05月24日(Wed) 13:50〜15:30 C会場(ウインクあいち-9F 902会議室)
演題番号 | 2C2-3 |
---|---|
題目 | Laplacian-Constrained Tri-factorization for Feature Association Learning |
著者 | Zhai HongJie(IST, Hokkaido University) Haraguchi Makoto(IST, Hokkaido University) |
時間 | 05月24日(Wed) 14:30〜14:50 |
概要 | In this paper, we study a problem of feature association learning. That is, given two object sets with their own feature sets, the learning task is to find associations between features, where part of associations is already presented. The proposed method is based on an idea: With the given associations, all the objects can be clustered into object clusters. By representing features with the object clusters, some new associations between remaining features can be determined. By repeating this process, we are able to get all the feature associations. We formalized this problem as a kind of Non-negative Matrix Tri-factorization (NMF). The method consists of two main parts: (1) it performs Non-negative Matrix Tri-factorization on two object sets. (2) during factorization, it uses a laplacian constraint to guarantee the associated features to have close vectors after they are projected into the embedded spaces. |
論文 | PDFファイル |