Feedback as the way forward: sensory predictions in the primate face processing hierarchy

• Applicant: Dr. Caspar M. Schwiedrzik
• Subject Area: Cognitive, Systems and Behavioural Neurobiology
• Term: since 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 405847822

The world is full of structure. Thunder follows lightning, and traffic lights turn from red to green. The brain exploits such patterns to make predictions about what will happen next. By anticipating instead of reacting, we can, eg. drive off just when the traffic light is green, avoiding angry honks. Predictions attune us to our environment to improve perception. In disorders like autism and psychosis, the ability to predict seems impaired. Yet, how predictions are implemented in the brain is poorly understood, especially for visual object recognition and at the single neuron level. Here, we aim to close this gap by testing the proposal that predictions are instantiated by feedback connections from higher to lower brain areas during stimulus processing. This radically departs from current models of object recognition that rely only on feedforward connections from lower to higher brain areas. We will capitalize on the face-processing system of macaque monkeys, a highly evolved face-processing system similar to that of humans. This system consists of distinct, hierarchically organized brain areas with unique, well-known functional properties. We will uncover the origins and consequences of predictions by exploiting a key discovery we recently made: as higher areas code information differently than lower areas, feedback carries a functional signature of higher areas that we can reveal in lower areas when predictions are violated. Building on this, we will study how predictions are passed across brain areas, how they impact information processing, and whether higher areas are causally relevant for predictions: we will localize the face-processing system in monkeys with neuroimaging and perform parallel electrophysiological recordings to establish the dynamics and directionality of information flow between higher and lower face areas during predictable and unpredictable sequences of faces (Aim 1). We will obtain causal evidence for the role of feedback in predictions by reversibly inactivating higher face areas during neuroimaging (Aim 2). Finally, we will test with neuroimaging whether humans employ the same neural solution as monkeys to make predictions about faces, and use the human conceptual system to investigate whether sensory and abstract predictions equally rely on feedback to face areas to improve perception (Aim 3). This comprehensive research across species (humans, monkeys), spatial scales (brain networks - single neurons), and levels of abstraction (sensory, conceptual) will provide a stringent test of the role of feedback in predictions, and unravel a new quality of information processing beyond current feedforward theories of face perception. Thereby, we will not only illuminate the neural mechanisms of object recognition, a core task of our visual system, but also the information-processing roots of predictive deficits in conditions like autism, and offer a blueprint for object-recognition computer algorithms that build on feedback.

• DFG Programme: Independent Junior Research Groups