Online learning the consensus of multiple correspondences between sets.

Abstract

When several subjects solve the assignment problem of two sets, differences on the correspondences computed by these subjects may occur. These differences appear due to several factors. For example, one of the subjects may give more importance to some of the elements’ attributes than another subject. Another factor could be that the assignment problem is computed through a suboptimal algorithm and different non-optimal correspondences can appear. In this paper, we present a consensus methodology to deduct the consensus of several correspondences between two sets. Moreover, we also present an online learning algorithm to deduct some weights that gauge the impact of each initial correspondence on the consensus. In the experimental section, we show the evolution of these parameters together with the evolution of the consensus accuracy. We observe that there is a clear dependence of the learned weights with respect to the quality of the initial correspondences. Moreover, we also observe that in the first iterations of the learning algorithm, the consensus accuracy drastically increases and then stabilises.

Citation

MORENO-GARCÍA, C.F. and SERRATOSA, F. 2015. Online learning the consensus of multiple correspondences between sets. Knowledge-based systems [online], 90, pages 49-57. Available from: https://doi.org/10.1016/j.knosys.2015.09.034

• Journal Article Type: Article
• Acceptance Date: Sep 28, 2015
• Online Publication Date: Oct 8, 2015
• Publication Date: Dec 31, 2015
• Deposit Date: Feb 5, 2020
• Publicly Available Date: Feb 5, 2020
• Journal: Knowledge-based systems
• Print ISSN: 0950-7051
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 90
• Pages: 49-57
• DOI: https://doi.org/10.1016/j.knosys.2015.09.034
• Keywords: Consensus; Learning weights; Correspondence between sets; Linear solver; Hamming distance
• Public URL: https://rgu-repository.worktribe.com/output/816408