NeurIPS 2020

Learning Deep Attribution Priors Based On Prior Knowledge

Meta Review

Three referees vote for acceptance, one leans towards rejection. The pros and cons of this paper have been discussed intensively. In my opinion, the rebuttal carefully addressed most of the relevant points of criticism raised in the initial reviews, and my final impression is that this paper does indeed contain relevant contributions, so I also recommend acceptance. However, please consider revising your paper to address reviewer R1’s remark on the prior assumption that every input data point should have the same attributions.