MANGER uses RND-computed observation novelty to give each agent a different number of extra Q-learning updates, improving sample efficiency and behavioral diversity in cooperative MARL.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Novelty-Guided Data Reuse for Efficient and Diversified Multi-Agent Reinforcement Learning
MANGER uses RND-computed observation novelty to give each agent a different number of extra Q-learning updates, improving sample efficiency and behavioral diversity in cooperative MARL.