357a6fdf7642bf815a88822c447d9dc4-AuthorFeedback.pdf

Reviewer231 We added a discussion about extending the proposed method to abroader class of losses, which is an interesting32 direction. Our view is that current common practice for making decisions with VAEs, such as using the a single posterior36 approximation both for calculating predictive densities for and model learning, lacksformal justification. Our37 approach removes this unjustifiable restriction.