Condition-guided adversarial generation test method and system for deep neural network

Abstract

The invention provides a condition-guided adversarial generation test method and system for a deep neural network, and the method comprises the steps: collecting a needed data set and a correspondinglabel, carrying out the grouping of the needed data set and the corresponding label, obtaining a needed batch processing pool data set, selecting a seed set through employing a heuristic algorithm, and carrying out the condition-guided adversarial generation training set. The target of the test generation process is to maximize the network coverage rate of the test suite, obtain a generated picture set and input the picture set into a corresponding network as a training set for testing, and if the generated picture set enables the coverage rate of the original network to be improved, the pictures are added into a batch processing pool as a batch. According to the method, the condition-guided adversarial generation network is used, the labels of the pictures are used as conditions to generate the pictures, and the generation scale can be reduced. The test case is confronted and generated under the guidance of the coverage rate, the neuron coverage rate of a given network or system can be maximized, and the precision of the deep neural network to be tested is improved.

Description

technical field [0001] The invention relates to a test case generation method, in particular to a conditional and controllable test case generation method, which belongs to the field of artificial intelligence testing. Background technique [0002] In recent years, Deep Neural Networks (DNNs) have been widely used in various application fields such as image recognition, natural language processing, malware detection, self-driving cars, etc. due to their high accuracy and efficiency. However, as more and more safety-critical applications start to adopt DNNs, deploying DNNs without thorough testing can create serious problems, such as possible accidents during autonomous driving. Therefore, efficient and reliable testing of deep neural network-based systems is imminent. [0003] Fuzzing is a good way to test DNNs. Traditional fuzzing mutates through bit / byte flips, block replacements, and intersections between input files, etc., but these methods result in too many inputs to...

AI Technical Summary

Problems solved by technology

The mutation algorithm of the above method to generate test cases uses the low-level visual information of the image, ignoring the high-level semantics of the image, and the middle and low-level visual information often represents the detailed information of the image.

And a large number of invalid test cases are generated, and the test cases we need cannot be generated constrainedly, wasting resources and time

Images