arXiv cs.CL· Zhongxing Xu, Zhonghua Wang, Zhe Qian, Shiyan Su, Ming Hu, Xiaocheng Zou, Wei Feng, Mingquan Lin, Yuyin Zhou, Yifan Peng, Hamid Rezatofighi, Zongyuan Ge·· 10 小时前AI 评分31
重新思考尖峰：用步感知解码缓解 MDLM 幻觉
Rethinking in Spikes: Mitigating Hallucinations in MDLMs with Step-Aware Decoding
AI 导读
研究者提出 StepRefine，一种即插即用的解码策略，用于缓解多模态扩散语言模型（MDLM）的幻觉。该方法通过测量每个去噪步相对邻近时间窗口的熵偏差来识别异常步，并在熵尖峰处用概率加权嵌入保留多样推理假设、进行多轮连续细化，再由细化控制器监测分布稳定性以切回标准离散去噪。在多模态基准上，StepRefine 在不同 MDLM 上均显著缓解幻觉。
正文 · 原文
Authors:Zhongxing Xu, Zhonghua Wang, Zhe Qian, Shiyan Su, Ming Hu, Xiaocheng Zou, Wei Feng, Mingquan Lin, Yuyin Zhou, Yifan Peng, Hamid Rezatofighi, Zongyuan Ge
Abstract:Recent advancements in multimodal diffusion language models (MDLMs) have exhibited strong global modeling capabilities through iterative refinement. We observe that mask uncertainty follows an overall downward trend during iterative denoising, while a few denoising steps exhibit localized entropy spikes closely associated with hallucinations. We argue that anomalous steps can be identified by measuring the entropy deviation of each denoising step from its neighboring temporal window. We hypothesize that discrete token commitment discards the distributional information at masked positions, limiting the continued interaction between candidate semantics and the context during entropy spikes. With this goal, we present StepRefine, a plug-and-play decoding strategy that leverages semantic context to achieve reliable denoising. The core of our method lies in a step-aware switch between discrete denoising and latent refinement. The model employs probability-weighted embeddings at entropy spike steps to preserve diverse reasoning hypotheses and performs multiple rounds of continuous refinement. Meanwhile, a refinement controller monitors distributional stability to reduce unnecessary loops and switch the model back to standard discrete denoising. Through extensive experiments, StepRefine demonstrates significant hallucination mitigation across different MDLMs on multimodal benchmarks.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.12901 [cs.CL] |
| (or arXiv:2611.12901v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.12901 arXiv-issued DOI via DataCite |
Submission history
From: Zhonghua Wang [view email]
[v1]
Wed, 14 Oct 2026 07:52:29 UTC (4,035 KB)
来源：arXiv cs.CL · arxiv.org