[Python] What's the use case for `torch.randn_like` when`requires_grad=True`? [closed]

In PyTorch, requires_grad can be set to True in torch.randn_like, what's the use case for this?

Say we have $\epsilon ~ N(0, I)$, to my understanding, since $\epsilon$ is not a parameter, there is no need to pipe gradient through it.

Thanks in advance.

Continue reading...