教學大綱 Syllabus

科目名稱:進階資訊系統研發

Course Name: Advanced Information System Development

修別:選

Type of Credit: Elective

3.0

學分數

Credit(s)

25

預收人數

Number of Students

課程資料Course Details

課程簡介Course Description

Advanced machine learning mechanisms such as deep learning neural networks have rapidly automated decision processes in domains including government, business, finance, and healthcare. Unlike traditional rule-based AI approaches, decisions made by a modern AI system are often built upon numerous iterations over input data, making the system behave like a black box to humans. Therefore, Trustworthy AI research and practice aim to open the black box of highly complex AI systems and inspect justification for the cause and effect within their decision processes.

This course will introduce the recent developments of Trustworthy AI. We will cover topics in AI fairness, explainability, and safety, with a focus on approaches providing provable correctness and quality guarantee. The course will consist of lectures, tutorials, and paper presentations. The teacher will give lectures and tutorials on selected topics in a self-contained manner. The students will form into groups: each group will be responsible for giving one lecture, one tutorial, and one paper presentation. After taking this course, students will gain a general knowledge of Trustworthy AI, as well as a deep understanding of specific techniques for practicing and researching Formal AI fairness, explainability, and safety.

核心能力分析圖 Core Competence Analysis Chart

能力項目說明


    課程目標與學習成效Course Objectives & Learning Outcomes

    This course provides an overview of selected topics in AI fairness and explainability, focusing on the formal approaches. Specific topics that will be covered include machine learning fairness, formal explanations, logical query languages, property inference, robustness analysis, and constraint solving. This course is research-oriented and will equip the students with an algorithmic toolkit for their further study in related and more advanced topics.

    每周課程進度與作業要求 Course Schedule & Requirements

    教學週次Course Week 彈性補充教學週次Flexible Supplemental Instruction Week 彈性補充教學類別Flexible Supplemental Instruction Type

    週次

    Week

    課程主題

    Topic

    課程內容與閱讀資料

    Content and Reading Materials

    課程活動

    Teaching Activities

    學習投入時間

    課堂講授

    課程前後

    1

    Course
    Overview

    Algorithmic bias in data science

    Kick-off Meeting

    3

    5

    2

    Holiday

    3

    Fairness

    Fairness metrics: individual fairness, group fairness, local fairness, and global fairness.

    Fair model training: pre-processing, in-processing, and post-processing

    Structural equation modeling and counterfactual fairness

    Name Form and
    Group List Due

    3

    5

    4

    Lecture and
    Group Discussion

    3

    5

    5

    Lecture and
    Group Discussion

    3

    5

    6

    Lecture and
    Group Discussion

    3

    5

    7

    Lecture and
    Group Discussion

    3

    5

    8

    Formal XAI

    First-order logic

    Logical approaches to XAI

    Abductive explanations

    Contrastive explanations

    Transformer programs

    Property inference

    Lecture and
    Group Discussion

    3

    5

    9

    Lecture and
    Group Discussion

    3

    5

    10

    Lecture and
    Group Discussion

    3

    5

    11

    Lecture and
    Group Discussion

    3

    5

    12

    Lecture and
    Group Discussion

    3

    5

    13

    XAI

    Interpretable models

    Global model-agnostic methods

    Local model-agnostic methods

    Neural network interpretations

    Lecture and
    Group Discussion

    3

    5

    14

    Lecture and
    Group Discussion

    3

    5

    15

    Lecture and
    Group Discussion

    3

    5

    16

    Lecture and Group Discussion

    3

    5

    17

    Lecture and
    Group Discussion

    3

    5

    18

    TBD

    Reserved for unfinished lectures

    TBD

    3

    5

    授課方式Teaching Approach

    50%

    講述 Lecture

    40%

    討論 Discussion

    10%

    小組活動 Group activity

    0%

    數位學習 E-learning

    0%

    其他: Others:

    評量工具與策略、評分標準成效Evaluation Criteria

    GRADE DISTRIBUTION

    Literature presentation 30%
    Paper presentation 30%
    Assignments 30%
    Participation 10%
    Bonus exercises (~10%)

    CONTRIBUTION EVALUATION

    You are encouraged to attend each meeting with the assigned readings prepared in advance. When you attend a meeting, you must be on time and remain there for the entire meeting. A three-hour meeting will often consist of two hours of the lecturer presenting technical papers, followed by one hour of group discussion on a selected article. Your grades for "Lecture participation" will be determined by how actively you participate in the lecturer's presentation, for example, by answering questions from the lecturer, making comments, and initiating further discussions. The group discussion session will be led by the lecturer in the first few weeks and by the students afterward. Each group must take charge of at least one group discussion session. Your grades for "Group discussion participation" will be determined by how your group presents the selected article and leads the discussion (when your group is in charge), and how your group contributes to the discussion (when your group is an audience) during the group discussion session.

