☆ 4.7 Article

Eigen-guided deep metric learning

EXPERT SYSTEMS WITH APPLICATIONS (2022)

Adaptive Neighborhood Metric Learning

Kun Song et al.

Summary: In this paper, the authors propose a novel distance metric learning algorithm called adaptive neighborhood metric learning (ANML), which addresses the inseparable problem in metric learning. ANML adaptsively identifies inseparable similar and dissimilar samples in the training procedure and removes them while learning metric parameters. The proposed ANML can be used to learn both linear and deep embeddings, and it has interesting connections to other famous metric learning algorithms. The authors also introduce a log-exp mean function in ANML, which provides a new perspective for reviewing deep metric learning methods.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2022)

添加到收藏夹

Article Chemistry, Physical