Abstract: A system for performing software testing uses machine learning to extract features from a user interface of an app, classify screen types and screen elements of the user interface, and implement flows of test sequences to test the app. Training is performed to train the system to learn common application states of an application graph and to navigate through an application. In some implementations, the training includes Q-learning to learn how to navigate to a selected screen state. In some implementations, there is reuse of classifiers cross-application and cross platform.