Visual note-taking for deep learning