R1-ShareVL: Incentivizing Reasoning Capability of Multimodal Large Language Models via Share-GRPO
This work addresses a specific bottleneck in improving reasoning capabilities of MLLMs, which is incremental but could benefit AI systems requiring complex multimodal understanding.
The paper tackles the problem of sparse rewards and advantage vanishing in reinforcement learning for multimodal large language models (MLLMs) by proposing Share-GRPO, which expands the question space and shares reasoning trajectories, resulting in superior performance on six reasoning benchmarks.
In this work, we aim to incentivize the reasoning ability of Multimodal Large Language Models (MLLMs) via reinforcement learning (RL) and develop an effective approach that mitigates the sparse reward and advantage vanishing issues during RL. To this end, we propose Share-GRPO, a novel RL approach that tackle these issues by exploring and sharing diverse reasoning trajectories over expanded question space. Specifically, Share-GRPO first expands the question space for a given question via data transformation techniques, and then encourages MLLM to effectively explore diverse reasoning trajectories over the expanded question space and shares the discovered reasoning trajectories across the expanded questions during RL. In addition, Share-GRPO also shares reward information during advantage computation, which estimates solution advantages hierarchically across and within question variants, allowing more accurate estimation of relative advantages and improving the stability of policy training. Extensive evaluations over six widely-used reasoning benchmarks showcase the superior performance of our method. Code will be available at https://github.com/HJYao00/R1-ShareVL.