A Deep Learning Approach to Selecting Representative Time Steps for Time-Varying Multivariate Data

Abstract

We present a deep learning approach that selects representative time steps from a given time-varying multivariate data set. Our solution leverages an autoencoder that implicitly learns feature descriptors of each individual volume in a latent space. These feature descriptors are used to reconstruct respective volumes for error estimation during network training. We then perform dimensionality reduction of these feature descriptors and select representative time steps in the projected space. Unlike previous approaches, our solution can handle time-varying multivariate data sets where the multivariate features can be learned using a multichannel input to the autoencoder. We demonstrate the effectiveness of our approach using several time-varying multivariate data sets and compare our selection results with those generated using an information-theoretic approach.