Peer review Q8PJ9UV4

The authors investigate the accuracy of quaternion transformers and the accuracy of Lottery Ticket Hypothesis (LTH) in pruning (quaternion) transformers. Specifically the Vision Transformer (ViT) and nanoGPT architectures are investigated. The authors performed an extensive evaluation of their proposed approach on existing transformer architectures. Through experimental validation, the authors show that incorporating quaternions in a transformer architecture and performing pruning using LTH increases the computational efficiency while maintaining performance. Furthermore, the implementation is available open source as a PyTorch package. It would be better to formulate the 4 research questions as research contributions. Then it is also easier to emphasise the novelty in this paper. For example, quaternion transformers exist. Does this paper mean that this extensive evaluation is not available in literature? If this evaluation is done here for the first time, it could be more emphasised. Also in the literature study in this paper, no references are made to existing Quaternion Transformers networks. Idem for the Lottery Ticket Hypothesis there are gaps in the literature study, for example: https://arxiv.org/abs/2005.03454 Does this paper tackle an additional problem related specifically to the ViT or nanoGPT architecture? I am not in expert in the current state of the art regarding Transformers, but I think the paper overall contains an interesting contribution that is worth sharing. Although I would say the current shortcomings needs to be addressed before the paper can be accepted.