SLFE: A New Semi-supervised Local Feature Extraction Algorithm

In big data era, how to extract feature of big data stream is one of the challenges existing in machine learning recently. Feature extraction method has attracted much attention for its effective application to data classification. Traditional classification algorithms may take less advantage of labeled samples information. Based on the property that data with the same class label locate in one manifold and data labeled different classes locate in corresponding manifolds, we combine the labeled manifold information and local distance. Online learning and Out-of-Sample problems are recent hot topics. To solve these problems, a novel algorithm named Semi-supervised Local Feature Extraction (SLFE) is proposed in this paper. First we extract feature in semi-supervised way, then we define a novel manifold similarity to construct local tangent space matrix incrementally. In order to obtain multi-manifold objective function, we utilize an explicit mapping function to overcome the Out-of-Sample problem. Experiments have been carried out on several UCI datasets and real world image datasets(such as ORL, MNIST)with comparisons to some related semi-supervised feature extraction methods. The experiment results demonstrate that the proposed algorithm can obtain better performance.