| 課程編碼 Course Code | 中文課程名稱 Course Name (Chinese) | 英文課程名稱 Course Name (English) | 總學分數 Credits | 總時數 Hours |
|---|---|---|---|---|
| 3603119 | 智慧醫療大語言模型應用 | Large Language Model Applications for Smart Healthcare | 3.0 | 3 |
| 中文概述 Chinese Description | 本課程以可落地的醫療情境為核心，培養學生運用大型語言模型（LLM）建置智慧醫療輔助系統之能力，並以可重複、可量化的實驗流程強化工程實作與可靠性思維。課程從醫療任務定義與提示工程實驗出發，帶領學生將醫療資料轉譯為明確的模型規格，並透過對照測試比較不同提示策略在正確性、一致性與可讀性上的差異。接著進入醫療文本資料處理與去識別實作，涵蓋資料清理、敏感欄位遮罩與資料治理，確保資料使用符合隱私與合規要求。核心技術部分聚焦檢索增強生成(RAG)，引導學生完成文件切分、向量化索引、檢索參數調校與引用證據回覆，建立可追溯、可驗證的回答機制以降低幻覺風險。此外本課程亦強調可靠性評估與安全防護，包含小型測試集建置、錯誤類型歸因、prompt injection測試與拒答策略設計，建立可維運的風險控管能力。最後課程將以6小時期末專題進行實作展示與成果發表，要求學生交付可操作之實驗與技術報告，呈現端到端系統整合能力與醫療應用的實務價值。 | |||
| 英文概述 English Description | This course focuses on deployable healthcare use cases and trains students to build smart healthcare assistants with large language models (LLMs) through reproducible, measurable experiments. Students start with task definition and prompt-engineering labs, translating healthcare data into clear model input/output specifications and running controlled comparisons for correctness, consistency, and readability. The course then covers healthcare text processing and de-identification (data cleaning, sensitive-field masking, and governance) to meet privacy and compliance needs. Core modules emphasize Retrieval-Augmented Generation (RAG), including document chunking, embedding-based indexing, retrieval tuning, and evidence-cited answers to improve traceability and reduce hallucinations. Reliability and security are addressed via compact test sets, error attribution, prompt-injection testing, and refusal-policy design. The course ends with a 6-hour capstone demo and presentation, requiring exe | |||
備註：