|
The Hilbert-Schmidt Independence Criterion (HSIC) was originally design-ed to measure the statistical dependence of distribution-based Hilbert spaces embedding in statistical inference. In recent years, due to the validity and efficiency of this standard, it has been witnessed that this criterion can tackle a large number of learning problems owing to its effectiveness and high efficiency[1]. Unlike traditional binary classifications or multi-class single label problems, one goal of multi-label problems may be related to multiple labels. The rich relationship between labels makes the analysis of multilabel problems more complex. To solve the problem of how to make use of the relationship between features and labels, a multi-label feature selection algorithm based on HSIC is presented. This method uses Lasso (the least absolute shrinkage and selection operator) to solve a non-convex HSIC problem, and converts it to solve a lasso optimization problem, which can effectively calculate the global optimal solution. Finally, experiments show that our algorithm can improve the performance of multi-label classification.
|
