Boosting Multimodal Reasoning with MCTS-Automated Structured Thinking
Jinyang Wu, Mingkuan Feng, Shuai Zhang, Ruihan Jin, Feihu Che, Zengqi Wen, Jianhua Tao
2025-02-06
Summary
This paper talks about a new method called AStar, which improves how multimodal large language models (MLLMs) solve problems that involve both text and visuals. It uses a special technique called Monte Carlo Tree Search (MCTS) to make the models better at reasoning and understanding complex tasks.
What's the problem?
Multimodal AI models are good at combining text and images, but they struggle with complex visual reasoning tasks. Current methods often require a lot of data and computing power, which makes them inefficient and limits their ability to extract insights effectively.
What's the solution?
The researchers developed AStar, a system that uses MCTS to create structured thinking patterns for the models. By focusing on high-level reasoning with minimal data, AStar combines the model's internal abilities with external guidelines to improve accuracy while using fewer resources. Experiments showed that AStar performs better than other advanced models like GPT-4o on benchmarks like MathVerse.
Why it matters?
This research is important because it helps AI models become more efficient and accurate at solving complex problems that require both text and visual reasoning. It provides a way to balance performance and efficiency, making these models more practical for real-world applications like education, science, and technology.
Abstract
Multimodal large language models (MLLMs) exhibit impressive capabilities but still face challenges in complex visual reasoning. While recent efforts attempt to enhance MLLMs' reasoning by incorporating OpenAI o1-like structured thinking through explicit search structures or teacher-guided distillation, they often struggle to balance performance and efficiency. A critical limitation is their heavy reliance on extensive data and search spaces, resulting in low-efficiency implicit insight extraction and data utilization. To address this, we propose AStar, an Automated Structured thinking paradigm for multimodal reasoning via Monte Carlo Tree Search (MCTS). AStar automatically derives high-level cognitive reasoning patterns from limited data using MCTS-powered hierarchical structures. Building on these explicit patterns, we design a unified reasoning framework that seamlessly integrates models' internal reasoning capabilities and external reasoning guidelines, enabling efficient inference with minimal tree iterations. This novel paradigm strikes a compelling balance between performance and efficiency. Extensive experiments demonstrate AStar's effectiveness, achieving superior accuracy (54.0%) on the MathVerse benchmark with a 7B backbone, surpassing GPT-4o (50.2%) while maintaining substantial data and computational efficiency.