Explainable Artificial Intelligence

Course, master, summer 2023, L.079.05806

Content

Explaining the predictions of machine learning models is important in an increasing number of applications. For example, bank customers would like to know why their loan was denied; machine learning engineers would like to debug and improve their models; managers would like to ensure regulatory compliance. This course aims to explain the predictions of machine learning models and introduces different explanation methods to do so. Explanation methods can be distinguished whether they are specific to a certain model or model-agnostic and whether they explain an individual prediction or the entire model.

The students learn to explain the predictions of machine learning models, to choose an appropriate explanation method, and to implement explanation methods.

The course covers:

  • Introduction (e.g., importance of interpretability, evaluation of interpretability, datasets used in case studies)
  • Interpretable models (e.g., linear regression, logistic regression, decision trees, decision rules)
  • Global model-agnostic methods (e.g., partial dependence plots, permutation feature importance, global surrogate models)
  • Local model-agnostic methods (e.g., LIME, Anchors, SHAP, counterfactual explanations)
  • Model-specific methods (e.g., for neural networks)

Related Work

Organization

Link to PAUL: L.079.05806 Explainable Artificial Intelligence (in English)

Link to Panda: Explainable Artificial Intelligence

Lecture

  • Instructor: Dr. Stefan Heindorf
  • Location: F1.110
  • First Date: April 4, 2023
  • Last Date: July 11, 2023
  • Time: Tuesdays, 16:15 - 17:45

Tutorial

  • Instructor: Dr. Stefan Heindorf
  • Location: F1.110
  • First Date: April 24, 2023
  • Last Date: July 3, 2023
  • Time: Mondays, 9:15 - 10:45 (biweekly)

Mini Project

  • Instructor: Dr. Stefan Heindorf
  • Location: F1.110
  • First Date: April 18, 2023
  • Last Date: June 27, 2023
  • Time: Tuesdays, 17:45 - 19:15

Contact

Stefan Heindorf