MInTRL: Off-policy Intervention can boost On-policy RL

Researchers introduced MInTRL, an off-policy intervention method that boosts on-policy reinforcement learning with verifiable rewards, according to the paper's headline. The approach addresses the limitation that on-policy RL restricts learning to trajectories the current policy can discover itself, while off-policy methods such as supervised fine-tuning can leverage external knowledge. The work has not yet been evaluated in the provided source beyond its stated premise.

Reinforcement learning with verifiable rewards is typically performed on-policy, keeping training data close to the current policy but limiting learning to trajectories that the policy can discover itself. Off-policy methods such as supervised fine-tuning, on the other hand, can leverage external kn