Number of RNS Parts
The paper recommends starting with m = 2 and increasing it until accuracy plateaus. An RNS epoch processes about |E|/m induced edges across all batches. The number of RNS batches per epoch strongly affects accuracy even when other training…
1 sources - 4 claims
The paper recommends starting with m = 2 and increasing it until accuracy plateaus. An RNS epoch processes about |E|/m induced edges across all batches. The number of RNS batches per epoch strongly affects accuracy even when other training hyperparameters are independently tuned. Increasing m can damage topology by increasing diameter or removing too much connectivity.