Priors in perception: How foreknowledge changes neural computations

One of the most fundamental functions of the human brain is to predict future events on the basis of the recent past. Prediction plays a prominent role in models of seemingly disparate cognitive functions such as perception, motor control, and language processing. We often only become aware of these predictions when they are violated. For example, when opening the door of our house does not result in the expected auditory squeak, we are surprised, whereas if we hear the expected squeak at the expected time, we fail to notice it. Sensory predictions can originate from our own actions, but they can also by generated by probabilistic associations with other sensory events that have been learned from previous experience. Although there is ample behavioral evidence for the predictive nature of perception, the incorporation of prior knowledge in the neural processes underlying perception is still poorly understood.

My long-term research goals are to understand how prior expectations (‘priors’) about the sensory world influence the neural computations that give rise to perception. In my lab, we view perception as a process of probabilistic inference, in which bottom-up input is continuously compared with top-down expectations. Perception arises from this recurrent interaction as the brain settles on a particular interpretation of the world. My lab uses a variety of approaches (neuroimaging, electrophysiology, brain stimulation and psychophysics) to tackle this problem of predictive perception at several levels of granularity - from activity in sensory populations to neural systems and behavior. Moreover, we employ sophisticated population analysis methods to obtain feature-specific responses from neural recordings, and aim to link empirical data to computational models of perceptual inference.

In doing so, we have gained novel insights into how priors change sensory neural computations and behavior. We recently showed that priors can lead to a reduction of sensory activity, but a concurrent increase in informational content in the sensory response. This suggests that priors may sharpen the sensory representation, and it is an important step towards understanding the functional significance of activity modulations induced by prior expectation. Secondly, we are studying the link between prior expectation and the closely related concept of selective attention. Whereas these concepts have been treated as interchangeable by many scholars, we have begun to provide empirical evidence for their distinct computational roles and neural implementations, as well as their synergistic interaction.

My lab will take this research line forward by connecting current data to both more fine-grained computations within cortical circuits (‘zooming in’) and a network investigation of the communication between sensory regions with memory- and decision-related structures during the acquisition of prior expectations (‘zooming out’). Connecting these levels of description will allow for a more complete linkage between computational models of predictive perception, sensory neural data and system behavior. This will provide a deeper account of the predictive nature of the neural computations that implement human perception, and may help to build models that can account for the deficits that arise when these computations go wrong, such as in autism and schizophrenia.