Evolving cellular automata for diversity generation and pattern recognition: deterministic versus random strategy

TL;DR

This paper compares deterministic cellular automaton strategies to random transformations in evolving agents for pattern recognition, showing that automaton rules improve diversity and efficiency in matching complex patterns.

Contribution

It introduces a genetic algorithm with inheritable cellular automaton rules for agent rearrangements, demonstrating enhanced pattern recognition over random strategies.

Findings

Automaton-based agents outperform random transformations in pattern matching.

The population reaches a stable coexistence of different automaton rules.

Deterministic rules increase diversity and efficiency in evolving systems.

Abstract

Microbiological systems evolve to fulfill their tasks with maximal efficiency. The immune system is a remarkable example, where self-non self distinction is accomplished by means of molecular interaction between self proteins and antigens, triggering affinity-dependent systemic actions. Specificity of this binding and the infinitude of potential antigenic patterns call for novel mechanisms to generate antibody diversity. Inspired by this problem, we develop a genetic algorithm where agents evolve their strings in the presence of random antigenic strings and reproduce with affinity-dependent rates. We ask what is the best strategy to generate diversity if agents can rearrange their strings a finite number of times. We find that endowing each agent with an inheritable cellular automaton rule for performing rearrangements makes the system more efficient in pattern-matching than if transformations are totally random. In the former implementation, the population evolves to a stationary state where agents with different automata rules coexist.