BNS-GCN: Random Boundary Vertex Sampling to Accelerate Distributed GCN Training
MLSys’22 paper code BNS-GCN believes that the communication overhead of distributed GCN training is proportional to the number of boundary points. On this basis, in order to reduce communication and memory usage, before each epoch, they sample the boundary points (Boundary node sampling), through a large number of The experiments verified the performance of BNS-GCN, […]
BNS-GCN: Random Boundary Vertex Sampling to Accelerate Distributed GCN Training Read More »