An approach to supervised distance metric learning based on difference of convex functions programming

作者：

Highlights：

• We develop a metric learning method (DML-dc) for nearest-neighbor classification.

• DML-dc is based on difference of convex functions (DC) programming.

• DML-dc uses the ramp loss function to avoid the influence of outliers.

• Extensive experiments on several benchmark data sets show the effectiveness of DML-dc.

摘要

•We develop a metric learning method (DML-dc) for nearest-neighbor classification.•DML-dc is based on difference of convex functions (DC) programming.•DML-dc uses the ramp loss function to avoid the influence of outliers.•Extensive experiments on several benchmark data sets show the effectiveness of DML-dc.

论文关键词：Distance metric learning,Nearest neighbor,Linear transformation,DC programming

论文评审过程：Received 8 January 2018, Revised 26 March 2018, Accepted 24 April 2018, Available online 25 April 2018, Version of Record 16 May 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.04.024