Open Problem: Active Representation Learning