8511df98c02ab60aea1b2356c013bc0f-AuthorFeedback.pdf

We thank the reviewers for their consideration of our paper and their insightful suggestions that1 will beincluded intherevision. The use of an attention23 mechanism in our approach significantly benefits from a learning rate schedule (to prevent early24 saturation or instability).25 Anexample with threshold 0.5 (where the effect is strongest) is in Figure 1. Using otherwise the same hyperparameters, this model achieves75.5 3.8%ARI as opposed to37 98.8 0.3%forSlotAttention.