发表机构
University of Technology Sydney; The University of Sydney; CSIRO; ELLIS Institute Finland; University of Turku（悉尼科技大学; 悉尼大学; 澳大利亚联邦科学与工业研究组织; 芬兰ELLIS研究所; 图尔库大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
将多路径LLM推理视为无线通信中的多样性合并问题，证明多数投票为最优，提出自适应K规则，在5个模型12个基准上保留96%-103%准确率。
AI 中文摘要
自一致性（SC）等多路径推理方法采样$K$条推理路径并选择最频繁的答案。然而，随着$K$的增加，其收益迅速趋于平稳，现有方法无法预测这种饱和何时发生。我们将多路径LLM推理形式化为无线通信中的多样性合并问题：每条路径是一个带噪声的信道观测，路径正确性的成对相关性将投票的设计效应有效样本量限制在有限上限。广义最小二乘（GLS）分析表明，在可交换性下，潜在嵌入的最优对称线性合并器是均匀的，支持多数投票作为标准SC中的自然默认，同时在异构提示模板分支下为加权或剪枝留出空间。在5个模型和12个基准上，提示模板多样性在57个有效单元中的55个中降低了路径相关性，对开放式问答的影响最强。我们推导出一种自适应K规则，使用四路径试点选择$K^*$，在数学、问答和NLU任务上保留了MV@$K{=}32$准确率的96%--103%。
英文摘要
Multi-path reasoning methods such as self-consistency (SC) sample $K$ reasoning paths and choose the most frequent answer. However, their gains quickly plateau as $K$ increases, and existing methods do not predict when this saturation will occur. We formalize multi-path LLM reasoning as a diversity combining problem from wireless communications: each path is a noisy channel observation, and the pairwise correlation of path correctness caps the design-effect effective sample size of the vote at a finite ceiling. Generalized least squares (GLS) analysis shows that, under exchangeability, the optimal symmetric linear combiner of latent embeddings is uniform, supporting majority vote as the natural default in standard SC while leaving room for weighting or pruning under heterogeneous prompt-template branches. Across 5 models and 12 benchmarks, prompt-template diversity reduces path correlation in $55$ of $57$ valid cells, with the strongest effect on open-ended QA. We derive an Adaptive-K rule that uses a four-path pilot to select $K^*$, retaining $96$--$103\%$ of MV@$K{=}32$ accuracy across Math, QA, and NLU.
CommentsAccepted by NeurIPS 2026
