistical physics and network science.