Fairness, Explainability, and Accountability for Machine Learning

Course Description

Instructor: Hoda Heidari
Office hours: Tuesdays 11:40 AM - 1:00 PM and by appointment
Lectures: Thursdays 11:40 AM - 1:00 PM

Overview. Machine Learning (ML) tools increasingly make or inform high-stakes decisions that affect people's lives. Several recent studies and articles have demonstrated that if not done carefully, ML tools can disparately harm/benefit various individuals and social groups. It is, therefore, important for ML designers and practitioners to understand the ethical and societal-level implications of their choices and designs, so that they can contribute to reducing (and not unintentionally amplifying) existing/historical injustices through their work. This course aims to

Topics that will be covered include:

Prerequisite knowledge. Students are expected to have an introductory knowledge of ML. (For example, the student has taken an Introduction to Machine Learning course before).

Format and deliverables. The course will be entirely online, and lectures/presentations will be delivered via Zoom. In the first severn sessions of the course, I will give lectures to cover the basics of the research area. The remaining sessions are dedicated to student presentations on their projects. Project deliverables include a class presentation and paper (due at the end of the semester). The midterm exam will be take-home.


Important Announcements and Statements

Wellness This semester is unlike any other. We are all under a lot of stress and I know that attending Zoom classes and meetings all day can become overwhelming. Please make sure to take frequent breaks and move regularly, get enough sleep, eat well, and reach out to your support system or me if you need to. In addition, please take advantage of the many excellent resources that the university offers to support your overall health and wellness during these challenging times (see, for example, Counseling and Psychological Services).

Diversity. As the title of the course suggests, one of our key goals will be to promote fairness, diversity, and inclusion. We all come from different backgrounds and life circumstances and our diverse paths shape the perspectives we bring to the classroom. This diversity is extremely valuable as we engage with difficult topics such as unfairness and discrimination through the use of ML. While I expect there to be rigorous class discussion, I ask that you engage in discussion with care and empathy for the other members in the classroom. This is particularly important in handling disagreements. Remember: every one of us has a role to play in creating a more inclusive environment.

If any of our class meetings conflict with your religious events, please let me know so that we can make alternative arrangements for you. If you have a disability, I encourage you to discuss your accommodations and needs with me as early in the semester as possible. I will work with you to ensure that accommodations are provided as appropriate.

Technology. This semester involves the regular use of technology during class. Research has shown that divided attention is detrimental to learning, so I encourage you to close any windows not directly related to what we are doing while you are in class. Please turn off your phone notifications and limit other likely sources of technology disruption, so you can fully engage with the material, each other, and me. This will create a better learning environment for everyone.

Textbook. We will not be using any particular textbook. I will make sure the references cited in the course are uploaded (or made accessible otherwise) in a timely fashion. If you encounter any problem accessing any of the course-related resources, please let me know.

Class presence and participation. Class presence and participation points are given to encourage your active class participation and discussion. You will be rewarded with a perfect score as long as you frequently come to class and actively contribute to the class discussion during recitations and lectures.

Academic integrity. For a clear description of what counts as plagiarism, cheating, and/or the use of unauthorized sources, please see the University's Policy on Academic Integrity. In particular, the policy states that assistance from campus resources (Academic Development, the Global Communication Center, and the Academic Resource Center at CMU-Q) is permitted, but no collaboration is allowed unless specifically permitted by a course instructor.

Late/Make-up work policy. Due dates for the exam, your class presentation, and the written project reports will be announced on this website. If you experience circumstances that prohibit you from submitting your work on time, please let me know as soon as you can. I will evaluate these instances on a case-by-case basis.


Schedule

My lectures (the first 7 sessions of the course) will cover the following topics:
  1. Introduction (Sep 3)
  2. Sources of Unfairness (Sep 10)
  3. Statistical Notions of Fairness (Sep 10, 17)
  4. Individual Notions of Fairness (Sep 17)
  5. Guest lecture (Sep 24)
  6. Fairness Mechanisms (Oct 1)
  7. Data Pre-processing Methods (Oct 1)
  8. Tradeoffs and Impossibilities (Oct 8)
  9. Counterfactual Fairness (Oct 8)
  10. Software Tools for Fairness (Oct 15)
  11. Long-term Algorithmic Impact (Oct 15)
  12. Explainability (global methods) (Oct 22)
  13. Explainability (local methods) (Oct 22)
  14. Governance and Accountability (Oct 29)
  15. Wrap-up and Midterm Prep (Nov 5)
  16. Student Presentations (Nov 12--end of semester)

Projects

Information about course projects will be announced here.

Each team will be assigned a topic plus 1-2 research papers on the topic. The team is expected to read, understand, and reproduce the results in the main papers, write a critical survey of the papers as well as other relevant articles on the topic, propose directions for future work, and partially explore some of these directions.


Performance Assessment

The final grade will be computed based on:
  • 30% written mid-term exam
  • 60% project (40% written report + 20% class presentation)
  • 10% participation in class discussions