Integrated circuit chip devices and related products

Abstract

The invention provides an integrated circuit chip device and related products. The integrated circuit chip device includes an external interface and a processing circuit. The external interface is used to receive training instructions; the processing circuit is used to determine the first layer input data, the first Layer weight group data and the operation instructions included in the first layer, quantize the first layer input data and the first layer weight group data to obtain the first layer quantized input data and the first layer quantized weight group data, from the preset Query the first-layer quantized input data and the first-layer output data corresponding to the first-layer quantized weight group data in the output result table of the first layer, and input the first-layer output data as the second-layer input data to the n‑1 layer to execute The output data of the nth layer is obtained by operation; the gradient of the output data of the nth layer is determined according to the output data of the nth layer, and the reverse operation of the nth layer is obtained according to the training instruction. By quantifying the input data and weights, the calculation results can be directly queried, reducing the amount of parameters and calculations during training.

Description

technical field [0001] The invention relates to the field of neural networks, in particular to an integrated circuit chip device and related products. Background technique [0002] The existing neural network training methods generally use the backpropagation algorithm, and the learning process consists of a forward propagation process and a back propagation process. In the process of forward propagation, the input data passes through the hidden layer through the input layer, is processed layer by layer and transmitted to the output layer. If the expected output data cannot be obtained at the output layer, then switch to backpropagation, calculate the weight gradient of each layer in the backpropagation layer by layer, and finally use the calculated weight gradient to update the weight value, which is In one iteration of neural network training, this process needs to be repeated many times throughout the training process until the output data reaches the expected value. Thi...

AI Technical Summary

Problems solved by technology

If the expected output data cannot be obtained at the output layer, then switch to backpropagation, calculate the weight gradient of each layer in the backpropagation layer by layer, and finally use the calculated weight gradient to update the weight value, which is One iteration of neural network training, this process needs to be repeated many times throughout the training process until the output data reaches the expected value. This training method has problems of excessive parameters and calculations during the training process, and low training efficiency.

Images