Computational Modelling of Learning and Decision-Making
(former group - now in collaboration with the Ernst-Abbe-Hochschule/University of Applied Sciences Jena)
Computational modelling describes the process of simulating the structure and behavior of a complex system based on a mathematical model. The goal of computational modelling is to describe and understand the system and its interaction with the environment.
In our research, we develop and establish computational models for learning and decision making in humans. The models link human behavior to underlying brain structure and function assessed by functional neuroimaging. Aim of our approach is to provide mechanistic formulations of learning and decision making processes that foster their systematic examination and provide quantitative predictions of behavior and brain function in both healthy and diseased condition.
Our models are currently applied, in particular, in the context of obesity research. For this, we worked in collaboration with Annette Horstmann and her group as part of the O’Brain Project and in sub-project A5 of the Collaborative Research Center 1052 ‘Obesity Mechanisms’ (previous funding periods) at Leipzig University. Our group is also part of the Integrated Research and Treatment Center (IFB) Adiposity Diseases at the University of Leipzig Medical Center.
In our work, several classes of computational models are of relevance.
Cognitive models of behavior are used to decompose complex behavior into mathematically tractable latent variables such as prediction errors or decision boundaries, this way formalizing and testing hypotheses about internal cognitive mechanisms that cannot be accessed directly by experimental manipulation. We relate parameters derived from these models (e.g. prediction errors or response certainties) to underlying neuronal mechanisms through model-based functional neuroimaging. Using model parameters to inform the statistical analysis of neuroimaging data accounts for inter-individual differences in the observed measurements and thus helps to explain behavioral and functional alterations related to particular subject populations or diseases.
Mechanistic modular brain models explain observable behavior on the grounds of the intact or deficient performance and interaction of particular model components, such as specific cortical and sub-cortical brain structures or neurotransmitter systems.
Brain connectivity models describe the global behavior of large-scale functional brain networks. They are used to identify functional brain networks underlying specific cognitive tasks, or individual brain regions that play a key role in particular brain networks.
Our research integrates both mathematical and experimental work. Computational modelling, structural and functional neuroimaging, and functional connectivity analysis are accompanied by behavioral and physiological assessments, quantitative meta-analyses of neuroimaging data, Bayesian and multivariate statistical modelling. The integration of several academic disciplines is expected to deepen and extend our knowledge about learning and decision-making in general as well as about their population- or disease-specific alterations.
Our research was funded by the German Research Foundation (DFG) and the German Federal Ministry of Education and Research (BMBF).