The paper introduces a new approach to model monotonic functions by integrating the output of a non-negative neural network. All reviewers thought the idea is interesting, original, and worth pursuing. In particular, it does seem to offer more flexibility compared to existing approaches. Please revise some of the claims on the expressiveness (compared to existing work) as suggested by the reviewers. If possible, more thorough experiments could better demonstrate the effectiveness of the approach (reviewer 1 provided some great suggestions).