Deep neural network test sufficiency method based on variable intensity combination test

Abstract

The invention discloses a deep neural network test sufficiency method based on a variable intensity combination test. The method includes: utilizing a variable intensity combination test technology toperform relation extraction on neurons in the deep neural network according to the model weight, extracting neuron combinations with different intensities, and assessing the coverage condition of theneuron activation state in the neural network according to the neuron activation state in the neuron combinations; and evaluating the model test sufficiency according to the calculated coverage rate.According to the method, the neuron state space is effectively reduced, corresponding neuron combinations are extracted according to different action relationships, and coverage rate calculation is carried out. If the test case can reach a relatively high coverage rate, the sufficiency of the test case can be better proved, so that the scientificity and credibility of the test criterion can be improved.

Description

technical field [0001] The present invention proposes a test adequacy criterion of a deep neural network based on a variable intensity combination test, which is used for evaluating the test adequacy of a deep neural network. The invention relates to the technical field of deep learning testing. Background technique [0002] Deep learning is a method of data representation learning in machine learning, which extracts features and converts data through the cascade of multi-layer nonlinear processing units. Deep learning has been officially proposed since 2016, which has made a revolutionary breakthrough in artificial intelligence. In recent years, deep learning has developed rapidly and has been applied to many safety-critical fields, such as autonomous driving, intelligent medical care and so on. However, due to the lack of effective testing technology, security accidents occur frequently. [0003] The system with deep learning as the core has the characteristics of high-...

AI Technical Summary

Problems solved by technology

However, due to the lack of effective testing technology, security accidents occur frequently

[0003] The system with deep learning as the core has the characteristics of high-dimensional input, multiple hidden layers, and low-dimensional output. These make deep learning systems very different from traditional software systems, and traditional software testing techniques cannot be applied in deep learning.

The model based on deep neural network also has the characteristics of application diversity, complex use scenarios and large amount of data, which makes it face many challenges when testing this type of system, such as: few suitable and efficient test methods, insufficient test data and Can't wait for data quality

[0004] At present, it is difficult for deep learning systems to achieve 100% accuracy through conventional training, and testing for deep neural networks is still in its early stages

Because deep neural networks are different from traditional software, the coverage criteria in traditional software testing techniques are not applied to deep neural network models, so for deep neural networks, there is a lack of coverage criteria for evaluating test adequacy

Images