Revisiting deep neural network test coverage from the test effectiveness perspective

نویسندگان

چکیده

Many test coverage metrics have been proposed to measure the Deep Neural Network (DNN) testing effectiveness, including structural and non-structural coverage. These are based on fundamental assumption: they correlated with effectiveness. However, assumption is still not validated sufficiently reasonably, which brings question usefulness of DNN This paper conducted a revisiting study existing from effectiveness perspective, effectively validate assumption. Here, we carefully considered diversity subjects, three criteria, both typical state-of-the-art metrics. Different all studies that deliver negative conclusions coverage, identified some positive their perspective. In particular, found complementary relationship between practical usage scenarios promising research directions for these

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Relationship Between Test Effectiveness and Coverage

A model has recently been developed that relates test coverage and defect coverage. It has been shown that this model can be used to estimate the residual defect density. Defect finding capability of sets of tests with different coverage values has recently been studied by Frankl and Iakounenko. They have defined measure test effectiveness, and have evaluated it for programs containing a single...

متن کامل

Biodiesel production from tomato seed and its engine emission test and simulation using Artificial Neural Network

In this study, tomato seed oil was used to produce Biodiesel fuel. To reduce the percentage of free fatty acids, the oil was reacted at a temperature of 40, 50, and 60°C with a mixture of sulphuric acid and the industrial methanol with different molar ratios of oil. The best conversion efficiency was achieved at 60°C and a molar ratio of 1:9. In the transesterification step, biodiesel was produ...

متن کامل

test

چکیده ندارد.

15 صفحه اول

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Journal of software

سال: 2023

ISSN: ['1796-217X']

DOI: https://doi.org/10.1002/smr.2561