Natural Language Processing: Shortcuts and Biases in Artificial Intelligence

What it is about

You will learn how shortcuts and data thinking in NLP models lead to biased or unreliable results and which methods and metrics exist to detect and mitigate these problems. In the end you will be able to understand, assess and apply current techniques for analysis and debiasing of language models in your own projects.

What you will be able to do

• Fundamentals of NLP: distributed semantics and language models
• Critical reading and assessment of current papers on interpretable AI
• Understanding fairness in ML and ethical aspects of bias
• Improvement of presentation and communication skills

What the module consists of

• SeminarTopics introduced by instructors, followed by paper-based student participation with presentations

Teaching method

• Seminar format with lecture and paper-readingConveying fundamentals by instructors and deepening through independent reading, presenting and discussion to foster critical thinking and presentation skills