Zetao Chen - Supervised Feature Selection based on Generalized Matrix learning Vector Quantization

The curse of dimensionality” refers to the problem when analyzing high-dimensional data. Feature selection is the task of choosing a smaller feature subset which can capture the data property and predict the label information. The search strategy in feature space is of great importance. A lot of feature weighting algorithms have been proposed to rank the features with their contributions for classification and provide a search direction in the feature space. Generalized Matrix LVQ is a prototype-based supervised classification method whose distance matrix can account for pairwise correlations of features. Its application on feature selection has not yet been discovered. For feature selection, new energy is introduced during its training to provide more discriminative information of feature relevance and provide a search strategy in the feature subset selection. In my talk, I will introduce this GMLVQ-based feature selection algorithm and explore its application on different data set. Its comparison with other state-of-the-art feature selection algorithms will also be presented.