CIRCUIT-LINK · Linking self-supervised learning to cortical circuits for uncertainty processing

The brain has the remarkable ability to maintain a robust model of the world even when sensory data is ambiguous and noisy. To achieve this, it dynamically prioritizes sensory observations or internal predictions depending on their reliability. When observations are unreliable, e.g., in low visibility conditions, they are weighted less compared to predictions, and vice versa. While previous studies have proposed intriguing hypotheses on how neural circuits could solve this dynamic weighting problem, a clear and empirically testable connection between theory and neural mechanisms is still missing. The main reason is that current theories are highly simplified, and are not able to quantitatively predict responses of a sensory system that is adapted to a highly complex environment.This project will investigate how the brain solves the dynamic weighting problem via a theory- and data-driven approach. I will develop a theoretical framework that enables solving the problem in realistic scenarios, by combining rigorous and interpretable Bayesian modeling with modern self-supervised deep learning methods. This will provide insights into the determinants of dynamical weighting in diverse environments. In simulations I will test the hypothesis that the same weighting mechanisms are beneficial across environments, meaning that pre-structuring world models with these mechanisms consistently improves uncertainty processing. Finally, to understand if and how these mechanisms are implemented in cortex, I will collaborate with experimental researchers at FMI to record neural responses to complex stimuli with uncertainty in mice. By quantitatively predicting these responses we can directly measure the success of the developed theory, and assign functional roles to individual neurons in the brain.The results of this project will elucidate how specific cortical circuit elements enable to model an uncertain world - one of the most fundamental abilities of the brain.