Stabilising Experience Replay for Deep Multi-Agent Reinforcement Learning

I had seen this paper before, but I was confused by the "Fingerprint". This kind of makes more sense.