VERDICT: Training-Free Step-Wise Verification of Multimodal Reasoning via Disagreement-Aware Consensus

📄 arXiv: 2608.10665v1 📥 PDF

作者: Rohit Sinha, Kunal Tilaganji, Tanuja Ganu, Nagarajan Natarajan, Amit Sharma, Vineeth Balasubramanian

分类: cs.AI, cs.CV, cs.GT

发布日期: 2026-08-11

备注: European Conference on Computer Vision 2026

💡 一句话要点

提出VERDICT以解决多模态推理验证问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 多模态推理 验证方法 训练无关 不一致性 共识评分

📋 核心要点

  1. 现有的验证方法依赖昂贵的标注监督，且在不同任务间表现不一致，难以有效验证推理链的正确性。
  2. 本文提出的VERDICT方法通过不一致性信息进行逐步验证，避免了对训练数据的依赖，具有领域无关性。
  3. 在六个基准测试中，VERDICT方法在性能上超越基础模型，证明了跨模态一致性在验证中的有效性。

📝 摘要（中文）

多模态大型语言模型常常生成包含细微错误的推理链，导致错误答案。现有的验证方法存在显著局限性，通常需要昂贵的标注监督，且在跨任务性能上不一致。本文提出了一种名为VERDICT的训练无关的逐步验证方法，通过明确的跨模态不一致性来计算共识分数，从而实现有效的推理步骤验证。VERDICT在六个基准测试中表现出色，较基础模型提升了最多5.95%。

🔬 方法详解

问题定义：本文旨在解决多模态推理中的验证问题，现有方法通常需要大量标注数据，且在不同任务间的表现不稳定，难以有效捕捉推理步骤的有效性。

核心思路：VERDICT方法通过将跨模态的不一致性作为关键信息，提出了一种训练无关的逐步验证机制，利用不一致性来识别有效的推理步骤。

技术框架：VERDICT的整体架构包括多个冻结的验证器，这些验证器之间的评分被视为一个协调博弈，通过闭合形式的解来计算共识分数，进而实现不一致性过滤和稳定性排序。

关键创新：VERDICT是首个明确将跨模态不一致性结构化并可操作的训练无关验证器，其通过不一致性信息提供了有效的验证信号，与传统方法相比具有本质的区别。

关键设计：该方法的设计包括不一致性评分的计算、共识分数的闭合形式解，以及对推理步骤的稳定性意识排序，确保了验证过程的高效性和准确性。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

在六个基准测试中，VERDICT方法相较于基础模型提升了最多5.95%的性能，且在与需要大量监督的领域特定评估者的比较中表现出色，证明了跨模态一致性在验证中的重要性。

🎯 应用场景

VERDICT方法在多模态推理验证中的应用潜力巨大，能够广泛应用于自然语言处理、计算机视觉等领域，提升模型的推理准确性和可靠性。未来，该方法可能推动更高效的验证机制的发展，减少对标注数据的依赖。

📄 摘要（原文）

Multimodal large language models often generate reasoning chains containing subtle errors that lead to incorrect answers. Current verification approaches have notable limitations. Existing approaches either require expensive labelled supervision with inconsistent cross-task performance or aggregate scores from multiple sources by simple aggregations, missing a key insight: when these scores disagree, that disagreement itself carries important information about whether a reasoning step is truly valid or not. We formalise this as a coupled scoring problem among disparate, frozen verifiers, interpretable as a coordination game with a unique closed-form equilibrium where agreement signals valid steps while disagreement reveals instability. Towards this end, we propose a training-free domain-agnostic step-wise verification approach we call VERDICT: VERification via Disagreement-Informed Coupled Thresholding. To our knowledge, VERDICT is the first training-free verifier that makes the structure of cross-modal disagreement explicit and actionable. It computes consensus scores through a closed-form solution, enabling both disagreement-aware filtering and stability-conscious ranking of reasoning steps. Evaluated across six benchmarks, \method consistently improves over the base model by up to +5.95%, and performs competitively with domain-specific critics that demand extensive supervision, demonstrating that cross-modal agreement provides robust verification signals without task-specific adaptation and Training-Free Verification