CSPO: Alleviating Reward Ambiguity for Structured Table-to-LaTeX Generation
For researchers working on structured document generation from images, CSPO provides a method to alleviate reward ambiguity in RL fine-tuning of multimodal LLMs, leading to more faithful LaTeX code generation.
CSPO introduces a reinforcement learning framework that uses component-specific rewards for structure, style, and content to improve table-to-LaTeX generation, outperforming conventional single-reward RL methods on hierarchical evaluation metrics.
Tables contain rich structured information, yet when stored as images their contents remain "locked" within pixels. Converting table images into LaTeX code enables faithful digitization and reuse, but current multimodal large language models (MLLMs) often fail to preserve structural, style, or content fidelity. Conventional post-training with reinforcement learning (RL) typically relies on a single aggregated reward, leading to reward ambiguity that conflates multiple behavioral aspects and hinders effective optimization. We propose Component-Specific Policy Optimization (CSPO), an RL framework that disentangles optimization across LaTeX tables components-structure, style, and content. In particular, CSPO assigns component-specific rewards and backpropagates each signal only through the tokens relevant to its component, alleviating reward ambiguity and enabling targeted component-wise optimization. To comprehensively assess performance, we introduce a set of hierarchical evaluation metrics. Extensive experiments demonstrate the effectiveness of CSPO, underscoring the importance of component-specific optimization for reliable structured generation.