DeepFinder: A New White-Box Testing Framework for Deep Neural Networks

DNNs have made important developments in the past decades, however, due to some potential problems of DNNs leading to security incidents, more rigorous and systematic testing of DNNs is desired to ensure the correctness of DNNs. Therefore, we design DeepFinder, a new white-box testing framework for neural networks, to find the corner cases of neural networks by generating testing examples. In this paper, we design comparative experiments to show the advantages of DeepFinder, which can find the corner cases of neural networks better and these corner cases are more evenly distributed.