Hello every one,
I wanted to extend the DGL with general binary operations for message function as well as reduce function, I have few questions?
- First question is it possible for all the message and reduce functions? (Note: i am studying corresponding equivalent operations in binary system for current message functions such as: Op: add/sub/mul/div/dot and similarly for reduce functions sum/max/min/mean/prod.
- If it is possible, then do i will need to implement it for both GPUs and CPUs myself or just a generic C APIs will do the work for both GPUs and CPUs.
- I assume it will significantly improve the GNNs speed and memory optimizations for both training and inference for large-scale graphs.
The resources currently i am studying are:
Bi-GCN: Binary Graph Convolutional Network
Binary Graph Neural Networks
Matrix Algebra in a Binary Finite Field