Agent57 — how did it achieve above-human on all Atari?
hardAnswer
- Combined: (1) meta-controller balancing exploration ↔ exploitation across an episode by choosing among a family of policies with different exploration weights.
- (2) NGU (Never Give Up) intrinsic motivation: episodic + long-term.
- (3) Distributed R2D2 backbone (recurrent DQN + distributed).
- Reached human-level on all 57 Atari games (Puigdomènech et al. 2020).
- Milestone in general RL.
Check yourself — multiple choice
- Random
- Meta-controller across policies with different exploration weights + NGU (episodic + long-term intrinsic) + distributed R2D2 backbone → above-human on all 57 Atari games
- Same as DQN
- Not real
Agent57: meta-controller + NGU + R2D2; above-human on all Atari.
#exploration#deep-rl
Practise Reinforcement Learning
214 interview questions in this topic.