The reviewers had low confidence in their assessments. Overall, I think that the idea as presented is sound and novel. However, the clarity of presentation could be significantly improved: reviewers found key parts of the exposition to be lacking in motivation; Eq 3 was not clear to one reviewer; and another reviewer felt that the details of the learning procedure should have been included in more detail in the paper rather than in the supplementals. I also think the extension to models for multiple types needs to be made more explicit.