Classificação não supervisionada para autômato celular elementar do Wolfram

Abstract

Cellular Automata are gaining ground in many research areas such as biology, physics and chemistry. But as more parameters are added, it becomes impossible to find features that may be interesting for a particular application. This work has objective by unsupervised machine learning methods like texture descriptors and convolutional autoencoder to determine which Elementary Cellular Automaton rules belong to the classes determined by Wolfram. To do unsupervised learning, histograms were generated with the Local Binary Pattern Variance (LBPV) and the Convolutional Autoencoder, then using the histograms in the K-Means model to analyze whether the model is able to separate the data into four clusters according to the classification proposed by Woflram. In the end, it was observed that both methods did not were able to perform such classification, and the LBPV showed the closest results.