Dynamic belief updating in mice, rats, humans and models

The environment is structured in time, allowing humans and animals to make predictions about the future and adjust behaviour.

To this end, they perform ‘dynamic belief updating’ whereby beliefs about event probabilities are iteratively revised in light of new observations. However, this strategy is challenged by two properties of the world, namely stochasticity and volatility.

These features imply that an unexpected observation might either simply reflect noise in the generative process or indicate that the process itself has covertly changed.

While remedying to stochasticity requires to integrate over past observations so as to cancel out noise from the estimates, the countermeasure for volatility instead dictates to discard past observations so as to compute the most up-to-date estimates. How do humans and animals strike this difficult balance?

In this interdisciplinary proposal, I offer to tackle this question using a unique combination of human cognitive neuroscience and rodent neurobiology, all guided by computational modelling. Authorities in these fields (Valentin Wyart, Ph.D. and Benjamin B.

Scott, Ph.D.) will provide me with the missing skillset which, combined with my pre-existing expertise, will put me in an unprecedented position to unravel how mammalian brains factor in uncertainty during dynamic belief updating.

Notably, I will test the idea that a sensory gain control mechanism gates the impact of novel observations onto past beliefs, and that this sensory gain is tuned by noradrenaline release from the locus coeruleus.

The findings will inform neurocomputational theories of learning as well as contribute to a better understanding of cross-species similarities and differences, which is key for more fruitful translational research programs and better optimized use of animals in research.

Furthermore, the acquired skillset will ideally position me as an independent researcher equipped with a cutting-edge approach at the frontiers of fields and species.