Beyond Post-Hoc Rewards: Privileged Reasoning Exploration for Multimodal Embeddings

📄 arXiv: 2610.32474v1 📥 PDF

作者: Tianyu Yang, Yijun Tian, Lars Benedikt Kaesberg, Jan Philip Wahle, Bela Gipp, Terry Ruas, Daniel Kurzawe

分类: cs.AI, cs.CL

发布日期: 2026-10-23

💡 一句话要点

提出PILOT框架以优化多模态嵌入模型的推理探索

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

关键词: 多模态嵌入 推理探索 关系监督 变分推理 信息检索 深度学习

📋 核心要点

  1. 现有方法在训练嵌入模型时未能有效指导推理探索，导致检索相关线索的利用不足。
  2. PILOT框架通过将推理视为潜变量，结合先验和后验分布，优化推理探索过程。
  3. 在MMEB-V2基准上，PILOT在所有12个任务组中超越基线，视觉文档检索任务得分提升显著。

📝 摘要（中文）

随着多模态大语言模型（MLLMs）的进步，多模态嵌入技术取得了显著进展。现有方法通过自生成推理增强嵌入输入，但传统的检索监督未能有效指导推理探索。本文提出PILOT框架，将推理建模为潜变量，利用先验和后验分布来探索推理模式。实验表明，PILOT在78个任务的MMEB-V2基准上超越了主要基线，整体得分提升3.2至3.3分，尤其在视觉文档检索任务中表现突出。

🔬 方法详解

问题定义：本文要解决的问题是现有多模态嵌入模型在推理探索中的不足，传统方法未能有效利用检索相关的推理线索，导致嵌入质量受限。

核心思路：PILOT框架的核心思路是将推理建模为潜变量，通过先验和后验分布的结合，利用训练时的关系证据来指导推理探索，从而提高嵌入模型的性能。

技术框架：PILOT框架包括两个主要模块：先验推理和后验推理。先验推理仅基于输入条件，而后验推理则结合额外的关系证据进行推理探索。通过重要性加权变分目标，促进先验和后验的共同演化。

关键创新：PILOT的创新在于将推理视为潜变量，并通过先验-后验共同演化框架来优化推理探索，这一方法与传统的基于输入的推理优化方式有本质区别。

关键设计：在损失函数设计上，采用重要性加权变分目标，确保推理探索的有效性。同时，模型结构设计上，确保先验和后验推理模块能够共享信息并相互促进。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，PILOT在78个任务的MMEB-V2基准上超越主要基线，整体得分提升3.2至3.3分，尤其在视觉文档检索任务中，得分提升达6.1和6.4分，展现出显著的性能优势。

🎯 应用场景

该研究的潜在应用领域包括信息检索、自然语言处理和计算机视觉等。通过优化多模态嵌入模型的推理探索，PILOT框架可以在实际应用中提升检索系统的准确性和效率，具有重要的实际价值和未来影响。

📄 摘要（原文）

Advances in Multimodal Large Language Models (MLLMs) have driven substantial progress in multimodal embeddings. Recent methods exploit generative capabilities by augmenting embedded inputs with self-generated reasoning to produce more informative embeddings. However, conventional retrieval supervision for training embedding models specifies pairwise relevance relations without explicitly guiding the reasoning used for input augmentation. Such reasoning is neither directly observed nor uniquely determined by the input, allowing multiple plausible traces to introduce semantic cues of varying retrieval utility. Existing approaches largely optimize reasoning conditioned on the input alone, limiting the exploration of retrieval-relevant traces. We introduce PILOT, a framework that models reasoning as a latent variable and instantiates its prior and posterior distributions as distinct reasoning modes of a shared MLLM. The prior conditions only on the input, whereas the posterior uses additional training-time relational evidence to guide reasoning exploration. To translate the posterior's broader exploration into inference-time improvements, we derive an importance-weighted variational objective and instantiate it as a prior-posterior co-evolution framework that improves both modes under a shared inference-compatible reward while selectively transferring posterior trajectories to the prior according to their utility and prior reachability. On the 78-task MMEB-V2 benchmark, PILOT outperforms the primary baseline across all 12 task groups at both 2B and 7B parameters, improving overall scores by 3.2 and 3.3 points, respectively, with the largest gains in visual-document retrieval (+6.1 and +6.4 points). Our results show that relational supervision is most valuable not as a post-hoc reward but as a signal for shaping how reasoning is explored in training reasoning-enhanced embedding models.