纠正上下文学习中的演示捷径
Rectifying Demonstration Shortcut in In-Context Learning
AI总结：
针对大语言模型上下文学习中依赖演示语义先验的“演示捷径”问题，提出演示感知的上下文校准方法，在两种任务设置、三类大模型上均实现显著效果提升。
摘要：
大语言模型（LLM）能够借助其上下文学习（ICL）能力，仅通过少量演示即可解决各类任务。然而，LLM在进行ICL预测时，往往依赖其预训练得到的演示语义先验，而非输入-标签关系。在本研究中，我们将这一现象命名为“演示捷径（Demonstration Shortcut）”。\n以往研究主要聚焦于提升预定义任务的ICL预测效果，而本研究旨在纠正演示捷径，从而使LLM能够从演示中有效学习新的输入-标签关系。为此，我们提出了上下文校准（In-Context Calibration），这是一种具备演示感知能力的校准方法。\n我们在两种设置下评估了所提方法的有效性：（1）采用标准标签空间的原始ICL任务；（2）标签空间被替换为语义无关token的任务学习设置。在两种设置下，上下文校准均展现出显著提升，其效果在三种LLM系列（OPT、GPT和Llama2）的多种配置下均具备通用性。
英文摘要：
Large language models (LLMs) are able to solve various tasks with only a few demonstrations utilizing their in-context learning (ICL) abilities. However, LLMs often rely on their pre-trained semantic priors of demonstrations rather than on the input-label relationships to proceed with ICL prediction. In this work, we term this phenomenon as the 'Demonstration Shortcut'. While previous works have primarily focused on improving ICL prediction results for predefined tasks, we aim to rectify the Demonstration Shortcut, thereby enabling the LLM to effectively learn new input-label relationships from demonstrations. To achieve this, we introduce In-Context Calibration, a demonstration-aware calibration method. We evaluate the effectiveness of the proposed method in two settings: (1) the Original ICL Task using the standard label space and (2) the Task Learning setting, where the label space is replaced with semantically unrelated tokens. In both settings, In-Context Calibration demonstrates substantial improvements, with results generalized across three LLM families (OPT, GPT, and Llama2) under various configurations.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
