一种优化神经网络映射算法——吸收算法

本文揭示了神经计算的本质──并行分布处理，并以此为基础提出了时间步的概念。分析了映射算法的两个重要概念──负载均衡和通讯开销，并提出了映射分配准则。在神经网络的映射分配中引入图论的有关思想，提出了一种优化的神经网络映射算法──吸收算法。最后给出了重要的试验结果，这些数据表明吸收算法是一种有效的映射算法。

An Optimal Neural Network Mapping Algorithm-Absorb Algorithm

In this paper the important feature of artificial neural networks-parallel distributed processing is discussed thoroughly,so the essential features of neural computing are revealed, and the important conception of timestep is introduced. At the same time, two critical concepts of mapping algorithm-balancing the working load of PEs and the communicatin overhead among PEs are explained more thoroughly, and the main allocation criterions are presented. Adopting the ideas of graph theory, an optimal neural network mapping algorithm-Absorb Algorithm is advanced. This mapping algorithm shows remarkable efficiency.Some interesting experimental results are presented.