Common and distinct neural correlates of trial-by-trial expectancies for feature-based as compared to spatial attention

Since our brain can only process a small amount of sensory information at a time, attention systems select specific aspects for prioritized processing – such as features or locations – that are expected to be most relevant in a given context. The present thesis investigated responses of visual attention systems to expected and unexpected sensory events during uncertainty, when the probability of sensory events needs to be inferred from environmental observations. In the spatial attention domain these processes have already been described with the help of computational models and the underlying neural mechanisms have been explored with functional neuroimaging. However, it is not known whether other attentional subsystems such as feature-based attention behave similarly during uncertainty. 
\nThe core aim of this thesis was to characterize the modulation of attentional deployment by inferred probabilities during feature-based as compared to spatial attention. This was realized using a Posner-cueing paradigm in which feature and spatial cues were presented. Classically these cues predict the color or the location of an upcoming target with a fixed high probability, so that participants expect the cues to be valid and respond slower and less accurate when attention needs to be reoriented to invalidly cued targets. In the present paradigm, this probability was varied by changing the percentage cue validity (%CV) unpredictably over time. 
\nIn a behavioral experiment (Experiment 1) three different color cues were used to identify cue-related factors which influence the effects of probabilistic inference on feature-based attention and to establish an experimental paradigm for the comparison of feature-based and spatial attention systems. It was observed that all color cues affected attentional deployment. However, probability-dependent effects differed depending on the level of cue abstraction: More automatically processed cues required more observations of cue-target outcomes to establish probabilistic learning than abstract cues. 
\nExperiment 2 employed functional magnetic resonance imaging (fMRI) to investigate the computational and neural mechanisms that modulate probabilistic inference using the experimental paradigm from Experiment 1 with abstract feature and spatial cues. The results indicated that probabilistic inference follows similar principles for both attention systems. However, their neural implementations seemed to be confined to domain-specific subsystems: The right temporoparietal junction (TPJ) was particularly involved in spatial attention, while the left intraparietal sulcus (IPS) was most crucial for feature-based attention. However, the left anterior IPS showed an effect of probability-dependent attention in both attention systems. These findings provide novel insights into the generality and specificity of the functional basis of visual attention, suggesting that probabilistic inference can distinctively affect each attentional subsystem, but that the left anterior IPS may establish probabilistic inference in a domain general manner. 
\nTaken together our findings speak against a unitary visual attention network. Rather, we propose that depending on the nature of the selected aspects different attentional subsystems are activated following expectancy violations and these processes can even differ within one single domain such as feature-based attention.