How do we take into account uncertainty when we are making predictions?
Experiments involving selective inactivation of the hippocampus or its subregions demonstrated that the hippocampus is critical for temporal association memory. Sequential activity patterns, spanning over multiple time scales and characterizing the hippocampal population activity both during externally driven and spontaneous states, may be crucial for forming associations between temporally separated events. These data suggest that the hippocampus implements a model of events organised in time. Such a dynamic generative model could be used to continuously provide predictions about possible future events for decision making circuits.
Our current research focuses on testing a crucial prediction of this theory: that uncertainty increases towards predictions into the more distant future. In collaboration with Gergő Orbán (Wigner RCP), we are contrasting alternative models of representing uncertainty in a neuronal activity and our preliminary analysis using various publicly availably hippocampal datasets indicates that the population activity is consistent with a sampling based representation. This suggests that the spatiotemporal representation maintained by the hippocampus represents uncertainty and therefore is well equipped to contribute to optimal planning. For more information see this preprint.