Tensor-based embedding for graph-based semi-supervised approaches

This paper presents a novel approach to multiclass classification tasks, utilizing tensor-based embeddings for graph-based semi-supervised learning. The proposed method utilizes a tensor decomposition algorithm to create embeddings that capture the essential features of the data. These are used by various graph-based semi-supervised approach to construct a graph capable to propagate the information from labeled to unlabeled nodes, classifying available data. The proposed method was tested on hyperspectral datasets. The results demonstrate the potential of such combinatory tensor-based semi-supervised approaches.