LLM Hallucination Detection via Low-rank Multilinear Decomposition

查看原文 / View Original 编辑 / Edit

Computer Science > Computation and Language arXiv:2610.36191 (cs) [Submitted on 26 Oct 2026] Title:LLM Hallucination Detection via Low-rank Multilinear Decomposition Authors:Devansh Sharma, Faizul Rakib Sayem, Shahana Ibrahim View a PDF of the paper titled LLM Hallucination Detection via Low-rank Multilinear Decomposition, by Devansh Sharma and 2 other authors View PDF HTML (experimental)

计算机科学 > 计算与语言 arXiv:2610.36191 (cs) [提交于 2026 年 10 月 26 日] 标题：通过低秩多线性分解进行大语言模型幻觉检测作者：Devansh Sharma, Faizul Rakib Sayem, Shahana Ibrahim 查看题为《通过低秩多线性分解进行大语言模型幻觉检测》的论文 PDF，作者为 Devansh Sharma 及其他 2 位作者查看 PDF HTML（实验性）

Abstract:Large language models (LLMs) frequently generate fluent yet factually incorrect responses, raising concerns about their reliability. Recent hallucination detectors increasingly exploit internal representations, but extracting informative signals from high-dimensional hidden states remains challenging. Existing approaches often rely on selected activations or pooling-based compression, potentially overlooking discriminative information distributed across Transformer layers, feature modes, and tokens.

摘要：大语言模型（LLM）经常生成流畅但事实错误的回答，引发了对其可靠性的担忧。近期的幻觉检测器越来越多地利用内部表征，但从高维隐藏状态中提取信息信号仍然具有挑战性。现有方法通常依赖于选定的激活值或基于池化的压缩，这可能会忽略分布在 Transformer 层、特征模式和标记（token）中的判别性信息。

We propose LOTUS (Low-rank Order-statistic TUcker-based Subspace learning), a multilinear low-rank decomposition framework for hallucination detection that extracts compact representations from LLM hidden states while reducing dependence on labeled data. Our approach first summarizes variable-length token representations using complementary statistical descriptors that capture peak activations, activation distributions, and cross-layer changes.

我们提出了 LOTUS（基于低秩顺序统计 Tucker 分解的子空间学习），这是一个用于幻觉检测的多线性低秩分解框架，它能从 LLM 隐藏状态中提取紧凑的表征，同时减少对标注数据的依赖。我们的方法首先使用互补的统计描述符来总结变长标记表征，这些描述符能够捕捉峰值激活、激活分布以及跨层变化。

We then leverage low-rank multilinear decomposition to extract compact representations that capture the underlying structure of LLM hidden states, which are subsequently used to train a lightweight hallucination classifier. Experiments across multiple LLMs and question-answering benchmarks demonstrate improved hallucination detection performance over recent state-of-the-art baselines.

随后，我们利用低秩多线性分解来提取能够捕捉 LLM 隐藏状态底层结构的紧凑表征，并将其用于训练轻量级的幻觉分类器。在多个 LLM 和问答基准测试上的实验表明，该方法在幻觉检测性能上优于近期最先进的基准模型。