一种通用有效的神经网络映射算法

首先分析了神经网络映射的本质，神经网络拓扑结构和神经计算过程，在充分考虑负载均衡与通讯开销的基础上，引入了时间步的思想，提出了一种通用有效的神经网络映射算法，最后给出的对多种拓扑结构的神经网络的测试结果证明了该算法的通用性和有效性。

A General-purpose Efficient Neural Network Mapping Algorithm

Neural network mapping algorithm is an important research content in the field of neural netork virtual implementation.The efficiency of the algorithm affects the virtual implementation of neural network.In this paper,the nature of neural network mapping,the neural nctwork topology and the neurocomputing process have been analysised at first.Based on thorovgh consideration of load-balance and communication-cost,the thesis gives a general-purpose efficient neural network mapping algorithm.introducing the idea of timestep.Finally,the experimenting results of neural networks with different topologies are presented,which proves the generality and efficiency of the algorithm.