Neural Network Optimization Method

Abstract

The present application relates to a neural network optimization method. The neural network optimization method comprises the following steps: presetting modeling parameters, wherein the modeling parameters include network parameters and hardware parameters; Constructing a neural network energy consumption model based on the modeling parameters; The neural network energy consumption model and theneural network time model are subjected to double-objective optimization. The present application models the neural network in terms of time and energy consumption from the viewpoint of the hardware computing flow of the network, At the same time, the dominant modeling parameters of time and energy consumption are analyzed. The neural network model is improved by improving the modeling parameters,array segmentation method and buffer segmentation method to optimize the time and energy consumption of neural network.

Description

technical field [0001] The present application relates to the technical field of artificial neural networks, in particular to a neural network optimization method. Background technique [0002] With the rise of neural network technology, neural network hardware adapted to different application scenarios has emerged as the times require. The reasoning and prediction ability of the neural network is strong but the amount of calculation is large, so how to improve the calculation speed of the neural network and reduce the energy consumption of the neural network has become the key. [0003] In related technologies, both the neural network training process and the inference process have relatively urgent requirements for the acceleration of network computing. Network training is basically done on the cloud using GPUs. Different hardware parallelization methods and communication methods will greatly affect the speed of neural network training. Therefore, the calculation speed of...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome at least to a certain extent the research on neural network computing in related technologies focuses on either low energy consumption or accelerated computing speed, without considering that there may be some contradictions between the acceleration and energy perception of neural networks in complex application environments , when reducing energy consumption, it can be at the expense of speed or energy, and reducing the calculation time of the neural network may cause more energy consumption. This application provides a neural network optimization method

Images