Func-R1 improves math function reasoning in multimodal ai models
Func-R1: Incentivizing Mathematical Function Reasoning in Multimodal Large Language Models
Computation and Language
Summary
Multimodal AI models often struggle to understand math functions correctly when both text and images are involved, mixing up visual cues and logic. The authors found that current models sometimes ignore important visual information when solving math problems. They created Func-R1, a new approach that helps AI carefully examine visuals and think through math step-by-step. Their method uses a special training process to better connect what the model sees with the math reasoning it needs to do. Func-R1 outperforms other models, including a leading version from GPT, on challenging math function tasks.
What this means in practice
- •For ai developers: Build smarter AI systems that combine images and math text for more accurate function problem solving.
- •For software engineers: Improve multimodal AI tool accuracy in scientific and educational software that involves math function reasoning.
Authors
Mingze Yin, Xiaohan Wang, Dian Li, Haichao Yao, Yilin Zhao, Youjun Chen, Gang Liu, Jintai Chen, Yiheng Zhu, Chang-Yu Hsieh, Aimin Pan
Abstract
Performing deliberate mathematical reasoning in visual contexts is a hallmark of advanced Multimodal Large Language Models (MLLMs) and requires a sophisticated synthesis of perceptual grounding and symbolic logic. However, in the realm of mathematical functions, our investigation reveals a critical modality interference phenomenon: even advanced models, while performing textual computational reasoning, tend to disregard or misinterpret essential visual cues. To address this challenge, we propose Func-R1, which synergistically harmonizes precise visual perception and rigorous logical reasoning. Concretely, built upon an explicitly decoupled architecture, we employ a hierarchical post-training framework to progressively identify critical visual evidence and conduct in-depth theoretical reasoning. Furthermore, the Perception-Aligned Theoretic Optimization (PATO) strategy is proposed to steer policy updating towards internalizing fundamental theoretical properties while dynamically rectifying heterogeneous visual information throughout the reasoning process. Extensive experiments across diverse benchmarks demonstrate that Func-R1 delivers the optimal performance among open-source MLLMs, even surpassing GPT-5 with an 8.4% improvement on MathVerse's function-oriented tasks.