Methods and Issues in Explainable AI— Free Online Course

This workshop surveys contemporary methods in explainable AI (XAI) with a focus on practical use: motivation and regulatory context, trade-offs between interpretability and performance, and integration of explainability into standard ML workflows. We highlight deep-learning-specific challenges and interpretable architectures for vision and forecasting. Course materials and recordings are available online.

Learning outcomes

• Recognize when and why explainability is required (technical and regulatory drivers).
• Use a concise taxonomy of XAI methods and select appropriate approaches for a task.
• Integrate explainability into the ML workflow (EDA, feature engineering, model selection, evaluation).
• Differentiate post-hoc explanations from intrinsically interpretable models and assess explanation quality.

Syllabus (brief)

1. Introduction: motivation, taxonomy, and interpretability vs. black-box trade-offs.
2. Post-hoc methods: feature attributions, partial dependence, LIME, SHAP.
3. Deep-learning techniques: saliency maps, integrated gradients, data-valuation methods.
4. Interpretable vision models: prototype and prototype-tree approaches (ProtoTreeNet).
5. Interpretable forecasting: probabilistic and attention-based models; analyst-in-the-loop case studies.

Materials

Slides, notebooks, and recordings are linked from the course page. See the provided LMS link for full resources. Note that a free registration is neccessary in order to access the content.