Randomly connected networks generate emergent selectivity and predict decoding properties of large populations of neurons

摘要

What do we learn about neural circuit organization and function from recordings of large populations of neurons? For example, in population recordings in the posterior parietal cortex of mice performing an evidence integration task, particular patterns of selectivity and correlations between cells were observed. One hypothesis for an underlying mechanism generating these patterns is that they follow from intricate rules of connectivity between specific neurons, but this raises the question of how such intricate patterns arise during learning or development. An alternative hypothesis, which we explore here, is that such patterns emerge from generic properties of certain random networks. We find that a random network model matches many features of experimental recordings, from single cells to populations. We suggest that such emergent selectivity could be an important principle in brain areas in which a broad distribution of selectivity is observed.