Device operation neural network test method and device

Abstract

The invention discloses a device operation neural network test method and a device. In the embodiment of the invention, the method for testing the operation of the neural network by the to-be-tested device can comprise the steps: generating a reference model of the neural network through the description information of the neural network; calling a simulator to process a predetermined image according to the reference model of the neural network to obtain an expected result of the neural network; operating the neural network through a to-be-tested device to process the predetermined image so asto obtain a real result of the neural network; and determining whether the to-be-tested device can normally operate the neural network or not according to the real result of the neural network and theexpected result of the neural network. According to the method and the device, the test for the device operation neural network condition can be automatically completed without manually constructingtest data and preparing an expected calculation result, so that the labor load can be remarkably reduced, the test efficiency is improved, and the test accuracy is higher.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a testing method, device, equipment and storage medium for a device running a neural network. Background technique [0002] Neural network is an important application basis in the current artificial intelligence field, and it is more and more widely used in face recognition, classification, feature extraction, machine recognition, driverless driving and other fields. In order to speed up calculations, more and more artificial intelligence chips are implemented based on neural network algorithms. [0003] In practical applications, after the neural network is loaded into the hardware (for example, a dedicated chip), it is necessary to test the performance of the neural network operation performed by the hardware loaded with the neural network (for example, whether the result is correct, whether the accuracy meets expectations, etc.). In the process of testing the ...

AI Technical Summary

Problems solved by technology

In the process of testing the condition of the neural network running on the hardware, it is necessary to manually construct test data such as image data, parameter data, and instructions related to the neural network, and manually prepare the expected calculation results. The data volume is huge and the work complexity High, very low test efficiency

Images