In brief (compressed)

Matthijs likes data, patterns, algorithms, and information theory. As Professor of Explainable Machine Learning and leader of the EDA group at LIACS, he develops theoretically grounded, human-centred methods that help domain experts discover what matters in complex data and make trustworthy predictions and decisions.

Research

Matthijs’ primary research interests are exploratory data mining and interpretable machine learning. He develops methods that identify relevant structure in data and express it in forms people can inspect directly. These include probabilistic rule sets, subgroups, decision trees, histograms, patterns, and explanations of anomalous observations. Rather than treating interpretability solely as an explanation added after a model has been trained, he is especially interested in models that are interpretable by construction.

His signature approach is to identify patterns that matter: succinct descriptions that characterise relevant structure in data. Information-theoretic concepts, particularly the Minimum Description Length (MDL) principle, provide a principled way to balance how well a model explains the data against its complexity. Because the same data may support several models that perform almost equally well, he also studies human-guided mining and learning: how domain knowledge, user feedback, and interactive exploration of alternative models can help identify results that are not only accurate, but also meaningful and useful. Although much of this work starts from tabular and other structured data, it increasingly extends to event sequences, dynamic graphs, time series, and spatiotemporal data.

Matthijs is also interested in connecting interpretable learning with causal questions. Recent research on subgroup discovery investigates which understandable groups experience the greatest effects from a treatment or intervention, thereby connecting the discovery of relevant structure with reasoning about cause and effect. This is one example of how the EDA group combines its two complementary research themes.

Many of these questions originate in interdisciplinary collaborations. His work spans health and life sciences, human behaviour and society, and high-tech industrial systems. In these collaborations, practical challenges expose limitations of existing methods and inspire new algorithms and theory. In turn, interpretable models help domain experts obtain clearer explanations, more reliable predictions, and new domain knowledge.

Leadership, education and mentoring

At LIACS, Matthijs leads the Explanatory Data Analysis group and serves as programme director for the MSc Computer Science, MSc Creative Intelligence & Technology, and the Business Studies specialisation. He is also a member of the LIACS Management Team.

He teaches Statistics and Information Theoretic Data Mining, emphasising a sound understanding of the fundamentals and a critical, responsible approach to data science and artificial intelligence. He has supervised numerous researchers, from MSc students to PhD candidates and postdoctoral researchers. Supporting their academic and personal development is an important part of his approach to leadership.

Background and academic service

Matthijs was previously associate professor (2020–2026), tenure-track assistant professor (2017–2020), and senior researcher (2015–2017) at Leiden University. Before coming to Leiden, he was a postdoctoral researcher at KU Leuven (2011–2015) and Utrecht University (2009–2011). He defended his PhD thesis, Patterns that Matter, at Utrecht University in February 2010.

He has received several best-paper and reviewer awards and was awarded NWO Rubicon, FWO Postdoc, NWO TOP2, and NWO TTW Perspectief grants. He chairs the IDA Council and is an Action Editor for the journal Data Mining and Knowledge Discovery. He also contributes to the international data mining and machine learning community through conference organisation, editorial work, and reviewing.

For a detailed CV and further information, see patternsthatmatter.org.