Responsible Machine Learning 2400-SZD-QPE-RML
During this class we will discuss:
• Basics of machine learning.
• Global methods of explaining the model, such as permutation importance of variables, Partial dependence profiles.
• Methods of local model explanation, such as Shapley values, LIME, Break-Down, Ceteris paribus.
• Methods of fairness analysis of a model.
• We will discuss examples of literary failures related to ML models.
Apart from the seminar formula, students will prepare a short essay on cases of using responsible ML.
This essay and its presentation will be the basis for the credit.
Bibliography: Fairness and machine learning
Limitations and Opportunities
Solon Barocas, Moritz Hardt, Arvind Narayanan
https://fairmlbook.org/
Explanatory Model Analysis
Explore, Explain, and Examine Predictive Models. With examples in R and Python.
Przemyslaw Biecek and Tomasz Burzykowski
https://pbiecek.github.io/ema/
A Responsible Machine Learning Workflow with Focus on Interpretable Models, Post-hoc Explanation, and Discrimination Testing
Navdeep Gill, Patrick Hall, Kim Montgomery, Nicholas Schmidt
https://www.mdpi.com/2078-2489/11/3/137
Rodzaj przedmiotu
Efekty kształcenia
By the end of the course, the student will be familiar:
- with basic machine learning techniques,
- with the area of eXplainable Artificial Intelligence,
- with the area of fairness and transparency of Machine Learning.
Kryteria oceniania
By project. The course grade will be based on the preparation of a use case for Responsible ML, which will be described in the form of a short essay in an open ebook. See for example https://pbiecek.github.io/xai_stories/.
The readability of the description, relevance for ML modelling and innovation related to new RML applications will be assessed.
Literatura
Fairness and machine learning
Limitations and Opportunities
Solon Barocas, Moritz Hardt, Arvind Narayanan
https://fairmlbook.org/
Explanatory Model Analysis
Explore, Explain, and Examine Predictive Models. With examples in R and Python.
Przemyslaw Biecek and Tomasz Burzykowski
https://pbiecek.github.io/ema/
A Responsible Machine Learning Workflow with Focus on Interpretable Models, Post-hoc Explanation, and Discrimination Testing
Navdeep Gill, Patrick Hall, Kim Montgomery, Nicholas Schmidt
https://www.mdpi.com/2078-2489/11/3/137
Więcej informacji
Dodatkowe informacje (np. o kalendarzu rejestracji, prowadzących zajęcia, lokalizacji i terminach zajęć) mogą być dostępne w serwisie USOSweb: