为大模型推理提供分组条件下的可靠性能保障，兼顾效率与安全性。
Conditional Performance Guarantee for Large Reasoning Models
- 通过输入空间分组实现条件化风险控制，替代传统边际保证。
- 在异质场景下，计算效率比传统方法提升显著，且风险可控。
- 适用于需要可解释性与可靠性保障的复杂推理任务。
大型推理模型虽通过扩展思维链展现出强大性能，但计算成本仍高。可能近似正确（PAC）推理通过自适应切换思考与非思考模式，提供了高效推理的统计保障，但仅在边际情况下成立，无法提供精确的条件覆盖。本文提出G-PAC推理框架，通过划分输入空间，在群体层面提供类PAC保证。我们设计了两种实例：针对已知分组结构的群组PAC（G-PAC）推理，以及针对未知分组的聚类PAC（C-PAC）推理。理论证明了两者均能实现群体条件下的风险控制，且在异质设置中，分组策略可严格提升效率。在多个推理基准上的实验表明，G-PAC与C-PAC成功实现了群体条件风险控制，并保持显著的计算节省。
原文摘要 · Abstract (English)
Large reasoning models have shown strong performance through extended chain-of-thought reasoning, yet their computational cost remains significant. Probably approximately correct (PAC) reasoning provides statistical guarantees for efficient reasoning by adaptively switching between thinking and non-thinking models, but the guarantee holds only in the marginal case and does not provide exact conditional coverage. We propose G-PAC reasoning, a practical framework that provides PAC-style guarantees at the group level by partitioning the input space. We develop two instantiations: Group PAC (G-PAC) reasoning for known group structures and Clustered PAC (C-PAC) reasoning for unknown groupings. We prove that both G-PAC and C-PAC achieve group-conditional risk control, and that grouping can strictly improve efficiency over marginal PAC reasoning in heterogeneous settings. Our experiments on diverse reasoning benchmarks demonstrate that G-PAC and C-PAC successfully achieve group-conditional risk control while maintaining substantial computational savings.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。