Shift Before You Learn: Enabling Low-Rank Representations in Reinforcement Learning | Read Paper on Bytez