Neural Mechanisms and Functional Purpose of Dynamic Dimensionality States in Sensory Cortex

Over the past decade, measurements of brain activity have advanced to record a rapidly growing number of neurons simultaneously. Strikingly, these large-scale recordings have revealed that brain activity can often be effectively described using far fewer variables than the number of recorded cells, hence confining the neural activity to a low-dimensional space. The magnitude of this network feature has been observed to change dynamically, varying when we pay attention, perceive new information or memorise past experiences. Despite these observed correlations to task performance, a fundamental understanding of how network dimensionality influences sensory processing is missing. Additionally, it is unknown what biological network properties, such as network connectivity or the balance of different cell types, give rise to this dynamic phenomenon. We will address these questions by combining data analysis of large-scale calcium recordings from mice with the development of new artificial models that replicate these characteristics. Together, these approaches will yield testable hypotheses of how dimensionality causally influences the perception of the world around us, which we will put to the test by perturbing brain activity in mice during different dimensionality states. In short, we will develop a low-level understanding of how an internal state of a network, effective dimensionality, can modulate the network's computational capacity. This network trait is fundamental to biological and artificial systems alike, and without a quantitative understanding of its mechanisms, we critically limit our knowledge of how to make sense of our surroundings