Review for NeurIPS paper: Better Set Representations For Relational Reasoning

Weaknesses: In terms of the proposed method, there is little novelty over DSPN (Zhang et al., 2019): SRN is largely a re-branding of DSPN. Indeed, the attached code implementation of SRN simply imports and applies the DSPN module. This is not necessarily a bad thing: it is perfectly fine to re-use prior work, but I think that giving an existing module a new name leads to unnecessary confusion. There is no experimental comparison to related works that propose iterative models along similar lines of reasoning (e.g. All baselines are the same form of model ablation (ablation of the SRN network).