DeepMutation++

DeepMutation++ supports both typical types of neural networks, i.e., FNNs and RNNs. Given a DNN model and a set of test data under analysis, it first leverages the provided mutation operators to generate a set of high quality DNN mutants with a user specified quality threshold. After a certain number of mutants are generated, DeepMutation++ analyzes the behavior differences of original DNN and the generated DNN mutants against the provided test inputs for robustness analysis. Moreover, DeepMutation++ specially features a dynamic mode to analyze the vulnerable input segments for a RNN model based on the mutant DNNs generated on-the-fly. Finally, DeepMutation++ outputs the analysis reports that indicate test data quality and DNN robustness at the same time.