Neural-network computing system and methods

Abstract

The disclosure discloses a neural-network computing system. The system includes: an I / O interface, which is used for I / O of data; a memory, which is used for temporarily storing a multi-layer artificial-neural-network model and neuron data; an artificial-neural-network chip, which is used for executing multi-layer artificial-neural-network operation and a back-propagation training algorithm thereof, wherein data and a program from a central processing unit (CPU) are accepted, and the above-mentioned multi-layer artificial-neural-network operation and the back-propagation training algorithm thereof are executed; the central processing unit CPU, which is used for data transportation and starting / stopping control of the artificial-neural-network chip, is used as an interface of the artificial-neural-network chip and external control, and receives results after execution of the artificial-neural-network chip. The disclosure also discloses a method of applying the above-mentioned system forartificial-neural-network compression encoding. According to the system, a model size of an artificial neural network can be effectively reduced, data processing speed of the artificial neural network can be increased, power consumption can be effectively reduced, and a resource utilization rate can be increased.

Description

technical field [0001] The present disclosure relates to the technical field of artificial neural network processing, and more particularly relates to a neural network computing system and method. Background technique [0002] Multi-layer artificial neural network is widely used in the fields of pattern recognition, image processing, function approximation and optimization calculation. Especially in recent years, due to the continuous deepening of research on backpropagation training algorithms and pre-training algorithms, multi-layer artificial neural networks have become more and more popular in academia and industry due to their high recognition accuracy and good parallelism. Widespread concern. [0003] With the sharp increase in the amount of calculation and memory access of artificial neural networks, general-purpose processors are usually used in the prior art to process multi-layer artificial neural network operations, training algorithms and compression coding, by ...

AI Technical Summary

Problems solved by technology

Since the GPU is a device specially used to perform graphics and image calculations and scientific calculations, without special support for multi-layer artificial neural network operations, it still requires a lot of front-end decoding work to perform multi-layer artificial neural network operations, which brings a lot of problems. additional cost

In addition, the GPU has only a small on-chip cache, and the model data (weights) of the multi-layer artificial neural network need to be repeatedly moved from off-chip. The off-chip bandwidth has become the main performance bottleneck, and it has brought huge power consumption overhead.

Images