Deep RL agents' best experienced trajectories are 2-3 times better than their learned policy's average return, suggesting exploitation and optimization issues dominate exploration challenges.
The theory of dynamic programming
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Is Exploration or Optimization the Problem for Deep Reinforcement Learning?
Deep RL agents' best experienced trajectories are 2-3 times better than their learned policy's average return, suggesting exploitation and optimization issues dominate exploration challenges.