可信任人工智慧：統計機器學習與最佳化

This course centers on Trustworthy Artificial Intelligence (TAI) and integrates methods from statistical machine learning and optimization to examine challenges faced by AI systems in real-world applications, including interpretability, uncertainty, reliability, and decision quality. The course guides students to understand, from the perspectives of data, models, decision-making, and deployment, that AI models should not only pursue predictive accuracy but also be understandable, verifiable, correctable, and capable of supporting robust decisions in complex and changing environments. Topics include foundational concepts of trustworthy AI, explainable AI, uncertainty quantification, and data-driven decision-making and optimization. Through homework assignments, paper reading, oral presentations, and a final project, students will learn to analyze limitations of AI systems, select and design appropriate statistical machine learning and optimization methods, and translate analyses into practically meaningful or research-relevant models, evidence, and decision recommendations. After completing this course, students will be able to: - Understand core concepts of trustworthy AI, statistical machine learning, and optimization, and apply them appropriately to data analysis, model evaluation, and decision-support problems. - Analyze trustworthy AI issues such as interpretability, uncertainty, and reliability, and evaluate their impacts on model usage and decision quality. - Develop the ability to read academic papers, structure problems, and communicate research ideas clearly on trustworthy-AI-related topics. - Integrate statistical machine learning, optimization, and application contexts to design new analytical methods or validation procedures, and present effectiveness and limitations in a project format. Course learning goals include: - Critical thinking: Understand the systemic nature of trustworthy AI problems, application contexts, and sources of risk. - Research skills: Learn how to search for, read, organize, and critique trustworthy AI literature. - Integrative skills: Design new methods and validation pipelines, and clearly explain advantages, limitations, and applicable scenarios. Course Scope: This course does not aim to comprehensively cover all modern AI topics, such as LLMs, computer vision, VLMs, reinforcement learning, or agentic AI. Instead, it mainly approaches trustworthy AI from the perspectives of statistical machine learning, uncertainty quantification, explainability, reliability, and optimization-based decision-making. However, final project topics are open-ended and may involve any AI-related application, as long as they address trustworthy AI issues. 本課程以可信任人工智慧為核心，結合統計機器學習與最佳化方法，探討人工智慧系統在真實應用場域中所面臨的可解釋性、不確定性、可靠性、與決策品質等議題。課程引導學生從資料、模型、決策與系統部署的角度，理解 AI 模型不只是追求預測準確率，也必須能被理解、被檢驗、被修正，並在複雜且變動的環境中支援穩健決策。 課程內容涵蓋可信任 AI 的基本概念、可解釋人工智慧、不確定性量化、資料驅動決策與最佳化等主題。學生將透過作業、論文閱讀、口頭報告與專題實作，學習如何分析 AI 系統的限制，選擇與設計合適的統計機器學習與最佳化方法，並將分析結果轉化為具實務意義或研究價值的模型、證據與決策建議。 修習本課程後，學生將能夠： - 掌握可信任 AI、統計機器學習與最佳化的基本概念，並能正確應用於資料分析、模型評估與決策支援問題。 - 理解可解釋性、不確定性、分佈變動與模型更新等可信任 AI 議題，並能評估其對模型使用與決策品質的影響。 - 培養學術論文閱讀、問題整理與研究表達能力，能針對可信任 AI 相關主題提出清楚的研究觀點。 - 整合統計機器學習、最佳化與應用情境，設計新的分析方法或驗證流程，並以專題形式呈現其有效性與限制。 本課程的學習目標，包含： - 思辨能力：理解可信任 AI 問題的系統性、應用脈絡與風險來源。 - 研究能力：學習如何搜尋、閱讀、整理與評論可信任 AI 相關論文。 - 整合型能力：設計新方法、建立驗證流程，並能清楚說明方法的優點、限制與適用情境。 課程範圍： 本課程並非旨在完整涵蓋所有現代人工智慧主題，例如大型語言模型、電腦視覺、視覺語言模型、強化學習或 Agentic AI。相較之下，本課程主要從統計機器學習、不確定性量化、可解釋性、可靠性，以及以最佳化為基礎的決策制定等角度切入可信任人工智慧。期末專題主題採開放形式，可涵蓋任何 AI 相關應用；但專題內容需回應可信任人工智慧相關議題。

- It is highly recommended that students have taken courses related to "Probability & Statistics" and "Machine Learning". Background in "Operations Research" is helpful. - Students without a background in probability, statistics, or machine learning may need substantial online coursework or self-study to keep up with the course. - Students are expected to be able to program in Python and have experience with LaTeX. - This is a highly research project-oriented course. - 建議修習過「機率與統計」與「機器學習」相關課程為佳；具備「作業研究」基礎則有助於課程學習。 - 若沒有機率、統計或機器學習背景，可能需要投入相當程度的線上課程或自主學習，才能跟上本課程內容。 - 修課學生需具備 Python 的程式能力，並具有 LaTeX 使用經驗。 - 本課程是高度研究專題導向（research project-oriented）的課程。

無備註

教師未提供此項資料

Homework (30%) Midterm Paper Presentation (10%) Midterm Project Proposal Presentation (10%) Final Project (40%) Participation (10%) 作業（30%） 期中論文報告（10%） 期中提案報告（10%） 期末專題（40%） 課程參與（10%）

教師未提供此項資料

Statistical Learning and Trustworthy AI Foundations - Barocas, S., Hardt, M., and Narayanan, A. (2023), Fairness and Machine Learning: Limitations and Opportunities, MIT Press. - Hastie, T., Tibshirani, R., and Friedman, J. (2009), The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd ed., Springer. Explainable AI (XAI) - Explanatory Model Analysis (https://ema.drwhy.ai/) - Interpretable Machine Learning. A Guide for Making Black Box Models Explainable (https://christophm.github.io/interpretable-ml-book/) Uncertainty Quantification (UQ) - Uncertainty Quantification Dictionary (https://dictionary.helmholtz-uq.de/content/landing_page.html) - Abdar, M., Pourpanah, F., Hussain, S., Rezazadegan, D., Liu, L., Ghavamzadeh, M., Fieguth, P., Cao, X., Khosravi, A., Acharya, U. R., et al. (2021), "A review of uncertainty quantification in deep learning: Techniques, applications and challenges," Information Fusion, 76, 243--297. - Gawlikowski, J., Tassi, C. R. N., Ali, M., Lee, J., Humt, M., Feng, J., Kruspe, A., Triebel, R., Jung, P., Roscher, R., et al. (2023), "A survey of uncertainty in deep neural networks," Artificial Intelligence Review, 56, 1513--1589. - Smith, R. C. (2024), Uncertainty Quantification: Theory, Implementation, and Applications, SIAM. Decision-Making and Optimization Methods - Kochenderfer, M. J. (2015), Decision Making Under Uncertainty: Theory and Application, MIT Press. - Bertsimas, D., and Kallus, N. (2020), "From predictive to prescriptive analytics," Management Science, 66(3), 1025--1044. - Elmachtoub, A. N., and Grigas, P. (2022), "Smart predict, then optimize," Management Science, 68(1), 9--26.

地點

Room 419, 4F, Assembly Building I 綜合一館 419

時間

Please make an appointment with the professor. 請跟老師預約時間

聯絡方式

Email: yuhsinhung@nycu.edu.tw