Fast rule identification and neighborhood selection for cellular automata

Area of Science:

• Computational Science
• Complex Systems
• Machine Learning

Background:

• Cellular automata (CA) rule extraction from data is an underexplored area.
• Existing methods for CA rule identification are inefficient, particularly in neighborhood selection.

Purpose of the Study:

• To develop a novel, efficient approach for identifying cellular automata (CA) rules and selecting appropriate neighborhoods from observed data.
• To provide a unified framework for both deterministic and probabilistic CA identification.

Main Methods:

• Developed a parameter-linear identification algorithm for CA rule extraction.
• Employed a minimum variance criterion for parameter estimation.
• Utilized an incremental procedure for initial neighborhood selection, followed by redundant cell removal and Bayesian Information Criterion for neighborhood size determination.

Main Results:

• The proposed algorithm effectively identifies CA rules and neighborhoods.
• Experimental results demonstrate superior performance compared to leading CA identification algorithms.
• The method is efficient and handles both deterministic and probabilistic CA models.

Conclusions:

• The novel approach offers a significant advancement in CA rule extraction.
• The algorithm's efficiency and accuracy make it a valuable tool for analyzing complex systems.
• This unified framework simplifies and improves the process of CA model identification.