误差反向传播的过程中，链式求导会经过好几层神经元。最后第一层的误差经过

In the process of error back propagation, the chain derivative will pass through several layers of neurons. Finally, the error of the first layer after the weight distribution is the penultimate layer error, the penultimate layer of the error is assigned by the next layer of weights to become the penultimate layer error. . Finally propagated to the first layer, the wheel of error ran over all neurons. All weights and thresholds have been changed, and one training is completed. After many trainings, the weights and thresholds are optimized to a very good level, and then you can make predictions.

During the process of the reverse propagation of error, the chain-induced sequester passes through several layers of neurons. The error of the last layer is the error of the penultimate layer after the weight allocation, and the error of the penultimate layer through the allocation of the next layer of weight becomes the error of the penultimate layer.... and so on. Finally, it spreads to the first layer, and the wheel of error crushes all the neurons. Change all weights and thresholds, one training complete. After many trainingsessions, the weights and thresholds are optimized to a very good level, and then they can be predicted.

In the process of error back propagation, the chain derivation will go through several layers of neurons. Finally, the error of the first layer is the error of the penultimate layer after the weight distribution, and the error of the penultimate layer becomes the error of the penultimate layer after the weight distribution of the next layer..... And so on. At last, it spread to the first layer, and the wheel of error ran over all the neurons. All the weights and thresholds are changed, and the training is completed at one time. After a lot of training, the weights and thresholds are optimized to a very good level, and then can be predicted.<br>