NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:3697
Title:High-dimensional multivariate forecasting with low-rank Gaussian Copula Processes

This paper developed a method for probabilistic forecasting of high dimensional time series. The method parameterizes a structured covariance matrix with an RNN and incorporated a copula to account for dependence between series with heavy tailed marginals. Some of the reviewers had criticisms of the experiments. Specifically, comparing the method to other state of the art models instead of simpler time series models. One of these models a reviewer asked for was DeepGLO. Another reviewer took issue with the number of Monte Carlo samples used on the computation time. In all cases the authors addressed these issues with new experiments with results reported in the authors' response One reviewer also was concerned about the originality of the paper. However, the authors addressed this reasonably well in their response and looking at the paper the proposed approach solves an interesting problem and isn't a trivial combination of the constituent ideas. Overall, the I think that the authors have satisfactorily addressed the reviewers' main concerns and can be accepted.