论文部分内容阅读
本文提出一种用于有限环算术的数字式神经网络方法。它采用非对称的数字式神经网络结构，在整个矢量空间上具有唯一的平衡点，因而不存在计算误差。这种方法保持了神经网络的高度并行结构，能够实时完成有限环上的模运算。它还被应用于中国余数定理的实现。
In this paper, we propose a digital neural network method for finite-loop arithmetic. It uses asymmetric digital neural network structure, the entire vector space has a unique balance point, so there is no calculation error. This method preserves the highly parallel structure of neural networks and realizes modulo operation on finite rings in real time. It is also applied to the realization of the Chinese remainder theorem.