NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:6036
Title:Demystifying Black-box Models with Symbolic Metamodels

The reviewers agreed that this paper presents a valuable contribution; they appreciated the quality of the writing, the overall motivation of interpretability of the models, and the proposed approach to use gradient descent to learn symbolic models. The primary shortcomings that the discussion focused around were some of the limited evaluation included in the paper, and the justification of the proposed models as being “interpretable”. The response was useful in addressing some of these concerns.