Chart-RVR: Reinforcement Learning with Verifiable Rewards for Explainable Chart Reasoning

2025/10/13 by Sinha, Sanchit, Frunza, Oana, Rasul, Kashif +2
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2510.10973

Abstract

The capabilities of Large Vision-Language Models (LVLMs) have reached state-of-the-art on many visual reasoning tasks, including chart reasoning, yet they still falter on out-of-distribution (OOD) data, and degrade further when asked to produce their chain-of-thought (CoT) rationales, limiting explainability. We present Chart-RVR, a general framework that fine-tunes LVLMs to be more robust and explainable for chart reasoning by coupling Group Relative Policy Optimization (GRPO) with automatically verifiable rewards. Our framework comprises of three rewards that maximize: (i) correct chart-type classification, (ii) faithful chart table reconstruction, and (iii) process conformity. Applied to 3-billion-parameter LVLMs, Chart-RVR consistently outperforms standard supervised fine-tuning (SFT) on both in-distribution and out-of-distribution datasets, closing the OOD performance gap while improving rationale fidelity. The resulting models, the Chart-RVR-3B series, achieve state-of-the-art results on six chart-reasoning benchmarks spanning in-domain and OOD settings, surpassing all existing models of comparable size. Beyond accuracy, Chart-RVR yields more interpretable CoT rationales, strengthening trust and reliability - showcasing the power of verifiable rewards with GRPO for training reliable, interpretable chart-reasoning models.

Citations

• More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models
• SCI-Reason: A Dataset with Chain-of-Thought Rationales for Complex Multimodal Reasoning in Academic Areas
• ChartQAPro: A More Diverse and Challenging Benchmark for Chart Question Answering
• Boosting Chart-to-Code Generation in MLLM via Dual Preference-Guided Refinement
• Visual-RFT: Visual Reinforcement Fine-Tuning
• SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
• ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding
• Graph-Based Multimodal Contrastive Learning for Chart Question Answering
• Improve Vision Language Model Chain-of-thought Reasoning
• EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding
• ChartGemma: Visual Instruction-tuning for Chart Reasoning in the Wild
• ChartInsights: Evaluating Multimodal Large Language Models for Low-Level Chart Question Answering
• TinyChart: Efficient Chart Understanding with Visual Token Merging and Program-of-Thoughts Learning
• Chart-based Reasoning: Transferring Capabilities from LLMs to VLMs
• ChartBench: A Benchmark for Complex Visual Reasoning in Charts
• UniChart: A Universal Vision-language Pretrained Model for Chart Comprehension and Reasoning
• Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
• Reading and Reasoning over Chart Images for Evidence-based Automated Fact-Checking
• MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering
• Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding
• ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning
• PlotQA: Reasoning over Scientific Plots
• Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
• Proximal Policy Optimization Algorithms

Related

• Chart-R1: Chain-of-Thought Supervision and Reinforcement for Advanced Chart Reasoner
• BigCharts-R1: Enhanced Chart Reasoning with Visual Reinforcement Finetuning
• ChartM³: A Multi-Stage Code-Driven Pipeline for Constructing Multi-Dimensional and Multi-Step Visual Reasoning Data in Chart Comprehension
• Visual Programmability: A Guide for Code-as-Thought in Chart Understanding
• Breaking the SFT Plateau: Multimodal Structured Reinforcement Learning for Chart-to-Code Generation