Residual high-precision parallel computing with using of neural networks

The authors present the modeling high-precision parallel computing with high-bit data, based on the residue number systems with the conversion of numbers from the positional notation to the residual notation and back with the using of neural networks. Application of neural technology allows parallelize computations at the level of the algorithm, and the use of residue number systems makes it possible to improve the performance of high-precision computation through the transition to a low-bit data processing and application of parallel processing at the level of the elementary arithmetic operations. A neural network for the conversion of the positional notation to the residual notation and back is constructed. The dependence of the time of conversion from changing the bit has been got. The dependence of the time operations of addition, subtraction and multiplication for fixed-point numbers from changing the bit has been got.