On Proper Learnability between Average- and Worst-case Robustness | Read Paper on Bytez