本課程以「智慧型代理人」為主軸，前半學期涵蓋搜尋與問題求解、對抗搜尋與賽局、限制滿足問題、邏輯與知識表示、機率與貝氏網路、馬可夫模型與 HMM；後半學期涵蓋效用理論與決策網路、機器學習與類神經網路、馬可夫決策過程、強化學習、大語言模型，第 13 至 15 週聚焦大語言模型與 Agentic AI：LLM 代理人架構、工具使用與協定 (MCP、A2A)、多代理人系統、代理人評估與 Agentic AI 安全。課程採期末專題制，學員將實作一個會呼叫工具的 AI 代理人，解決自身工作場域的真實問題。
【115 學年度第 1 學期行事曆安排】
本課程自 2026/9/19 至 2027/1/16，共 18 週；實體課於每週六 16:10–19:05 在建 208 教室上課。
第 4 週（10/10）國慶日、第 11 週（11/28）選舉日停課，不到校上課；依原課綱分別進行限制滿足問題 (CSP)、馬可夫決策過程 (MDP) 的線上自主學習。
第 7 週（10/31）改為線上繳交期中簡報，不到校上課；報告範圍為第 1 至 6 週，不另安排期中口頭報告。第 8 週（11/7）面授貝氏網路抽樣、馬可夫模型與隱藏馬可夫模型 (HMM)；第 9 週（11/14）面授效用理論、決策網路與資訊價值。
第 10 週（11/21）面授機器學習與類神經網路，並繳交期末專題提案；第 14 週（12/19）為 LLM 代理人上機實作。
第 16 週（2027/1/2）辦理期末報告與專題發表，對應學校期末考試週 12/28–1/3，以發表與實作展示代替期末筆試。
第 17、18 週（1/9、1/16）為彈性教學週，不到校上課；分別完成期末專題程式最終版、書面報告及學習心得，書面報告於 1/16 截止。
各週主題與教學目標詳見 iCAN「課程進度」。
《 課程簡介 -- English 》
This course centered on the concept of intelligent agents. The first half covers problem solving and search, adversarial search and games, constraint satisfaction problems, logic and knowledge representation, probability and Bayesian networks, and Markov models and HMMs. The second half covers utility theory and decision networks, machine learning and neural networks, Markov decision processes, reinforcement learning, and large language models, with Weeks 13–15 focused on large language models and agentic AI: LLM agent architectures, tool use and protocols (MCP, A2A), multi-agent systems, agent evaluation, and agentic AI safety. The course concludes with a term project in which students build a tool-using AI agent for a real problem in their own workplace.
Academic calendar for Fall 2026 (Academic Year 115, Semester 1):
The course runs from September 19, 2026 to January 16, 2027. Regular classes meet on Saturdays, 16:10–19:05, in Jian 208.
There are no on-campus classes on October 10 (National Day, Week 4) or November 28 (election day, Week 11). Students follow the existing syllabus for asynchronous independent study of CSP and Markov decision processes (MDP), respectively.
Midterm slides covering Weeks 1–6 are due online on October 31 (Week 7), with no on-campus class or separate oral midterm presentation. Bayesian network sampling, Markov models and HMMs are taught on campus on November 7 (Week 8). Utility theory, decision networks and the value of perfect information are taught on campus on November 14 (Week 9).
Machine learning and neural networks are taught on campus on November 21 (Week 10), when project proposals are also due, and the LLM agent lab is on December 19 (Week 14).
The final presentation and demonstration take place on January 2, 2027 (Week 16), within the university final assessment week of December 28–January 3, in place of a written final exam.
January 9 and 16 (Weeks 17–18) are flexible teaching weeks with no on-campus class: finalize the project code in Week 17 and submit the written report and learning reflection by January 16. See the iCAN course schedule for weekly details.
|