aaf2979785deb27864047e0ea40ef1b7-AuthorFeedback.pdf

Practicalityandclaims: Intermofpractical1 use of the EM training we agree and explicitly acknowledged that it is computationally demanding. Those insights are gained despite the computational limits of the practical EM learning as they rely on8 the analytical derivations only. We will add further analysis and discussions on this. EM is one common strategy to do so based on the posteriorp(z|x); once z14 is inferred, one can then do the maximization of the (now estimated) log-likelihood.Adding cost analysis of the15 algorithm: Wewill addintheappendix exact computation times andfurther details foreach oftheexperiment and16 different architectures for the EM learning as well as each step involved (partition finding, region triangulation, per17 region integration). Reviewer #2: We thank the reviewer for their appreciation of the paper.