DQN training slows down significantly over time

Question

I am training a DQN on the pong gym environment to replicate the original DQN "Human-Level Control..." paper. My algorithm works fine and converges on a smaller te