One of the more interesting examples of a custom layer I've seen. Faster fully connected layers w/ Tensor Networks. Idea in a nutshell: replace the usual `Dense` layer w/ a matrix product, for faster inference & similar representational power.
-
-
Very cool! Thanks for the links to TensorLy, will check out.
-
We have implementations of tensor regression layers with TensorLy and various backends (here, PyTorch):https://github.com/JeanKossaifi/tensorly-notebooks/blob/master/05_pytorch_backend/tensor_regression_layer_pytorch.ipynb …
End of conversation
New conversation -
-
-
Yep! And that one wasn't the first either. This one uses tensor trains in FCL all the way back in 2015. https://arxiv.org/abs/1509.06569 The idea has been around for a while, but for some reason it hasn't really caught on.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.