论文部分内容阅读
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。
Firstly, the essence of neural network mapping, neural network topological structure and neural computing process are analyzed. Based on the consideration of load balancing and communication overhead, the idea of time step is introduced and a universal and effective neural network mapping algorithm is proposed. Finally, The test results of neural networks with many topological structures are given to demonstrate the generality and validity of the proposed algorithm.