Semi-supervised Learning by Deep Learning Techniques and Information Theory
Semi-supervised, Deep Learning, Labeling, Clustering, Classification, Information Theory.
The expressive growth of modern data sets, combined with the difficulty of obtaining information about labels, has made semi-supervised learning one of the problems of practical importance in modern data analysis. In most cases, obtaining a set of data with enough examples to induce a classifier, can be costly, as it is necessary to carry out data labeling by a specialist. Unlabeled data is easier to obtain but more difficult to analyze and interpret. In of semi-supervised learning problem, there is a database formed by a small part labeled and a larger part not labeled, being possible two aspects: semi-supervised classification and semi-supervised clustering. With this, this work aims to apply models that use Deep Learning techniques in semi-supervised learning, where a neural network is trained, in this case, an autoencoder using unlabeled data. Then, an additional layer is embedded in the encoder. This new layer has its weights initialized by the K-means ++ algorithm and adjusted through the backpropagation algorithm using information theory learning. The labeled data is assigned to the clusters generated by the encoder, influencing the unlabeled, cluster by cluster, thus labeling the non-labeled data that was previously clustered. With the experiments carried out, it was noted that the satisfactory performanceof the proposed model when compared with other semi-supervised algorithms, both the classics such as self-training and co-training, as well as other more recent works, showing the proposed model feasibility for the learning semi-supervised problem.