Self‐labeled techniques, a semi‐supervised classification paradigm (SSC), are highly effective in alleviating the scarcity of labeled data used in classification tasks through an iterative process of self‐training. This problem was addressed by several approaches with different assumptions about the features of the input data, examples of these approaches being self‐training, co‐training, STRED, among others. This paper presents a framework for data self‐labeling based on deep autoencoder combined with a self‐labeled technique that takes into consideration cross‐entropy. The model uses the Encoder to reduce the dimensionality of the input that is submitted to a labeling layer. The weights of this layer are adjusted through the learning from a clustering performed in the Z space, which is the reduced dimensionality space. Results showed that the proposed method obtained competitive performance in relation to classic methods that are found in the literature.
The field of Deep Learning is in constant evolution, with new techniques and applications being developed by the day. Of those techniques, semi-supervised deep learning have promising results, especially in combination with the standard Convolutional Neural Network (CNN) architectures. CNNs attain state-of-the-art performance on various classification tasks assuming a sufficiently large number of labeled training examples. Unfortunately, labeling sufficiently large training data sets requires human involvement, which is expensive and time consuming. In semi-supervised learning there is not only a set of labeled samples (L), but also a set of unlabeled samples (U), which is generally greater than the first (U > L). This paper presents a semi-supervised model using a CNN supported by a Multilayer Perceprton (MLP) network, and a clustering process by k Nearest Labeled Neighbors. The results showed that the proposed model solves the semi-supervised learning problem over different scenarios.
The techniques of clustering and classification are frequently used to obtain patterns and classify new data. The combination of those techniques can be applied to problems where there is no label, using the clustering process to extract information that will assist the classification process. Usually, the clusters are analyzed by an expert to obtain that information, but this process can also be done by automatic labeling models. Those are models capable of identifying the most relevant characteristics and use them to create a label. This paper proposes a model of classification of clusters using labels and Fuzzy logic. The efficiency of the proposed model was evaluated by comparing the accuracy and standard deviation, as well as individually analyzed each rule through the metrics based on the contingency matrix. For different databases available in the UCI repository, the results show that the rules reached precision and specificity rates above 94 %, with accuracy and standard deviation similar to the algorithms. Resumo: As técnicas de agrupamento e classificação de dados são frequentemente utilizadas com a finalidade de extrair padrões e classificar novos elementos. A combinação de tais técnicas pode ser aplicada em bases que não se conhece o atributo classe, utilizando a interpretação dos grupos obtidos no processo de agrupamento para identificação de padrões que auxilie o processo de classificacão. Esta interpretação pode ser desempenhada por modelos de rotulação: modelos capazes de identificar características relevantes e utilizá-las na formação de rótulos. Este trabalho propõe um modelo de classificação de grupos utilizando os rótulos e lógica Fuzzy. O modelo proposto foi avaliado comparando a acurácia e desvio padrão, bem como analisado individualmente cada regra através da matriz de contingência. Para diferentes bases disponíveis no repositório UCI, os resultados mostram que as regras obtidas alcançaram taxas de precisão e especificidade acima de 94 %, com acurácia e desvio padrão similares aos de algoritmos comparados.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.