While training, any sort of screen interruption (rotation, popup, back button, etc.) will cause the app to stop and attempting to return execution of the app will result in a black screen. I have debugging information logging when I am reading the training/test data files and after the screen has been interrupted I can see data files still being read, but no tensorflow operations are running, so this is also a huge memory leak.