Official course description:
Full info last published 20/06-22

Algorithmic Fairness, Accountability and Ethics

Course info
Language:
English
ECTS points:
7.5
Course code:
KSALFAE1KU
Participants max:
40
Offered to guest students:
no
Offered to exchange students:
no
Offered as a single subject:
no
Programme
Level:
MSc. Master
Programme:
MSc in Data Science
Staff
Course manager
Associate Professor
Teacher
Associate Professor
Teacher
Associate Professor
Teacher
Assistant Professor
Course semester
Semester
Forår 2022
Start
31 January 2022
End
31 August 2022
Exam
Exam type
ordinær
Internal/External
intern censur
Grade Scale
7-trinsskala
Exam Language
GB
Abstract

This course provides an overview on the topics of bias and fairness in data, models, and algorithms, and on the associated ethical and accountability issues.

Description

Whether to provide predictions, information, assessments, or evidence of some sort, the use of data-driven technologies comes with important responsibilities. In this course, you will learn to recognize and quantify how bias can affect data and algorithms, to apply methods for interpreting algorithmic predictions, to mitigate or manage the effects of bias and improve fairness in a number of applications, and understand why ethical issues can arise when applying data science to real world problems. Ultimately, the class will teach you how to reason through these problems in a systematic manner and how to justify and defend your approach to dealing with them.

The course will cover the following subjects:

  • Bias Sources
  • Fairness Metrics
  • Explainability and Interpretability of Models
  • Debiasing Data and Models
  • Robustness of models
  • Limits of Predictability
  • Algorithms Auditing
  • Old and new philosophical paradigms in AI
  • General ethics - difference between fairness and justice
  • Selected topics from philosophy of science and ethical systems for bias management


Formal prerequisites

  • The prerequisite required for admission to the course is knowledge of introductory machine learning.
  • Students must be able to program. The default language is Python.


Intended learning outcomes

After the course, the student should be able to:

  • Identify potential sources of bias and quantify fairness in data and applications
  • Discuss strategies to mitigate or manage bias
  • Analyse the robustness and caveats of machine learning models and predictions
  • Apply tools to interpret/explain the predictions of a ML algorithm
  • Reflect on the philosophical and ethical implications of data science applications to real world problems
Learning activities

The course consists of lectures and exercises. Beyond lectures and exercise sessions, we will have various online activities to be done before and after classes. These online activities will include: readings and videos, discussion on forums, writing documents, and coding exercises. During the class, we will have group discussions, class discussions, quizzes, writing sessions, and various hands-on exercises based on the activities done at home.



Mandatory activities

During the course, there will be three mandatory assignments, which focus on specific parts of the final exam project. These mandatory assignments are checkpoints on the status of the project and opportunities to receive feedback during the course. Students that miss the deadline for the mandatory activities during the course are required to submit the assignment together with the final exam project.

The student will receive the grade NA (not approved) at the ordinary exam, if the mandatory activities are not approved and the student will use an exam attempt.

Course literature

  • More literature is published in the course page in LearnIT.


Student Activity Budget
Estimated distribution of learning activities for the typical student
  • Preparation for lectures and exercises: 15%
  • Lectures: 25%
  • Exercises: 20%
  • Assignments: 10%
  • Project work, supervision included: 30%
Ordinary exam
Exam type:
C: Submission of written work, Internal (7-point scale)
Exam variation:
C1G: Submission of written work for groups
Exam submisson description:
Group paper (max 10 pages) and code. Groups choose a relevant dataset and algorithm, formulate relevant questions relating to fairness, apply concepts and tools presented during the course, and discuss ethical implications with references to the arguments covered during the course.
All group members are expected to work together on the submission and each group member should invest roughly the same time. Every group is asked to submit a contribution statement to report (un)equal contributions. Individual grading will take these numbers into account.

Group submission:
Group
  • 2-3


reexam
Exam type:
C: Submission of written work, Internal (7-point scale)
Exam variation:
C1G: Submission of written work for groups
Group submission:
Group
  • 2-3

Time and date
Ordinary Exam - submission Fri, 27 May 2022, 08:00 - 14:00
Reexam - submission Wed, 27 Jul 2022, 08:00 - 14:00