Variational Gaussian Process Dynamical Systems

TL;DR

This paper introduces the variational Gaussian process dynamical system, a probabilistic model for nonlinear dimensionality reduction and dynamical learning in high-dimensional time series data, with automatic latent space dimension determination.

Contribution

It extends Gaussian process latent variable models with variational methods to jointly learn nonlinear embeddings and dynamics, automatically inferring latent space dimensionality.

Findings

Successfully applied to human motion capture data

Effective on high-resolution video sequences

Automatically determines latent space dimension

Abstract

High dimensional time series are endemic in applications of machine learning such as robotics (sensor data), computational biology (gene expression data), vision (video sequences) and graphics (motion capture data). Practical nonlinear probabilistic approaches to this data are required. In this paper we introduce the variational Gaussian process dynamical system. Our work builds on recent variational approximations for Gaussian process latent variable models to allow for nonlinear dimensionality reduction simultaneously with learning a dynamical prior in the latent space. The approach also allows for the appropriate dimensionality of the latent space to be automatically determined. We demonstrate the model on a human motion capture data set and a series of high resolution video sequences.