Layered Cellular Automata

Abstract

Layered Cellular Automata (LCA) extends the concept of traditional cellular automata (CA) to model complex systems and phenomena. In LCA, each cell's next state is determined by the interaction of two layers of computation, allowing for more dynamic and realistic simulations. This thesis explores the design, dynamics, and applications of LCA, with a focus on its potential in pattern recognition and classification. The research begins by introducing the limitations of traditional CA in capturing the complexity of real-world systems. It then presents the concept of LCA, where layer 0 corresponds to a predefined model, and layer 1 represents the proposed model with additional influence. The interlayer rules, denoted as f and g, enable interactions not only from adjacent neighboring cells but also from some far-away neighboring cells, capturing long-range dependencies. The thesis explores various LCA models, including those based on averaging, maximization, minimization, and modified ECA neighborhoods. Additionally, the implementation of LCA on the 2-D cellular automaton Game of Life is discussed, showcasing intriguing patterns and behaviors. Through extensive experiments, the dynamics of different LCA models are analyzed, revealing their sensitivity to rule changes and block size variations. Convergent LCAs, which converge to fixed points from any initial configuration, are identified and used to design a two-class pattern classifier. Comparative evaluations demonstrate the competitive performance of the LCA-based classifier against existing algorithms. Theoretical analysis of LCA properties contributes to a deeper understanding of its computational capabilities and behaviors. The research also suggests potential future directions, such as exploring advanced LCA models, higher-dimensional simulations, and hybrid approaches integrating LCA with other computational models.