Proseminar Wintersemester 2020/2021

Machine Learning

Prof. Dr. Katharina Morik
Informatik LS8

Trustworthy Machine Learning

Machine learning (ML) is a driving force for many successful applications in Artificial Intelligence. Machine learning is not just a class of algorithms but there are long and nested sequences of algorithms and, at the same time, algorithms are built upon other algorithms from libraries or tools. This makes it hard for users to understand machine learning models.
One approach to allowing an understanding of machine learning is to explain the learned model. It has been pointed out that also the data which are used for training the model need a careful inspection. 8 of the selected papers for this seminar deal with explainability. Among them are, of course, those that explain deep neural networks

We want to include no only pure computer science views, but also look at what psychology, sociology and ethics have to say about AI and ML, in particular. A view of the broad area of ethics and bias is represented by 3 carefully selected papers.

Deep neural networks are particularly challenging our understanding. Their function approximation capabilities are so huge that they are hard to understand and control. A comprehensive sound paper on the trustworthiness of deep neural networks is split into three parts, so that three students could study this work. All parts only need the introduction as an additional read, not the other parts.

The impact of learned models on staff selection, sales, and popularity is tremendous. The bias in data lead to a bias in real-world life. Actually, biased data may do harm to the health and success of people. Hence, the community of researchers in machine learning and other disciplines discuss how to establish fairness. 12 papers cover diverse aspects of fairness.

Date: Tuesdays, 14:15 - 16:00 h, online

Moodle Workspace

Topics, literature (excluding books) and dates: