NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:8844
Title:Input-Output Equivalence of Unitary and Contractive RNNs

Although the paper does not suggest any specific direction for designing better RNN architectures, it provides relevant and significant theoretical results on the expressivity of Unitary RNN. Thus it can be stated that the paper contributes to a better understanding of the capabilities of Unitary RNN and for this reason is worth to be accepted. The rebuttal contributed to clarify issues raised by reviewers.