90c34175923a36ab7a5de4b981c1972f-AuthorFeedback.pdf

Rebuttal for "Invertible Gaussian Reparameterization" (ID 10306): We thank the reviewers for their time and1 valuablefeedback. Weintroducedthesoftmax++ preciselytousethisfact.7 We note that the GS does not transformα into an approximate one-hot sample as claimed by R1. Reporting results on test set (R3): Thank you, this is an important point to discuss. Using statically binarized MNIST (R3): We are actually using this version of MNIST. On VIMCO and VQ-VAE (R4): Thank you for mentioning these papers, we will reference them.