Please enable JavaScript.
Coggle requires JavaScript to display documents.
对大规模Graph训练GNN (挑战 (Graph data (Large, Sparse), train model (Mini-batch…
对大规模Graph训练GNN
挑战
时间和空间
Graph data
Large
Sparse
train model
Mini-batch SGD
neighborhood expansion
Full-batch GD
策略：大图化小图
Sampling
GraphSAGE
固定采样大小
FastGCN
重要性采样
Adaptive Sampling
自适应采样、层间采样、跳链
Constant time embedding approximation
GraphSAINT
DGL
NodeFlow
block
sample
VRGCN
减小采样数量
Partitioning
ClusterGCN
embedding utilization
GPNN
多硬件环境考虑分布式存储计算
评估
memory
决定模型的可移植性
time per epoch
训练效率
convergence speed