T.Claassen (at) science.ru.nl / tomc (at) cs.ru.nl
For more contact details see here.
Main topics of interest:
Statistical Causal Inference (methods and principles)
My research focusses on causal discovery methods and their application to real-world experiments.
I am currently involved in the EU-FP7 projects MATRICS and Aggressotype, and the NWO-projects MoCoCaDi ('More Confidence in Causal Discovery'), and 'Causal Discovery from High-Dimensional Data in the Large-Sample Limit'. As of 1 April 2016 I also work 50% in the AMLab group of Dr Joris Mooij at the UvA.
The MoCoCaDi project aims to improve the use of 'proper' causal discovery methods in standard data analysis by providing more robust and flexible methods to identify likely cause-effect relations from observational data, as well as a principled quantification of both the strength and the reliability of this connection.
The projects MATRICS and Aggressotype deal with genetic, cognitive, and neuropsychological factors that underly human conduct disorder and antisocial behaviour in general, and the relation between callous-unemotional / aggressive traits and ADHD in children and young adults in particular.
Goal of our subproject is to build up a model that can explain the connection between the many associated risk factors identified in other sub-projects, aid understanding of the underlying make up of CD/CU conditions (subclasses) and help define possible intervention strategies on these risk-factors. It involves data sets as diverse as different DSM-V questionnaires (incl. self-evaluation reports), IQ/education data, brain-activation images, hormonal/heart-rate measurements, animal behavioural experiments, idem differences in brain-metabolites (MR/ASL/DTI), data on substance abuse/addiction, and psychopathic traits in non-institutionalised populations.
Related subjects I am working on:
Bayesian estimates for strength of cause-effect relations
Inference on data from multiple exeriments
Handling deterministic relations in causal structure learning
Extensions to more powerful/expressive model classes (e.g. cyclic models, mixed data, ADMGs)
Using GPU programming to speed up causal discovery algorithms
Ecology: modeling habitat selection in migratory species
Before joining the machine learning group of Tom Heskes I worked as a design/specification expert and systems architect for two of the main IT-companies in the Netherlands (Logica, AtosOrigin), doing projects for industry (Philips, Shell) and (semi)government (KNMI, LMD-Schiphol, Min.Def.)
S. Magliacane, T. Claassen, J. Mooij Joint Causal Inference. Technical Report, arXiv.org e-Print archive, November 2016. [pdf]
S. Magliacane, T. Claassen, J. Mooij Ancestral Causal Inference (ACI). Advances in Neural Information Processing Systems (NIPS), 2016. [pdf]
E.Sokolova, P.Groot, T.Claassen, T.Heskes Causal discovery from medical data: dealing with missing values and a mixture of discrete and continuous data. Accepted for the 15th Conference on Artificial Intelligence in Medicine (AIME), 2015. [pdf]
E.Sokolova, M.Hoogman, P.Groot, T.Claassen, A.Vasquez, J.Buitelaar, B.Franke, T.Heskes Causal discovery in an adult ADHD data set suggests indirect link between DAT1 genetic variants and striatal brain activation during reward processing. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 2015. [pdf]
Tom Claassen, Joris Mooij, and Tom Heskes. Proof Supplement to: Learning Sparse Causal Models is not NP-hard Technical Report, arXiv.org e-Print archive, November 2014.
E.Sokolova, P.Groot, T.Claassen, T.Heskes Causal Discovery from Databases with Discrete and Continuous Variables. In Proceedings of the 7th European Workshop on Probabilistic Graphical Models(PGM), 2014.
Tom Claassen and Tom Heskes. Bayesian Probabilities for Constraint-based Causal Discovery. In Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI), 2013.
Tom Claassen, Joris Mooij, and Tom Heskes. Learning Sparse Causal Models is not NP-hard. In UAI 2013, Proceedings of the 29th Conference on Uncertainty in Artificial Intelligence, 2013.
Tom Claassen and Tom Heskes. BCCD algorithm (abstract only) Faculty of Science, Radboud University Nijmegen. In Belgian-Dutch Conference on Machine Learning (BENELEARN) , 2013.
Tom Claassen and Tom Heskes. A Bayesian Approach to Constraint Based Causal Inference Faculty of Science, Radboud University Nijmegen. In UAI 2012, Proceedings of the 28th Conference on Uncertainty in Artificial Intelligence, 2012.
Tom Claassen and Tom Heskes. BCCD - Supplement Technical Report, Faculty of Science, Radboud University Nijmegen.
Tom Claassen and Tom Heskes. A logical characterization of constraint-based causal discovery. In UAI 2011, Proceedings of the 27th Conference on Uncertainty in Artificial Intelligence, 2011.
Tom Claassen and Tom Heskes. Proof supplement to: A logical characterization of constraint-based causal discovery. Technical Report, Faculty of Science, Radboud University Nijmegen, June 2011.
Tom Claassen and Tom Heskes. A structure independent algorithm for causal discovery. In European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 2011.
Tom Claassen and Tom Heskes. Arrowhead completeness from minimal conditional independencies. Technical Report, Radboud University Nijmegen, iCIS/IS/ML, Nov. 2010.
Tom Claassen and Tom Heskes. Causal discovery in multiple models from different experiments. In Advances in Neural Information Processing Systems 23, pp.415-423, 2010.
Tom Claassen and Tom Heskes. Learning causal network structure from multiple (in)dependence models. In Proceedings of the fifth European Workshop on Probabilistic Graphical Models, 2010.
Tom Claassen. Zooming in on Probability and Causality. CAPITS, 2008 (presentation).
Winner ICPM Causal Inference Challenge, (CRM Montreal), July 2016.
The Willem R. van Zwet Award for the best PhD-thesis in statistics and OR in 2012-2013.
Best Paper Award at the UAI2012 conference for A Bayesian Approach to Constraint-based Causal Discovery.
In June 2013 I defended my doctoral thesis entitled 'Causal Discovery and Logic' with a cum laude distinction. I was honoured to host the accompanying mini-symposium with contributions by top-level scientists from both inside and outside the Netherlands.
In 2016-2017 I teach the following courses:
Statistical Machine Learning (CS master, 6EC)
Research A/B, (CS/IS master, 3EC)
Previous subjects include:
Requirements Engineering (CS bachelor, 3EC)
Machine Learning in Practice, (CS master, 3EC)
Data Mining (CS/IS bachelor)
Wiskunde 1 for HBO-instroom (IS master)
Introduction to Pattern Recognition (CS master)
A shortlist of more and less pretentious interests:
language and meter in Old-English poetry (Beowulf)