Metric learning via perturbing hard-to-classify instances

作者：

Highlights：

• We give a Mahalanobis distance metric learning model by perturbing hard-to-classify instances.

• The proposed model makes the learned metric to avoid over-fitting the hard-to-classify instances.

• The model has closed-form solutions.

• Experiments show that the proposed model compares favorably to those state-of-the-art methods.

摘要

•We give a Mahalanobis distance metric learning model by perturbing hard-to-classify instances.•The proposed model makes the learned metric to avoid over-fitting the hard-to-classify instances.•The model has closed-form solutions.•Experiments show that the proposed model compares favorably to those state-of-the-art methods.

论文关键词：Metric learning,Hard-to-classify instances,Instance perturbation,Alternating minimization

论文评审过程：Received 31 December 2021, Revised 25 June 2022, Accepted 21 July 2022, Available online 22 July 2022, Version of Record 4 August 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108928