[论文笔记] PiPAD: Pipelined and Parallel Dynamic GNN Training on GPUs
PiPAD:PipelinedandParallelDynamicGNNTrainingonGPUsPiPAD:基于GPU的流水线并行动态GNN训练[Paper]PPoPP’23摘要提出了PiPAD,一个流水线并行的DGNN训练框架,用于GPU上的端到端性能优化.从算法和运行时层面重构了整体训练范式.1介绍动态图表示分类:连续时间动态图(ContinuousTimeDynamicGraphs,CT