层连分级式并行推理架构

The Classification Parallel Inference Framework with Hiberarchy

在分析现有并行推理架构的基础上,设计了一种改进的层连分级式并行推理架构(CPIFH)。该并行推理架构分为多个级别,按层次互相连接。通过减少并行推理架构中主机的工作量,进一步提高了并行推理的效率。从最后的比较结果可以看出,层连分级式并行推理架构能更好的提高并行推理的速度,增加并行推理运行的效率。对于并行推理而言,这种分级式的架构具有一定的应用潜力。

Based on the structure of the parallel inference system, an improved classification parallel inference framework with hiberarchy （CPIFH） is designed. This structure divided into some rank, joined each other according to the level. It has improved the efficiency of parallel inference by reducing the workload of the host computer in the inference framework. The test data shows that the classification parallel inference framework with hiberarchy can improve the speed of parallel inference, increase the efficiency of parallel inference. This classification structure has certain application potentiality for parallel inference.