Summary
the 2014 International Symposium on Nonlinear Theory and its Applications
2014
Session Number:D2L-B
Session:
Number:D2L-B4
Latent Variable Models Combined with Clustering
Daming Li, Quan Wang, Fei Wang, Xuan Wang,
pp.763-766
Publication Date:2014/9/14
Online ISSN:2188-5079
Summary:
Given a known dataset to learn a latent variable model, previous methods fail to focus on how to get labeled samples, some just choosing them randomly. As a result, it is likely that those methods gain models with a very limited representation capability. In this article, we propose a novel method in which we select representative samples to be labeled ones for latent variable models. To this end, the G-means clustering algorithm is adopted to automatically cluster latent variables and obtain those corresponding representative samples. We learn the Gaussian Process Latent Variable Model (GPLVM) and the Constrained Latent Variable Model (CLVM) respectively combined with the G-means, and compare them to those without clustering, in the context of non-rigid 3D reconstruction from monocular images. Our experimental results show that our methods present a more powerful representation capability.