Paper accepted at IEEE ICDM 2026
1 October 2026
We are excited to share our new paper, “Explaining Changes in Evolving Systems,” accepted to the Blue Sky Ideas Track at IEEE ICDM 2026.
The paper is authored by Prof. Eirini Ntoutsi, Prof. Mykola Pechenizkiy from Eindhoven University of Technology, and Prof. Jerzy Stefanowski from Poznan University of Technology.
AI systems increasingly operate in dynamic environments where data distributions, user behaviour, and model behaviour change over time. Nevertheless, most Explainable AI methods are designed for static settings and explain individual predictions at only one point in time.
This paper argues that XAI should move beyond explaining predictions and begin explaining change itself.
Key takeaways:
-
A temporal perspective on XAI. Explanations should reveal what changed, why it changed, which groups were affected, and whether model adaptation is appropriate.
-
A three-layer framework. Changes are studied across the data, model, and explanation layers, as well as through the interactions between them.
-
Explanation drift matters. Even models with similar predictive performance can produce substantially different explanations over time.
-
New requirements for dynamic XAI. The paper identifies temporal consistency, causal attribution, cross-layer coherence, actionability, suitable granularity, and computational efficiency as central properties.
-
A research agenda for evolving systems. New evaluation methods and monitoring tools are needed to support developers, end users, auditors, and regulators.
Explaining change should become a first-class objective of XAI research, enabling greater transparency, accountability, and more informed intervention in evolving AI systems.