de Buitléir, Amy. “Evolving Pattern-Seeking Artificial Life with Créatúr.” Masters, Athlone Institute of Technology (2011).
URL1 URL2
This thesis describes a research project to evolve an Artificial Life (ALife) population with sufficient intelligence to discover patterns in data, and to make survival decisions based on those patterns. As part of this research, Créatúr, a reusable software framework for automating experiments with artificial lifeforms, was built. An ALife species called wains was implemented using diploid reproduction, Hebbian learning and Kohonen Self-organising Maps, in combination with novel techniques such as using pattern-rich data as the environment and framing data analysis as an ALife survival problem. The data set used was the MNIST database of handwritten numerals. The first generation of wains mastered the numeral recognition task well enough to thrive. Evolution further adapted the wains to their
environment by making them a little more pessimistic (lowering the rate at which they learn from positive experiences, and raising the the rate at which they learn from negative experiences), and also by making their brains more efficient (reducing the number of patterns remembered, and discarding the least useful patterns less frequently).