HVGH: Segmenting High Dimensional Time Series Data Using VAE and HDP-GP-HSMM

Abstract

Humans recognize perceived continuous high dimensional information by dividing it into significant segments such as words and unit motions. We believe that such unsupervised segmentation is also an important ability for robots to learn topics such as language and motions. In this paper, we propose Hierarchical Dirichlet Processes-Variational Autoencoder-Gaussian Process-Hidden Semi-Markov Model (HVGH). The parameters of HVGH are estimated by mutual learning loop of VAE and HDP-GP-HSMM. Hence, HVGH can extract features from high dimensional time-series data and, simultaneously, divide it into segments in an unsupervised manner. In the experiment, we use the various motion-capture data and show that our proposed model can estimate the correct number of classes and accurate segments compared with baseline methods.