My research broadly fits under the area of artificial intelligence, spanning from constraint-based reasoning to algorithmic data analysis of ecological and biological data, and from explorative data analysis with statistical significance testing to finding subjectively interesting patterns. In particular, I am interested in the combination of data analysis tasks and efficient constraint reasoning and optimization methodology in solving problems in various application domains.

My early research focused on development and analysis of constraint reasoning techniques, in particular in the area of answer set programming (ASP), that is, logic programs under the stable model semantics. I have, e.g., developed methods for the verification of equivalence and introduced an intuitive and elegant module architecture for ASP. In many cases translations and transformations between different problems, program classes, and formalisms have been utilized. I have also studied proof systems and backdoors for ASP.

I have also worked actively in the areas of computational argumentation and data analysis. In the context of computational argumentation, I have provided characterizations for equivalence under various semantics for abstract argumentation frameworks and analyzed properties of resolution- based grounded semantics. On the data analysis side I have been involved in e.g., analysis of the spatial patterns resulting from data on vascular plants, mammals, and climate as well as developed an ASP-based algorithm for finding optimal evolutionary supertrees given conflicting and partially overlapping sets of evolutionary trees.

My most recent research focuses on developing novel methods for exploratory data analysis combining the capabilities of the human and the computer. This includes characterising confidence intervals for multivariate data and studying subjective interestingness in interactive visual data exploration and subgroup discovery. A further research direction is related to interpretability and explainability of black-box models, where statistical significance of visual patterns is studied, and a framework for detecting concept drift (i.e., estimating the generalization error of regression functions) when the ground truth is unknown, is introduced.