In today's rapidly evolving research landscape, machine learning has emerged as an indispensable tool, enabling unparalleled insights and efficiencies across diverse domains. As social scientists have begun to use machine learning tools in their research, the need to understand and explain the complex behavior of seemingly black box machine learning models has risen enormously. Why did a model arrive at its concrete prediction for a particular data instance? Which variables were most important for the model behavior globally and how do they interplay? Does the model show unfair and discriminatory behavior towards vulnerable groups?

 

This workshop introduces the field of interpretable machine learning, also often referred to as explainable AI (xAI), with R and Python. Throughout the workshop, we will build both a solid conceptual understanding as well as engage in hands-on experience in applying the most popular xAI techniques like SHAP, LIME, Counterfactual Explanations, Permutation Feature Importance and Partial Dependence Plots (PDP) as well as Individual Conditional Expectations Plots (ICE). In addition, we will get to know techniques designed not only to understand models' behavior in general, but also with a particular focus on fairness, bias, and discrimination. The focus of the workshop lies on supervised machine learning models for tabular data. During the coding labs of the workshop, participants can individually decide whether they prefer to implement the discussed techniques either with R or with Python.

 

Organizational Structure of the Course

Each day is structured around one three-hour block in the morning and one three-hour block in the afternoon. Each block consists of 90 minutes of conceptual lectures and discussions of the material as well as 90 minutes of hands-on lab sessions. During the labs, participants can individually decide whether they prefer to work in R or Python. During the lab sessions, lecturers will be available to support participants, to facilitate discussions about the application of the material, and for individual consultations on participants' projects as long as they relate to the application of explainable AI and machine learning in general.

 

Software and Hardware Requirements

The workshop will be based on the open-source programming languages of R and Python, among which the participants can individually choose. We follow the principles of 'Open Data,' 'Open Code,' and the integration of narrative text and code (no commercial software is needed). Depending on your choice of working with either R or Python, please install R and RStudio before the workshop or Python and an IDE of your choice (e.g. Spyder, VS Code), respectively. Participants will receive an email with further installation instructions before the workshop (e.g., regarding required R and Python packages).