Reviews: On the Recursive Teaching Dimension of VC Classes

The paper is very insightful - the authors quite nicely explain the approach they took for proving their results. The questions addressed, while interesting only for a fairly small subcommunity of the machine learning community, are really important in that subcommunity, and the authors have achieved a substantial breakthrough on an open problem posed in COLT 2015. I quite liked the idea to formulate the problem of finding a concept class with RTD 3/2 VCD as a SAT problem. In my eyes, the results should definitely be published, and they are important enough to deserve publication in a leading venue like NIPS. The paper is generally well written and easy to read, but there are a few minor (easy to fix issues (mostly just typos etc).