Reviews: Learning a Warping Distance from Unlabeled Time Series Using Sequence Autoencoders

In this paper, authors proposed a metric called warping distance to measure the distance between raw sequence. BetaCV is optimized to learn the parameters in the metric and the robustness of this metric to initial guess of clustering is proven. Compared with using Euclidean distance between sequences' latent representation, the proposed method shows some potentials to get better clustering results. My main concerns include: 1. I think authors may underestimate the power of autoencoder.