ARES uses attention weights from a return-predicting transformer to generate dense shaped rewards from fully delayed reward episodes, improving RL training in many test environments.
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
CONDITIONAL 1representative citing papers
citing papers explorer
-
Attention-Based Reward Shaping for Sparse and Delayed Rewards
ARES uses attention weights from a return-predicting transformer to generate dense shaped rewards from fully delayed reward episodes, improving RL training in many test environments.