arXiv cs.CL· Akinori Tanaka, Yukinao Akamatsu, Takuki Tomita, Yuji Hirono·· 2 天前AI 评分22
扩散语言模型作为开放量子系统：Superposition of Prompts 与 Quantum-Inspired Remasking
Diffusion Language Models as Open Quantum Systems
AI 导读
研究将扩散语言模型的连续时间前向与反向动力学嵌入开放量子系统框架，构建 τ-leaping 解缠方案，在每步同时更新多个 token 位置，实现基于叠加提示（SoP）和叠加思维（SoT）的生成。
正文 · 原文
Abstract:Diffusion language models generate text by reversing a classical diffusion process over token sequences. We show that their continuous-time forward and reverse dynamics can be embedded in the framework of open quantum systems, which describes quantum systems interacting with their environment. This connection provides a basis for extending diffusion sampling to superpositions of sentence states and for modifying the underlying generation dynamics. We construct a $\tau$-leaping unraveling scheme that simulates stochastic trajectories of these states while updating multiple token positions at each step. This enables generation conditioned on superposed prompts or thoughts, which we call Superposition of Prompts (SoP) and Superposition of Thoughts (SoT). The same formulation yields Quantum-Inspired Remasking (QIR): perturbing the masked-diffusion dynamics produces a confidence-based rule for revising previously generated tokens. Experiments with pretrained diffusion language models show that superposition-based generation can retain information from conditioning components not selected at the final measurement, while QIR improves LLM-assessed text quality when applied at an appropriate stage of generation.
| Comments: | 48 pages, 8 figures |
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL); Quantum Physics (quant-ph) |
| Report number: | RIKEN-iTHEMS-Report-26 |
| Cite as: | arXiv:2610.21672 [cs.LG] |
| (or arXiv:2610.21672v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21672 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Akinori Tanaka [view email]
[v1]
Thu, 15 Oct 2026 10:57:07 UTC (3,035 KB)
来源：arXiv cs.CL · arxiv.org