sweichwald

Hi there, thanks for raising this! It would be great if you could provide a minimum working example for how the mean computation results in points not on the manifold....

Toying around with `examples/backend-everywhere.py -b pytorch` is revealing: Currently, we decorate the cost function to indicate the backend, which is then used to obtain gradient/hessian if autodiff capability is provided...

Hi all :sunglasses: and thanks @nkoep ! > These classes shouldn't have a notion of the particular execution context at all. (Of course you could argue that by implementing manifold...