大模型生成结构化数据时，难以更新预训练的类别先验，导致罕见类无法生成。
Categorical Prior Lock-in: Why In-Context Learning Fails for Structured Data
- 通过上下文学习调整生成分布，但无法改变预训练的类别偏好
- 在高基数表格数据上，罕见类别生成率接近零，存在明显上限
- 微调可突破限制，但有记忆泄露和输出不稳定的隐患
大型语言模型（LLMs）越来越多地被用作结构化数据的条件生成器，依赖上下文学习（ICL）在不更新参数的情况下适应新分布。本文以高基数表格数据为受控测试案例，研究了在分布不匹配下ICL对结构化生成的局限性，发现一种结构性失效模式—— extit{类别先验锁定}：ICL无法更新模型从预训练中继承的词元分布先验。在两个7B参数的开源模型上，ICL虽能提升数值精度，但在类别分布上表现出急剧的上限，完全无法再现罕见类别。参数高效微调（LoRA）可克服此限制，但引入明显的记忆风险，且在某些情况下会破坏结构化输出的稳定性，揭示了适应性与隐私之间的根本权衡。
原文摘要 · Abstract (English)
Large language models (LLMs) are increasingly used as conditional generators for structured data, relying on in-context learning (ICL) to adapt to new distributions without parameter updates. We investigate the limits of ICL for structured generation under distribution mismatch, using high-cardinality tabular data as a controlled test case, and identify a structural failure mode we term \textit{categorical prior lock-in}: the inability of ICL to update the model's prior over token distributions inherited from pre-training. Across two 7B-parameter open-weight models, ICL improves numerical fidelity with additional examples but exhibits a sharp ceiling on categorical distributions, failing to reproduce rare classes entirely. Parameter-efficient fine-tuning (LoRA) overcomes these limitations but introduces measurable memorization risk and, in some cases, destabilizes structured output generation, highlighting a fundamental trade-off between adaptability and privacy.
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。