论文部分内容阅读
数据处理的步调一致是神经网络系统的重要问题.根据神经网络的特点,提出了新型的数据一致性算法.该算法的运行范围限制在请求神经元到根神经元之间,采用回溯方法寻找仲裁神经元,并采用逻辑时戳保证消息的时序性;算法采用残存树探测方法应对网络改造等问题.性能分析与仿真证明,该算法具有较低的消息复杂度、较短的响应延迟以及较好的容错性能.
The unanimity of data processing is an important issue of the neural network system.According to the characteristics of the neural network, a new data consistency algorithm is proposed.The operating range of the algorithm is limited to the request neuron to the root neuron, using the backtracking method to find the arbitration Neurons and use logical timestamp to ensure the timing of the message.The algorithm uses the residual tree detection method to deal with the network transformation and other issues.The performance analysis and simulation prove that the algorithm has lower message complexity, shorter response delay and better Fault tolerance.