Abstract
Gaussian Process Dynamical Models (GPDM) is a nonlinear dimensionality reduction technique for time series data that provides a probabilistic representation of the time series data in terms of Gaussian process priors. In this paper, we study a method based on GPDM to visualize states of time-series data. Conventional GPDM models are unsupervised, and therefore when labels of data are available, it is not possible to use this information. To overcome the problem, we propose a Supervised GPDM (S-GPDM) model which utilizes both data and their corresponding labels. In the experiments, we demonstrate that the S-GPDM models can locate related motion data closer together than conventional GPDM models.
