Multi-instance dictionary learning via multivariate performance measure optimization

作者：

Highlights：

• Different multivariate performance require different optimal multi-instance dictionaries.

• We propose the problem of learning performance-optimal multi-instance dictionary.

• We learn dictionary and classifier jointly to optimize a target performance measure.

• It outperforms both independent dictionary learning and performance optimization methods.

摘要

•Different multivariate performance require different optimal multi-instance dictionaries.•We propose the problem of learning performance-optimal multi-instance dictionary.•We learn dictionary and classifier jointly to optimize a target performance measure.•It outperforms both independent dictionary learning and performance optimization methods.

论文关键词：Multi-instance learning,Dictionary,Multivariate performance measures,Cutting-plane algorithm

论文评审过程：Received 31 May 2016, Revised 3 December 2016, Accepted 21 December 2016, Available online 29 December 2016, Version of Record 12 March 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2016.12.023