NeurIPS 2020

Benchmarking Deep Inverse Models over time, and the Neural-Adjoint method

Meta Review

The reviewers all liked some aspects of the paper, agreed that the presentation of the technical part can be improved (e.g just spotted a missing ^2 in eq 7) and that the main contributions are 1) a heuristic (eq 5) to regularise the inverse solutions to be close to the input data and 2) the empirical evaluation suite. The presentation can (and should) of course be improved but it is not likely that the method will change if the paper is rejected now and resubmitted to another conference. Therefore acceptance is recommended together with a strong encouragement to rework the presentation for the final version.