思考扩散：惩罚与引导扩散多模态语言模型中的视觉定位推理
Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models
针对扩散多模态大模型在思维链推理中过早生成答案且视觉定位不足的问题，提出位置与步数惩罚（PSP）和视觉推理引导（VRG），可复用于其他并行生成式多模态模型的推理增强。
原文依据 dMLLMs often generate the final answer token at a very early timestep
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论扩散大语言模型（dLLMs）正成为自回归（AR）大语言模型的有前景替代方案。近期，这一范式被扩展到多模态任务，催生了扩散多模态大语言模型（dMLLMs）。这些模型有望在保留大语言模型推理能力的同时，通过并行生成实现更快的推理。然而，当与思维链（CoT）推理结合时，dMLLMs 表现出两个关键问题。第一，我们观察到 dMLLMs 经常在非常早的时间步生成最终答案词元。这一趋势表明模型在充分推理之前就确定了答案，导致推理性能下降。第二，在初始时间步，dMLLMs 对视觉提示的依赖极小，展现出与自回归视觉语言模型根本不同的视觉信息利用模式。总之，这些发现表明 dMLLMs 倾向于在未充分基于视觉输入的情况下过早生成最终答案。为解决这些局限，我们提出了位置与步数惩罚（PSP）和视觉推理引导（VRG）。PSP 在早期时间步惩罚较后位置的词元，延迟过早的答案生成，并鼓励跨时间步的渐进推理。VRG 受无分类器引导启发，放大视觉定位信号以增强模型与视觉证据的对齐。在不同 dMLLMs 上的大量实验表明，我们的方法在准确率上最高提升 7.5%，同时相比使用四倍扩散步数的推理实现了超过 3 倍的加速。我们的代码将在发表后发布。
英文原文摘要 1701 字符
Diffusion large language models (dLLMs) are emerging as promising alternatives to autoregressive (AR) LLMs. Recently, this paradigm has been extended to multimodal tasks, leading to the development of diffusion multimodal large language models (dMLLMs). These models are expected to retain the reasoning capabilities of LLMs while enabling faster inference through parallel generation. However, when combined with Chain-of-Thought (CoT) reasoning, dMLLMs exhibit two critical issues.First, we observe that dMLLMs often generate the final answer token at a very early timestep. This trend indicates that the model determines the answer before sufficient reasoning, leading to degraded reasoning performance. Second, during the initial timesteps, dMLLMs show minimal dependency on visual prompts, exhibiting a fundamentally different pattern of visual information utilization compared to AR vision-language models. In summary, these findings indicate that dMLLMs tend to generate premature final answers without sufficiently grounding on visual inputs.To address these limitations, we propose Position & Step Penalty (PSP) and Visual Reasoning Guidance (VRG). PSP penalizes tokens in later positions during early timesteps, delaying premature answer generation and encouraging progressive reasoning across timesteps. VRG, inspired by the classifier-free guidance, amplifies visual grounding signals to enhance the model's alignment with visual evidence.Extensive experiments across various dMLLMs demonstrate that our method achieves up to 7.5% higher accuracy while delivering more than 3x speedup compared to reasoning with four times more diffusion steps. Our code will be released after publication.