Deep semi-supervised learning using Multi-Layered Extreme Learning Machines

Graph-based and deep learning-based semi supervised learning algorithms have been successfully used for semantic extraction from large unstructured data, alongside alternative methods like dimensionality reduction and embedding. Semantic information in the form of edges linking similar instances can be utilized to learn from unlabeled data and low-dimensional embeddings can be used to visualize and classify them. In this paper, we solve the problem of extending nonlinear embedding algorithms to Multi-Layered Extreme Learning Machines by plugging the network to auxiliary layers to improve the semi-supervised learning performance by further building on the structure assumption of data. Our model is significantly different from the previous models in two ways.