DS dimension

A tighter sample-complexity bound for multiclass learning

This paper closes a long-standing gap in multiclass sample complexity by tying it tightly to DS dimension, with implications for list learning too.