Optimizing the human learnability of abstract network representations