T- Hop: Tensor representation of paths in graph convolutional networks
This work addresses a domain-specific challenge in graph neural networks by providing an incremental improvement for encoding path information, potentially benefiting researchers and practitioners in graph-based machine learning.
The paper tackles the problem of encoding path information in graphs by introducing a 3-d tensor representation and connecting it to powered adjacency matrices, resulting in a method that reduces computational demands through dimensionality reduction and integrates into existing graph convolutional networks like MixHop.
We describe a method for encoding path information in graphs into a 3-d tensor. We show a connection between the introduced path representation scheme and powered adjacency matrices. To alleviate the heavy computational demands of working with the 3-d tensor, we propose to apply dimensionality reduction on the depth axis of the tensor. We then describe our the reduced 3-d matrix can be parlayed into a plausible graph convolutional layer, by infusing it into an established graph convolutional network framework such as MixHop.