DEEPRUBRIC: Evidence-Tree Rubric Supervision for Efficient Reinforcement Learning of Deep Research Agents
Researchers have introduced DeepRubric, a novel framework for constructing query-rubric pairs to improve the efficiency of reinforcement learning for deep research agents. This method synthesizes aligned query-rubric pairs by first identifying evaluation targets and then building an evidence tree to ensure rubrics accurately reflect the information needs of a given query. By training the DeepRubric-8B model with this approach, the researchers achieved comparable performance to existing state-of-the-art models while using significantly fewer computational resources. AI
IMPACT This framework could lead to more efficient training of AI agents for complex research tasks, reducing computational costs.