Heterogeneous feature selection by group lasso with logistic regression

The selection of groups of discriminative features by the extension of group lasso with logistic regression for high-dimensional feature setting is formulates as the heterogeneous feature selection by Group Lasso with Logistic Regression (GLLR).

Abstract

The selection of groups of discriminative features is critical for image understanding since the irrelevant features could deteriorate the performance of image understanding. This paper formulates the selection of groups of discriminative features by the extension of group lasso with logistic regression for high-dimensional feature setting, we call it as the heterogeneous feature selection by Group Lasso with Logistic Regression (GLLR). GLLR encodes a sparse grouping prior to seek after a more interpretable model for feature selection and can identify most of discriminative groups of homogeneous features. The utilization of GLLR for image annotation shows the proposed GLLR achieves a better performance.