Proseminar Wintersemester 2023/2024
"Trustworthyness, Interpretability and Fairness in Machine Learning"
Description
With the upcoming of the world's first comprehensive AI law in Europe and rising questions on the use of AI in safety-critical contexts, it is evident the necessity of studying the social impact of new technologies in applications. Fairness, Trustworthiness, and Interpretability have become popular terms in the Computer Science society, and works involving them are becoming increasingly broad. Fairness, referring to the property of methods of outputting results not favoring particular categories despite other under-represented communities, looks to quantify whether the presence of biased training data leads to unfair prediction in the regards of a minority; practical, real-world applications can be found in rankings of candidates in the HR departments, or decisions in courts and loans among others. Interpretability refers to our understanding of predictions or general outputs of machine learning models; models that are not interpretable are referred to as black-box models, and the research focuses on providing post-hoc interpretation methods for them. The interpretation should allow broader use of the technologies in critical contexts (health, security, privacy) and also allow for easier debugging of the methods through underlining, which are the mechanisms that brought particular outcomes. Finally, trustworthiness refers to the ability of the model to be trusted, for example, with respect to robustness to attacks.
This seminar will touch on the hot topics of machine learning, the intersection of computer science, and its psychological and social impact. You will learn more about the Fairness, Trustworthiness, and Interpretability of models and what this means from a practical point of view. Moreover, it will be an occasion to discuss the importance of such themes in machine learning and the communication of discoveries and methods. The seminar will be held exclusively in English.
Literature
- tba -
Organization and Enrollment Procedure
- tba -
Goals and Criteria for a successful seminar
In the Pro-Seminar, you will learn how to work yourself into a topic, research related literature, and answer questions to this topic. To this end, you need to read the subchapters assigned to you and use different sources to verify and extend the statements made. Also, by listening and engaging with the other presentations, you will get a comprehensive understanding of interpretable machine learning methods. You will learn to address constructive criticism on research topics. To train this, we will assign to you two students' reports to critically comment until the end of the semester. You need to participate in every part of this seminar to pass it.