    You will be evaluated after every meeting based on the following criteria. Please note that contributions are not equivalent to merely attending a meeting and talking. The quality of your comments and responses will also be an important component of the evaluation.

    Excellent Participation (A) : (1) regularly initiates and contributes to meeting discussions; (2) regularly indicates substantial knowledge and insights; (3) frequently facilitates other students in clarifying and developing their viewpoints. (4) frequently contributes to meeting discussions by helping others produce a synergistic understanding of the issues being discussed.

    Good Participation (B) : (1) frequently initiates and contributes to meeting discussions; (2) occasionally indicates substantial knowledge and insights; (3) occasionally facilitates others in clarifying and developing their viewpoints.

    Fair Participation (C) : (1) occasionally initiates meeting discussions; (2) occasionally contributes to meeting discussions; (3) indicates some knowledge and insights; (4) rarely responds constructively to the contributions from other students.

    Poor Participation (D/E) : (1) never or rarely initiates meeting discussions; (2) never or rarely contributes to meeting discussions; (3) actively inhibits or impedes the course of discussion; (4) exhibits defensive behavior such as aggression or withdrawal rather than being thoughtful and considerate of others' ideas.

    指定/參考書目Textbook & References

    Paper Reading List

    Algorithmic bias - Sina Fazelpour and David Danks
    https://compass.onlinelibrary.wiley.com/doi/10.1111/phc3.12760?af=R

    Are Algorithms Value-Free? - Gabrielle M. Johnson
    https://www.gmjohnson.com/uploads/5/2/5/1/52514005/are_algorithms_value_free_.pdf 

    Algorithmic injustice - Abeba Birhane
    https://www.sciencedirect.com/science/article/pii/S2666389921000155

    Data Owning Democracy or Digital Socialism? - James Muldoon
    https://www.tandfonline.com/doi/full/10.1080/13698230.2022.2120737

    Stop Explaining Black-Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead - Cynthia Rudin
    https://arxiv.org/pdf/1811.10154.pdf

    The Bias Dilemma - Oisín Deery and Katherine Bailey
    https://ojs.lib.uwo.ca/index.php/fpq/article/view/14292

    On the Advantages of Distinguishing Between Predictive and Allocative Fairness in Algorithmic DecisionMaking - Fabian Beigang
    https://link.springer.com/article/10.1007/s11023-022-09615-9 

    Transparency in Complex Computational Systems - Kathleen Creel
    https://www.cambridge.org/core/journals/philosophy-of-science/article/transparency-in-complex-computational-systems/4DB040EB28172CADF5F2858B62D0952C  

    Algorithmic and Human Decision Making - Mario Günther and Atoosa Kasirzadeh
    https://www.mario-guenther.com/_files/ugd/70b9dd_ff087ae509034fb9b126dcf783182457.pdf

    A modern Pascal's wager for mass electronic surveillance. - David Danks
    https://static1.squarespace.com/static/5f6d0320212a261d8716949f/t/621319146907794d4dba3724/1645418773886/Telos-PascalsWager-Pub.pdf

    The Surveillance Society - Oscar H. Handy Jnr.
    https://academic.oup.com/joc/article-abstract/39/3/61/4210548

    Risk Imposition by Artificial Agents - Johanna Thoma
    https://johannathoma.files.wordpress.com/2021/02/moral-proxy-problem-feb-2021.pdf 

    已申請之圖書館指定參考書目 圖書館指定參考書查詢 |相關處理要點

    書名 Book Title 作者 Author 出版年 Publish Year 出版者 Publisher ISBN 館藏來源* 備註 Note

    維護智慧財產權,務必使用正版書籍。 Respect Copyright.

    課程相關連結Course Related Links

    
                

    課程附件Course Attachments

    課程進行中,使用智慧型手機、平板等隨身設備 To Use Smart Devices During the Class

    需經教師同意始得使用 Approval

    列印