Type of Credit: Elective
Credit(s)
Number of Students
Advanced machine learning mechanisms such as deep learning neural networks have rapidly automated decision processes in government, business, finance, and healthcare domains. 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, focusing 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.
Note: This course expects students to have experience with Python programming and Google Colab. The curriculum and assignments will extensively draw upon mathematical concepts, algorithmic principles, and logical reasoning. Prospective students should ensure their comfort and familiarity with these areas prior to enrollment.
能力項目說明
This course offers a comprehensive exploration of select topics in AI fairness and explainability, with a particular emphasis on formal methodologies. The technical subjects we will delve into encompass formal fairness, heuristic and formal explanations, 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.
Upon successful completion of this course, students are expected to:
| 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 |
課程內容與指定閱讀 Content and Reading Assignment |
學習投入時間 Student workload expectation |
|
|
課堂講授 In-class Hours |
課程前後 Outside-of-class Hours |
|||
|
1 |
Overview |
Data science fundamentals |
3 |
6 |
|
2 |
Fair AI |
(Mid-Autumn Festival) |
3 |
6 |
|
3 |
Individual fairness and FTU |
3 |
6 |
|
|
4 |
Statistical and group fairness |
3 |
6 |
|
|
5 |
Case studies of fair machine learning |
3 |
6 |
|
|
6 |
Measurement of discrimination |
3 |
6 |
|
|
7 |
Pre- / in- / post-processing of fair machine learning |
3 |
6 |
|
|
8 |
Causal analysis and counterfactual fairness |
3 |
6 |
|
|
9 |
Explainable AI |
Interpretable machine learning models |
3 |
6 |
|
10 |
Blackbox explanation methods |
3 |
6 |
|
|
11 |
Whitebox explanation methods |
3 |
6 |
|
|
12 |
Neural network explanations |
3 |
6 |
|
|
13 |
NLP and attention-based explanations |
3 |
6 |
|
|
14 |
Propositional and first-order logic |
3 |
6 |
|
|
15 |
Abductive and contrastive explanations |
3 |
6 |
|
|
16 |
Logic-based approaches to XAI |
3 |
6 |
|
|
17 |
Final Project |
Self-study week |
3 |
6 |
|
18 |
Self-study week |
3 |
6 |
|
In-class Exercises 10%
Literature presentation 25%
Paper presentation 25%
Homework Assignments 40%
Bonus tasks ~10%
We encourage active engagement and participation in every meeting as an integral part of the learning process. Your contributions during lectures, such as asking questions, responding to in-class brain teasers or pop quizzes, and sharing your thoughts and insights, are highly valued. These contributions will directly impact your meeting participation grades (10 points), reflecting the level of your active involvement.
The quality and frequency of your contributions will be taken into account in the grading scheme. Your contributions to Lecture Participation will be evaluated after every meeting. Please note that contributions are more than just attending or talking in meetings. The quality of your contributions and your responses to others will be essential to the evaluation.
When doing homework assignments and writing case study reports, students must comply with the plagiarism regulations imposed by the university. Any piece of the work submitted by a student must be their own work. Copying or paraphrasing another person's work in their submission without explicit acknowledgment will be considered plagiarism, leading to the loss of all points for your assignment or report.
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 Decision‐Making - 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
| 書名 Book Title | 作者 Author | 出版年 Publish Year | 出版者 Publisher | ISBN | 館藏來源* | 備註 Note |
|---|