CompoNet grows one small module per task, each module attending to and composing the actions proposed by frozen earlier policies, yielding linear parameter growth and generally positive forward transfer in continual RL benchmarks.
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
-
Self-Composing Policies for Scalable Continual Reinforcement Learning
CompoNet grows one small module per task, each module attending to and composing the actions proposed by frozen earlier policies, yielding linear parameter growth and generally positive forward transfer in continual RL benchmarks.