Ethical Machine Learning (2024)

The course empowers students to identify and mitigate negative impacts of data and machine learning tools on vulnerable groups, promoting transparency and ethical solutions. It involves data analysis, application of machine learning models, and understanding optimization problems, with an initial assessment and leveling classes to ensure the necessary foundation.

Terminal Objectives

Students should be able to identify the use of data and machine learning tools that harm individuals and vulnerable groups and use and propose tools that reduce the impact and increase the transparency of such data and tools.

Prerequisites

The following are desirable skills for the student, these are not mandatory, but will increase the probability of the student’s knowledge absorption.

• Being able to perform a descriptive analysis of data (basic statistics, data visualization).
• Being able to apply and describe characteristics of machine learning models (logistic regression, decision trees, and ensembles) in tabular data.
• Being able to apply deep learning models to data of diverse nature.
• Being able to understand and model optimization problems in the context of machine learning.

Contents

• Part 1: Data and models
  • Data: Data sources, modalities, biases, privacy, and consent. (1 class)
    • Objective: Students should be able to prepare data that is reliable and less biased, preserving user privacy and consent.
  • Basic modeling: Basic concepts of machine learning, supervised learning, causal models. (1 class)
    • Objective: Students should be able to choose and potentially create machine learning models that are more suitable for the prepared data.
• Part 2. Fairness
  • Fairness: Detection and promotion of fairness in machine learning, maintenance of reliability in scenarios under change or attack. (5 classes)
    • Objective: Students should be able to choose and potentially create/modify more reliable and impartial machine learning models.
• Part 3: Explainability:
  • Explainability: Interpretability in machine learning, and creation of intervention mechanisms in machine learning. (5 weeks)
    • Objective: Students should be able to detect and modify a machine learning that fails to empower decision makers to understand it and detect flaws in its reliability and impartiality.

References

• Main reference:
  • Trustworthy Machine Learning: The book addresses, in general, general concepts of fairness and transparency.
• Additional references:
  • Fairness and machine learning: A freely available book that was made jointly by three professors from different universities who teach Fairness or Ethics in ML courses. The website contains the pages of the respective courses.
  • Practical Fairness: Fairness book from O’Reilly that I added to the list just to let you know it’s not worth it. It uses different nomenclatures than those most used in the literature and does not present the mathematical part well.
• General resources:
  • Trustworthy ML Initiative: It has several Trustworthy ML resources, such as general interest books, textbooks, documentaries, and course websites.
• General disclosure:
  • Weapons of Math Destruction: A book that presents several examples of unethical algorithms.
  • The Ethical Algorithm [URL inválido removido]: The Science of Socially Aware Algorithm Design: A book that presents the main areas of ethics in ML (Fairness, Transparency, and Privacy) to the general public.
  • Coded Bias: A Netflix documentary that can serve as motivation for the course.