Unsupervised Scalable Representation Learning for Multivariate Time Series

Time series constitute a challenging data type for machine learning algorithms, due to their highly variable lengths and sparse labeling in practice.