EasyDeepLearn

Why is naive DQN unstable and what specific fixes address which cause?

hard

Answer

  • (1) Correlated sequential data → replay buffer breaks correlation.
  • (2) Moving target Q'-target = Q → target network freezes target for N steps.
  • (3) Overestimation from noise in max → Double DQN.
  • (4) Reward magnitude / scale → reward clipping to [-1, 1] or reward normalization.
  • (5) Sparse reward → n-step returns or intrinsic curiosity.
  • Every DQN implementation combines at least (1) + (2); Rainbow adds all six.
Check yourself — multiple choice
  • Random
  • Correlation → replay; moving target → target net; overestimation → Double DQN; reward scale → clipping; sparse reward → n-step / curiosity — DQN needs replay + target net minimum
  • Not unstable
  • Only replay

DQN instability fixes: replay + target net + Double + clipping + n-step / curiosity.

#value-methods#deep-rl#engineering

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions