Grounding (in XAI / RAG)

在可解释 AI / RAG 语境中，指将 AI 输出锚定在可验证的外部证据上（如检索到的文档、专家推理、临床指南），而非仅依赖模型的参数化知识。

English Definition

"In XAI and RAG contexts, the practice of anchoring AI outputs in verifiable external evidence (e.g., retrieved documents, expert reasoning, clinical guidelines) rather than allowing the model to rely solely on its parametric knowledge."

常见错误

·不要把它直译为'接地'或'地面'，在 XAI/RAG 里特指将生成内容锚定到可验证来源的设计动作。

真实用例 · 3 条

"their outputs require grounding in current medical evidence to ensure reliability."

论文Pipeline extracted — review needed·模型输出需要锚定在最新医学证据上以确保可靠性

"MAP-X translates clinical evaluation frames into patient-comprehensible structures, uses RAG to ground the content."

论文Pipeline extracted — review needed·MAP-X 使用 RAG 来锚定内容

"The RAG has emerged as a promising approach to mitigate these risks by grounding the model outputs in verifiable domain-specific information."

论文Pipeline extracted — review needed·RAG 通过将模型输出锚定于可验证领域知识来缓解风险

由 pipeline 自动采集，待人工 review