Export Reviews, Discussions, Author Feedback and Meta-Reviews

This paper applies to the sticky HDP-HMM some recently developed machinery for split/merge style adaptation of the truncation level for variational inference in Bayesian nonparametric models and evaluates the resulting algorithm on several tasks, emphasizing identifying the true number of states present in a dataset. The algorithm is an improvement over the algorithm presented in [7] particularly because of these extra steps, and possibly due to the improved treatment of the variational factor on the top-level HDP weights. The paper is written very clearly. However, given previous work on these subjects the novel contributions here are incremental. Furthermore, the experiments are unenlightening and the synthetic experiment is very confusing.