Volume no :3, Issue no: 1, September (2015)

International Journal of Applied Mathematics and Machine Learning

MULTI-INSTANCE LEARNING ON INNER STRUCTURE OF BAGS VIA WEIGHTED MATRIX KERNEL

Author's: Haitao Xu, Liya Fan and Hongxia Zheng
Pages: [41] - [49]
Received Date: August 10, 2015
Submitted by: Jianqiang Gao.
DOI: http://dx.doi.org/10.18642/ijamml_7100121537

Abstract

Most previous approaches on multiple instance learning (MIL) had focus on the structures between bags, such as positive instance clustering and bag similarity. In this paper, we proposed a novel method which called weighted matrix kernel support vector machine (WMKSVM) to solve the MIL problems. For WMKSVM, we consider the inner bag structure and assign each instance a weight based on a distance metric between each pair of instances in the same bag. Experiments on six data sets have shown that WMKSVM performs better than other key existing MIL algorithms.

Keywords

machine learning, multiple instance learning, support vector machine, instance weighting, weighted matrix kernel.