Kocmánek, T. “Hyperneat and novelty search for image recognition.” Diss. Master’s thesis, Czech Technical University in Prague (2015).
URL1 URL2
This thesis is an initial attempt to investigate the use of Novelty Search in the NeuroEvolutionary algorithm called HyperNEAT on the subject of image recognition. Up until now this combination was used mainly for maze navigation or robot movement, but never in image recognition. The HyperNEAT uses an indirect encoding in evolution of Artificial Neural Networks. We present three novel approaches in feature detection, where the first of them shows promising results. It was concluded that the Novelty Search can play a significant role in the evolution of feature vectors in the image recognition domain