Text Classification with Cellular Automata Networks

Abstract

The thesis focuses on several options of exploiting reservoir computing ideas in the context of text classification, using cellular automata networks as the reservoir. The key innovative aspect of the study is related to finding options for computing with low-density networks, which should be easier to manufacture. The overarching idea is to extend Wolfram’s elementary cellular automata rules and implement them within the framework of random cellular automata networks. The primary objectives of this study are as follows: (i) To determine the optimal network topologies, (ii) to identify the most effective transition rules within these networks, and (iii) to develop a robust methodological framework for finding the best networks. By exploring various network configurations, the aim is to uncover the structural characteristics that facilitate efficient information propagation and decision-making within the cellular automata framework. Observations from the study indicate that there exist cellular automata network configurations which can function as efficient reservoirs. Moreover, all considered cellular automata networks are categorised into different groups based on their performance, providing a solid foundation for further research.