ca46c1b9512a7a8315fa3c5a946e8265-Reviews.html

Alternatively, if the algorithm performs well against any of the existing AC ones then it would be good to see its performance in various settings and detailed explanation of its properties. The latter would be useful in its own right whether or not it is related to summarizing posterior distribution's characteristics. A few specific comments: Line 19: It is correct to say that the MAP might not be good in situations where the posterior is diffuse, I guess you mean uniform-like? It might also be multimodal, skewed (mode and mean differ), high variance,for example, so the MAP is not necessarily a good choice although it depends on the (underlying) loss function. Line 52: Since a Dirichlet Process is a discrete random probability measure, when sampling from it it induces a random partition (the ties will belong to the same cluster).