Gaussian Process Dynamical Autoencoder Model

Dimension reduction realize extraction of substantial low dimensional latent structure in high-dimensional data. Due to recent developments in information and measurement technology, it becomes more important to develop dimension reduction algorithms for high dimensional time series data. Gaussian process dynamic model (GPDM) is a method that can obtain low dimensional latent variable representation by using Gaussian process state space model. However, it is difficult to obtain an appropriate latent variable representation of new data point in the GPDM. In this study, we propose a Gaussian Process dynamic autoencoder model (GPDAEM), which consists of Gaussian process state space model and Gaussian process encoder model, in order to estimate appropriate latent variables corresponding to additional new time series data. Experimental results on low dimensional latent variable representation of time series data show that the proposed GPDAEM has better performance than the existing Gaussian process based latent variable models.