双世界：基于等变性的证据型推理弃权（不执行）方法
Twin Worlds: Equivariance-Based Abstention for Evidence-Grounded Reasoning
浏览论文内容
中文总结 AI 辅助
本研究提出双世界框架，通过等变性弃权提升大语言模型知识密集型推理的可靠性，在四个基准和三个模型主干上优于基线方法，可识别答案未可靠基于证据的情况。
中文摘要 AI 辅助
知识密集型推理要求大语言模型（LLMs）将答案建立在提供的证据基础上。当证据不足时，模型应弃权（不执行）而非自信地生成无支撑的答案。现有弃权方法依赖不确定性估计或证据充分性检查，但均未测试由提供证据与模型内部记忆参数交互驱动的生成推理过程是否真正基于证据。一个关键影响因素是上下文中的实体提及会激活记忆关联，导致模型生成看似合理但无证据支撑的响应。我们提出 Twin Worlds（TW，双世界）框架，通过基于等变性的弃权提升知识密集型推理的可靠性：与要求输出保持不变的不变性不同，等变性要求输出在实体替换时相应变换。基于证据的模型应在实体被替换且关系保留时，生成的答案发生一致变化。TW 通过对原始输入进行保留关系结构且减少参数先验的类型化替换构建多个世界，并将等变性违反作为弃权信号。在四个基准和三个模型主干上，TW 可识别答案未可靠基于提供证据的情况，且优于基于不确定性和充分性的基线方法。
英文摘要
Knowledge-intensive reasoning requires Large Language Models (LLMs) to ground answers in provided evidence. When evidence is insufficient, it is desirable that models abstain rather than confidently generating unsupported answers. Existing abstention methods rely on uncertainty estimation or evidence sufficiency checks, but neither tests whether the reasoning process for generation, driven by the interaction of provided evidence and the model's internal memory parameters, is actually grounded in the evidence. A key contributing factor is that entity mentions in context activate memorised associations, causing models to generate plausible responses ungrounded in evidence. We propose Twin Worlds (TW), a framework for improving reliability in knowledge-intensive reasoning through equivariance-based abstention: unlike invariance, which requires outputs to remain unchanged, equivariance requires outputs to transform correspondingly under entity substitutions. A model grounded in the evidence should produce answers that shift consistently when entities are substituted while their relations are preserved. TW constructs multiple worlds via typed substitutions of the original input that preserve relational structure while reducing parametric priors, and uses equivariance violations as an abstention signal. Across four benchmarks and three model backbones, TW identifies when answers are not reliably grounded in the provided evidence and outperforms uncertainty- and sufficiency-based baselines.
发表机构
- RMIT University（皇家墨尔本理工大学）
- Jilin University（吉林大学）
- Nanyang Technological University（南洋理工大学）
机构由 AI 辅助整理，请以论文原文为准。
