Research on BP algorithm based on conjugate gradient

This paper uses the conjugate gradient method to optimize the calculation and achieve rapid calculation on the network weights and thresholds, simulates the traditional gradient descent and conjugate gradient algorithm of BP neural network, and discusses the training speed, fault-tolerant generalization ability of the method. The goal is to variously verify the superiority of conjugate gradient algorithm. The simulation results highlight the substantial increase in training speed. In particular, for the generalization ability of damaged network after training, using linear regression method to simulate can also obtain satisfaction result, which supports the conjugate gradient BP algorithm from the new angle.