Meta-RL agents trained on a strictly greedy objective show information-seeking exploratory behavior when the environment has recurring structure and the agent has memory, according to bandit and gridworld experiments.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Exploitation Is All You Need... for Exploration
Meta-RL agents trained on a strictly greedy objective show information-seeking exploratory behavior when the environment has recurring structure and the agent has memory, according to bandit and gridworld experiments.