Summer School 2026

Explainable Artificial Intelligence (XAI) in Economics and Finance

Overview

As machine learning models increasingly drive modern economic forecasting and financial risk assessment, their inherent complexity often renders them ‘black-boxes’ to human decision-makers. While these models offer unprecedented predictive power, the lack of transparency in how they arrive at specific conclusions poses significant challenges for accountability and regulatory compliance. This lecture series provides a comprehensive overview of Explainable Artificial Intelligence (XAI), bridging the gap between high-performing algorithms and the transparency required for critical central banking operations. By the end of the course, participants will be able to implement XAI techniques that enhance trust, accountability, and safety in autonomous financial systems. We will cover:

• XAI taxonomy and core techniques
• practical implementation with real-world examples
• theoretical guarantees and limitations

Organisation

• 5 sessions (3h lecture + 3h hands-on lab)
• bring your laptop for the lab sessions (make sure to have a Google Colab account set up)

Lectures and labs

• 1) Introduction to XAI, taxonomy, interpretable-by-design models. Lab: explaining random forests with MDA
• 2) Perturbation-based approaches: LIME and SHAP. Lab: explaining time-series forecasting with LIME
• 3) Concept-based XAI: concept-activation vectors, concept bottleneck models. Lab: concept-activation vectors and prediction of annual income
• 4) Counterfactual explanations: MACE and DiCE. Lab: counterfactual explanations for the German credit dataset
• 5) Explaining transformer-based models: introduction to the attention mechanism, inspecting attention patterns, gradient-based explanations. Lab: uncovering GPT activation

Prerequisites

• to be familiar with the basic concepts of machine learning (train / test data, loss function,…)
• to know about simple models such as linear regression and random forests
• to be proficient in Python, especially the Pytorch library

Useful resource

• general machine learning: Hastie, Tibshirani, Friedman, The Elements of Statistical Learning, Second Edition (2009)
• pytorch: https://docs.pytorch.org/tutorials/beginner/basics/intro.html

Trainer

• Damien Garreau (Julius-Maximilians-Universität Würzburg)

Participation is free of charge. Participants are responsible for arranging their own accommodation and travel.

For registration, please complete this form by 27 April 2026.

Selected participants will be notified by 11 May 2026.