English(EN) Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models

新的反射掩码技术增强了扩散模型的推理能力

作者 PulseAugur 编辑部 · [2 个来源] · 2026-06-15 13:34

研究人员推出了一种名为反射掩码（RM）的新型训练后技术，旨在增强掩码扩散模型（MDMs）的推理能力。与依赖顺序生成的自回归模型不同，RM允许MDMs通过多轮掩码和去噪迭代地改进先前的输出，类似于人类纠错。该方法包含历史参考（History Reference），一种无参数机制，利用中间去噪状态从先前的轮次中汲取见解。这种方法无需架构更改，并在文本生成、数独和图像编辑等各种任务中展示了持续的性能提升，将RM定位为MDM推理的基础要素。 AI

影响这项研究可能使扩散模型能够执行更复杂的推理任务，从而可能提高它们在创意生成和解决问题等领域的效用。

排序理由该集群包含一篇详细介绍AI模型新方法的学术论文。

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.CL TIER_1 English(EN) · Yanming Zhang, Yihan Bian, Jingyuan Qi, Yuguang Yao, Lifu Huang, Tianyi Zhou · 2026-06-16 04:00

Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models

arXiv:2606.16700v1 Announce Type: new Abstract: While reasoning on autoregressive (AR) models is often performed by chain-of-thought reasoning and reflection, their refinement of previous outputs still relies on fully sequential generation, even when only local edits are needed. …
arXiv cs.CL TIER_1 English(EN) · Tianyi Zhou · 2026-06-15 13:34

Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models

While reasoning on autoregressive (AR) models is often performed by chain-of-thought reasoning and reflection, their refinement of previous outputs still relies on fully sequential generation, even when only local edits are needed. In contrast, the masking mechanism in Mask Diffu…

报道来源 [2]

Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models

Multi-Turn Reflective Masking Elicits Reasoning in Mask Diffusion Models

相关实体

相关话题