A gated mixture of Q-functions with different discount factors, trained with undiscounted Bellman error, adapts its temporal horizon in small MiniGrid tasks, but its theoretical justification is circular and baseline comparisons are missing.
Sutton , abstract =
1 Pith paper cite this work, alongside 149 external citations. Polarity classification is still indexing.
1
Pith paper citing it
149
external citations · OpenAlex
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Adaptive Multi-Horizon Reinforcement Learning
A gated mixture of Q-functions with different discount factors, trained with undiscounted Bellman error, adapts its temporal horizon in small MiniGrid tasks, but its theoretical justification is circular and baseline comparisons are missing.