Is CEBRA invariant to permutations in the input channels? #182
-
|
Should I still get the same embedding even if I shuffle the order of the input neurons? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
Hi @icarosadero , you would not get the same embedding, but an embedding with very high consistency. So yes, the ordering does not matter. This applies if you train and embed from scratch. The input layer is not permutation invariant in the sense that you can train a model, input permuted channels, and expect to get the same embedding. Does that answer the Q? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it does. Thanks |
Beta Was this translation helpful? Give feedback.
Hi @icarosadero , you would not get the same embedding, but an embedding with very high consistency. So yes, the ordering does not matter.
This applies if you train and embed from scratch. The input layer is not permutation invariant in the sense that you can train a model, input permuted channels, and expect to get the same embedding.
Does that answer the Q?