General representational automata using deep neural networks

摘要: Abstract Unlabeled data representation constitutes a major challenge in mining. Different unsupervised learning methods such as clustering and dimensionality reduction form the basis of representations. The impact attribute combinations their interactions on is less addressed by models. A model supported with machine concepts can reveal more information about nature underlying data. We herein present novel minimum instance selection (UMAIS) labeling algorithm that selects categorical class label, attribute-based powerset generation (APSG) for describing formation relevant sets using correlation powerset. Using these algorithms, we diagrammatic known Representational Automata depict importance among correlated non-correlated attributes an unlabeled dataset. performed experiments two large-scale datasets from energy financial domains compared our approach other standard classifiers. Our obtains significantly better classification accuracy 92.187% 87.32% datasets, respectively, to 74% 82% linear classifier, respectively.