Evo-PI: Aligning Medical Reasoning via Evolving Principle-Guided Supervision

2026-06-30

AI Summary

Existing static supervision signals struggle to effectively guide multimodal large language models in complex reasoning tasks, leading to limited generalization and performance saturation. To address this, this work proposes the Evo-PI framework, which introduces— for the first time—iteratively evolvable, linguistically articulated reasoning principles into the supervision mechanism. Specifically, in medical visual question answering (VQA), Evo-PI establishes a co-evolutionary loop between reasoning principles and model behavior, enabling dynamic alignment and mutual refinement of supervision signals and reasoning capabilities. Built upon structured vision–language reasoning, the method integrates principle generation, evaluation, and evolution mechanisms, achieving significant performance gains across eight medical VQA benchmarks, with accuracy improvements up to 24.6%, thereby demonstrating its effectiveness and broad applicability.

Abstract

Despite recent progress, the reasoning capabilities of large multimodal language models (MLLMs) remain fundamentally constrained by static supervision, where fixed prompts, rules, or reward models provide non-adaptive guidance throughout training. Such static signals are often sufficient to enforce output formats, but fail to shape the underlying reasoning process, leading to brittle generalization and performance saturation in complex decision-making tasks. We propose Evo-PI, a principle-centric learning framework that treats reasoning principles as explicit, language-based supervision signals that can be generated, evaluated, and iteratively evolved. Instead of relying on fixed rewards, Evo-PI enables a co-evolutionary loop in which principles guide model reasoning, while model behaviors in turn refine the principles that supervise them. This dynamic alignment mechanism allows supervision to progressively adapt to the model's reasoning deficiencies. We instantiate Evo-PI in medical visual question answering as a high-stakes testbed requiring structured visual-textual reasoning. Across eight benchmarks and multiple model backbones, Evo-PI consistently improves reasoning accuracy, achieving gains of up to 24.6%. Our results suggest that evolving principle-guided supervision offers a scalable and general paradigm for training expert-aligned reasoning in MLLMs. Code is available at https://github.com/zhengxianda/Evo_PI.

Problem

• static supervision
• reasoning process
• large multimodal language models
• generalization
• medical reasoning

Innovation

• principle-guided supervision
• evolving supervision
• co-evolutionary learning
• medical visual question answering
• multimodal reasoning