Add Graph Feedforward Network as a Graph Unpooling Layer #10587
Description
🚀 The feature, motivation and pitch
Dear PyG team,
As a frequent user, I have noticed for some time that there is only one single unpooling layer available in PyG, namely the knn_interpolate layer. This layer is not trainable and lacks the expressive power required for many applications, which is why I believe it would be very interesting to incorporate other options for graph unpooling layers.
I developed a novel method extending linear layers to graphs in this paper: "GFN: A graph feedforward network for resolution-invariant reduced operator learning in multifidelity applications". The method is quite general, and notably can be used as a graph unpooling layer analagously to the knn_interpolate layer with the key difference that it is trainable and expressive.
I have already implemented a first attempt at how this unpooling could look in this PR #10588 and am looking forward to your feedback.
Alternatives
No response
Additional context
No response