HIGHLIGHTS
SUMMARY
As suggested,, a test metric does not have to be strongly related to adversarial samples, a specific type of defects corresponding to the robustness requirement of a neural_network. By contrast, a recurrent neural_network (RNN) processes an input sequence by iteratively taking inputs one by one. The proposed set of coverage metrics plays the role of such guidance - as suggested in Fig 2(Left), coverage-guided testing generates a set of test cases to exploit the internal behaviour of the neural_networks. The authors treat adversarial samples as a proxy to evaluate the effectiveness of the . . .
If you want to have access to all the content you need to log in!
Thanks :)
If you don't have an account, you can create one here.