Fantope Regularization in Metric Learning

Marc Law 1 Nicolas Thome 1 Matthieu Cord 1
1 MLIA - Machine Learning and Information Access
LIP6 - Laboratoire d'Informatique de Paris 6
Abstract : This paper introduces a regularization method to ex-plicitly control the rank of a learned symmetric positive semidefinite distance matrix in distance metric learning. To this end, we propose to incorporate in the objective function a linear regularization term that minimizes the k smallest eigenvalues of the distance matrix. It is equivalent to min-imizing the trace of the product of the distance matrix with a matrix in the convex hull of rank-k projection matrices, called a Fantope. Based on this new regularization method, we derive an optimization scheme to efficiently learn the distance matrix. We demonstrate the effectiveness of the method on synthetic and challenging real datasets of face verification and image classification with relative attributes, on which our method outperforms state-of-the-art metric learning algorithms.
Document type :
Conference papers
Liste complète des métadonnées

Cited literature [28 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01094074
Contributor : Marc Law <>
Submitted on : Thursday, December 11, 2014 - 3:49:59 PM
Last modification on : Thursday, March 21, 2019 - 1:14:16 PM
Document(s) archivé(s) le : Saturday, April 15, 2017 - 7:25:12 AM

File

fantope_regularization.pdf
Files produced by the author(s)

Identifiers

Citation

Marc Law, Nicolas Thome, Matthieu Cord. Fantope Regularization in Metric Learning. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2014, Columbus, Ohio, United States. pp.1051 - 1058, ⟨10.1109/CVPR.2014.138⟩. ⟨hal-01094074⟩

Share

Metrics

Record views

149

Files downloads

118