Compositional task representations in the mouse cortex for multi-tasking

Abstract

The human brain can rapidly learn and perform multiple cognitive tasks and tackle a novel cognitive task using pre-existing knowledge. Previous studies demonstrated the crucial roles of the prefrontal cortex in cognitive tasks and identified the compositionality of the neural representations of the cognitive tasks. However, the mechanism of multi-tasking at the neural network level is poorly explained due to a lack of spatial resolution in human subject studies, and difficulties in experiment design in animal models. In this thesis, an experimental paradigm consisting of a series of cognitive tasks was designed, and the neural activity of mice during multitasking was imaged with two-photon calcium imaging. Behavioural data analysis suggested that mice did not learn faster across the tasks. Neural imaging analysis reveals that the neurons in the brain regions tend to have higher activities during the first half of trials across tasks. Clusters of neurons having similar neural activity are identified using an unsupervised machine learning method in a few regions. The neural activity showed ramping up during the delay epoch of the trials. These neural patterns may underly the ability of multi-tasking in animals.