Event prediction

Music, spoken language, and others’ actions are prime examples of highly complex events. And yet it seems child's play for us to decode them quickly, but also to predict quite accurately how a melody, phrase or movement will progress. How do our brains accomplish this? In our lab, we focus particularly on observed actions as predictable events. Moreover, observed actions can be used to investigate how the brain predicts events via hierarchically organized networks. Currently, the lab is investigating the assumption of whether lower levels of the neuronal hierarchy are optimized for predictions of first-order sequential structures (e.g., dynamic spatiotemporal state changes) while higher levels are optimized for predictions of higher-order sequential structures (e.g., contextually relevant information). Further, we are interested in how functional and structural hierarchical networks in the brain interact to provide an appropriate environment to predict hierarchically organized stimuli in the world.

Other Minds

Anticipation, prediction, and surprise are ubiquitous in our daily lives. And this becomes particularly obvious in social interactions. There is hardly anything more demanding and complex than interacting and communicating in social relationships and structures. Gestures, facial expressions, intonation, mood, actions, and utterances of the other person need to be processed in fractions of seconds, and our behavior should be adapted and tuned to this information. This performance is hardly conceivable without the constant and active generation of predictions, which then 'only' have to be compared and adapted to what actually happened. The brain uses several intertwined networks for these processes; for instance, large networks underlying our episodic memory that enable our affective experience and understanding or networks hosting models for movements and actions are all involved in social interactions. Our group studies expectations and breaches of expectations at all of these levels.

Surprise, surprise!

Surprises are beautiful. But sometimes also terrible. Intriguingly, most surprises we do not even notice. Quite in contrast, our brains always respond immediately to even minimal surprises, or, using the more technical term, prediction errors. The constant processing of prediction errors ensures the adjustment of our brains’ expectations to reality. Notably, prediction errors can trigger several different processes: the brain can switch to a different internal model (such as when our conversation partner changes from English to French); it can adapt an existing model (such as when a friend has a new hairstyle); but it can also register variance and probabilities and acquire a new predictive model over time (such as when we move to a new city). But what determines which process is taking place? To answer these and similar questions, we study the consequences of prediction errors of different types and strengths, and the reduction of prediction errors by expectations.


Memories as internal models

Memories fade, which is quite normal, and we all know about it. But sometimes, we remember events differently than they actually were. Why do memories change? The mutability of memories has long been considered a deficit or capacity problem. Recently, however, it has been considered that it may also have a critical function. That is, as a repository of our internal predictive models, our memory should be subject to adaptation processes in order to be still useful when our environment changes. We investigate the role of prediction errors in learning processes that lead to changes in our memory. One working hypothesis we pursue is that the situations in which we retrieve memories are never quite the same, and thus memory retrieval is always associated with episodic prediction errors. These have the function of correcting the internal model, in this case memory. One of the techniques we often use in our experiments is to establish episodic memories and test whether they change in the long term due to subtle prediction errors in retrieval.


Although our brains have relatively few dopaminergic neurons, dopamine plays a fundamental role in many functions of our nervous system. In a nutshell, dopamine keeps us on track! To this end, dopamine enables two complementary superpowers: to persevere in pursuit of a goal, but also to change it when necessary. Our lab is particularly interested in the role of dopamine in the optimal balance of cognitive stability and flexibility, which is a basic characteristic of optimal predictions. Thus, predictions should be stable enough to survive minor irritations and simultaneously flexible enough to be revised if things out there really change. We have studied this delicate balance in different groups of patients whose dopaminergic supply is impaired, such as individuals with Parkinson's disease, schizophrenia, or ADHD, and we are further investigating its neural underpinning in both healthy and impaired populations.