Formation, deployment, and generalization of neural predictions

• Applicant: Dr. Ryszard Auksztulewicz
• Subject Area: Human Cognitive and Systems Neuroscience
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 507693881

When interacting with its environment, the brain is not only engaged in passive sensory processing, but also in predictive processing of sensory stimuli. Predictions need to be flexibly updated and generalized across appropriate contexts to aid perception and guide behavior. Despite research into the neural and computational mechanisms of predictive processing, the brain mechanisms mediating the formation, deployment, and generalization of sensory predictions are largely unknown. Preliminary work has identified qualitative differences in predictive mechanisms pertaining to different stimulus dimensions, and modulations of predictions by other cognitive factors (e.g., task relevance). However, it remains to be tested whether the neural mechanisms of predictions are shared or dissociable across phases of predictive processing (e.g., during the initial formation of predictions and their subsequent deployment), and whether predictions acquired in one task context (e.g., detecting a stimulus in noise) can be generalized to another context (e.g., discriminating specific stimulus features).Accordingly, this proposal focuses on testing whether the mechanisms of prediction signalling differ across phases and tasks. The project combines electrophysiology, neuroimaging, and modeling, allowing for a systematic testing of biophysically plausible, mechanistic explanations of non-invasive data. First, an empirical study using high-density electroencephalography (hd-EEG) in healthy human volunteers will disambiguate the effects of predictive processing phases on the dynamics of neural responses to predicted and unpredicted stimuli, and on prediction decodability based on neural activity. Another hd-EEG study will quantify effects of task requirements, and of prediction transfer across tasks, on neural responses and decodability. Second, parallel studies using functional magnetic resonance imaging will identify brain networks involved in predictive processing across its different phases and task requirements. Finally, the empirical results will be integrated in a coherent computational framework, unifying the behavioral and neural effects of flexible prediction signalling across phases and tasks. Thereby, the project will not only elucidate the neural mechanisms of flexible predictive processing, a core function of the brain, but also the role of predictions in learning, offering a blueprint for creating environments particularly conducive to learning and its transfer across contexts.

• DFG Programme: Research Grants